Relationships Between Patient Satisfaction, Quality, Outcomes and Ownership Type in US Hospitals An Empirical Study
Relationships Between Patient Satisfaction, Quality, Outcomes and Ownership Type in US Hospitals An Empirical Study
Ediyattumangalam R. Shivaji
itt
An Abstract
of a dissertation submitted to the Graduate School of Maharishi University
of Management in partial
fulllment of the requirements for the degree of
Doctor of Philosophy
May, 2012
Dissertation Supervisor: Dr. Bruce McCollum
Abstract
Ediyattumangalam R. Shivaji
Public concerns about rising health costs and deteriorating quality of service in the US
have become a serious issue. The Institute of Medicine (IOM)1 report brought out the need for
overhauling the US Healthcare thoroughly. This report recommended that healthcare executives
should focus on performance improvement, driven by process, data, and evidence rather than
relying on technology or working harder. Healthcare organizations face multiple objectives and
constraints, while implementing performance improvement,.
The design of the current study was nonexperimental and the study analyzed available
archival data on patient satisfaction, process of care quality measures and outcome of care
measures. The study tested nine research hypotheses about the relationships between these
measures. The study also brought out the main components contributing to patient satisfaction
and process of care quality measures.
The study used the public data on US hospitals, downloaded from the CMS database,
maintained by the Center for Medicare and Medicaid services, a federal government agency. Data
from over 4,500 hospitals were used in the analysis.
The major ndings are summarized as follows:
1. Five components of patient satisfaction were identied and the implications to hospitals
were discussed.
2. Nine research hypotheses were tested, and the evidence was mixed.
3. Mean outcome rates in Church owned hospitals were signicantly better than the other
seven groups and denitely not worse.
1
IOM. (2001) Crossing the Quality Chasm: A New Health System for the 21st Century. Committee on Quality of
Ediyattumangalam R. Shivaji
itt
A Dissertation
submitted to the Graduate School of Maharishi University of Management in partial
fulllment of the requirements for the degree of
Doctor of Philosophy
May, 2012
UMI 3523284
Copyright 2012 by ProQuest LLC.
All rights reserved. This edition of the work is protected against
unauthorized copying under Title 17, United States Code.
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
ii
c
2012
Ediyattumangalam R. Shivaji
itt
All Rights Reserved.
Graduate School, Maharishi University of Management
Faireld, Iowa
Transcendental Meditation technique, Maharishi TM-Sidhi program, Maharishi Vedic
Approach to Health, Maharishi Ayur-Veda , Science of Creative Intelligence, Maharishi Vedic
Science, and Maharishi University of Management are registered or common law trademarks
licensed to Maharishi Vedic Education Development Corporation and used with permission.
iv
r
In line with the Vedic scholastic traditions, I begin my work, humbly thanking all my teachers for
giving me the knowledge and skills that enabled me to write this dissertation.
In particular, I respectfully dedicate this work to the great teacher of these teachers, His Holiness
Maharishi Mahesh Yogi and his spiritual master Guru Dev Shankaracharya Swami Brahmananda
Saraswati.
Abstract
Public concerns about rising health costs and deteriorating quality of service in the US have
become a serious issue. The Institute of Medicine (IOM)1 report brought out the need for
overhauling the US Healthcare thoroughly. This report recommended that healthcare executives
should focus on performance improvement, driven by process, data, and evidence rather than
relying on technology or working harder. Healthcare organizations face multiple objectives and
constraints, while implementing performance improvement,.
The design of the current study was nonexperimental and the study analyzed available archival
data on patient satisfaction, process of care quality measures and outcome of care measures. The
study tested nine research hypotheses about the relationships between these measures. The study
also brought out the main components contributing to patient satisfaction and process of care
quality measures.
The study used the public data on US hospitals, downloaded from the CMS database, maintained
by the Center for Medicare and Medicaid services, a federal government agency. Data from over
4,500 hospitals were used in the analysis.
The major ndings are summarized as follows:
1. Five components of patient satisfaction were identied and the implications to hospitals
were discussed.
2. Nine research hypotheses were tested, and the evidence was mixed.
3. Mean outcome rates in Church owned hospitals were signicantly better than the other
seven groups and denitely not worse.
1
IOM. (2001) Crossing the Quality Chasm: A New Health System for the 21st Century. Committee on Quality of
Health Care in America, Institute of Medicine.
vi
4. Evidence was mixed for negative association between patient satisfaction and outcomes.
5. Evidence was mixed for negative association between process of care quality and outcomes.
The study found some empirical evidence for encouraging hospitals to adopt the qualities
friendship, compassion, joy of serving and equanimity advocated by the ancient Vedic
physician Charaka as the prime qualities required by healthcare professionals. The study has
many strengths such as identifying the principal components of satisfaction and quality, using the
complete CMS data on US hospitals and obtaining some empirical evidence on the relationships
between satisfaction, process-of-care quality and the outcomes. Some empirical evidence was
also obtained on the need for qualities like compassion among healthcare staff.
The study ndings are limited by the reliability of the archival data used. Statistical conclusion
validity issues were adequately controlled during testing, by adopting diagnostic techniques.
However, ambiguity of temporal precedence between outcomes and process of care quality
measures is a threat to the internal validity of testing their relationship. A subsequent larger study
requiring support from CMS is proposed.
The study ndings will assist hospitals in their performance improvement activities.
vii
Table of Contents
Copyright
ii
Approval
iii
Dedication
iv
Abstract
List of Tables
xx
List of Figures
xxv
xxix
1 Study Overview
1
Charakas concept of healthcare quartet. . . . . . . . . . . . . . . . . . 1
Problems of healthcare in US. . . . . . . . . . . . . . . . . . . . . . . 1
High cost. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Inefciencies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Errors and patient safety. . . . . . . . . . . . . . . . . . . . . . . . . . 2
IOM reports. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Joint Commission efforts. . . . . . . . . . . . . . . . . . . . . . . . . . 3
viii
ix
25
Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Healthcare Costs and quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Healthcare errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Studies on HCAHPS Survey Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Studies on quality data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Effects of Ownership type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Hansmanns theory of non-prot hospitals. . . . . . . . . . . . . . . . . . . . . . . 35
Relationship of patient satisfaction, quality and outcomes . . . . . . . . . . . . . . . . . 36
Strategy for searching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3 Methodology
39
Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Research Design and Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Description of Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Archival data retrieved. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Participants. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
HCAHPS survey instrument. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Time period for the downloaded data. . . . . . . . . . . . . . . . . . . . . . . . . . 41
Data coverage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Data preparation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Threats to validity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Statistical conclusion validity issues. . . . . . . . . . . . . . . . . . . . . . . 43
Internal validity issues. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
External validity issues. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Construct validity issues. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Analysis of Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Principal component analysis (PCA). . . . . . . . . . . . . . . . . . . . . . . . . . . 47
HCAHPS data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Multivariate normality. . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Multivariate outliers. . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Linearity assumption. . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Process of care quality data. . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Multivariate normality. . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Multivariate outliers. . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Missing data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Research question 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Research question 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Assumptions to be satised in testing. . . . . . . . . . . . . . . . . . . 53
Research question 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Research hypotheses under research question 3. . . . . . . . . . . . . . . . . 54
Research question 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Research question 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Relationship between outcomes and hospital ownership. . . . . . . . . . . . . 58
RQ 5.1 and RQ 5.2 - relationship between outcome variables and
ownership type. . . . . . . . . . . . . . . . . . . . . . . . . 60
xi
67
Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Research Question 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Principal component analysis of HCAHPS data (PCA). . . . . . . . . . . . . . . . 68
Hospital consumer assessment of healthcare providers and systems
(HCAHPS). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Survey method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
HCAHPS Sampling methods and participants. . . . . . . . . . . . . . . 68
Survey questions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Summary statistics of HCAHPS variables. . . . . . . . . . . . . . . . . 70
Checking validity of PCA assumptions. . . . . . . . . . . . . . . . . . . . . . 74
PCA results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Interpretation of principal components of HCAHPS scores. . . . . . . . . . . . . . . 79
Applying the PCA results to hospital performance improvement. . . . . . . . . . . 81
Research Question 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Research Hypothesis under research question 2 . . . . . . . . . . . . . . . . . . . . 82
Testing research hypotheses with OLS regression. . . . . . . . . . . . . . . . . . . 84
Testing OLS assumptions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Test results for research question 2. . . . . . . . . . . . . . . . . . . . . . . . 86
xii
Effect sizes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Summary of ndings. . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Regression diagnostic tests. . . . . . . . . . . . . . . . . . . . . . . . . 88
Sensitivity analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Research Question 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Process of care quality data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Validating PCA assumptions for process of care quality data. . . . . . . . . . 93
Results from PCA of process of care quality data. . . . . . . . . . . . . . . . 93
Rotating component axes of process of care quality data. . . . . . . . . 94
Interpretation of quality components . . . . . . . . . . . . . . . . . . . 95
Test results for research question 3. . . . . . . . . . . . . . . . . . . . . . . . 98
Process of care quality data by ownership groups in . . . . . . . . . . . 98
Research hypotheses under research question 3. . . . . . . . . . . . . . 98
Validating OLS regression assumptions for quality data . . . . . . . . . 104
Regression results for quality component 1 (heart attack/failure
related) . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Regression results for quality component 2 (pneumonia related) . . . 109
Regression results for quality component 3 (surgical care related) . . 113
Regression results for quality component 4 smoking cessation related 115
Regression results for quality component 5 prevention related . . . . 118
Summary of Regression results for RQ-3. . . . . . . . . . . . . . . . . . 121
Research Question 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Validating OLS regression assumptions for research question 4. . . . . . . . 123
Test results for research question 4. . . . . . . . . . . . . . . . . . . . . . . . 123
Regression results for quality component 1 (heart attack/failure
related) . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Regression results for quality component 2 (pneumonia related) . . . 126
xiii
140
xiv
xv
xvi
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6 Discussion and Conclusion
206
xvii
240
xviii
278
295
298
300
303
304
308
309
xix
310
xx
List of Tables
. . . . . . . . . . . . . . . . . . . . . 71
. . . . . . . . . . . . . . . . . . . . . 76
10
11
12
13
14
. . . . . . . . . . . . . . . . . . 99
15
. . . . . . . . . . . . . . . . . . 99
16
. . . . . . . . . . . . . . . . . . 99
17
. . . . . . . . . . . . . . . . . . 100
18
. . . . . . . . . . . . . . . . . . 100
19
20
21
. . . . . . . . . . . . . . . . . . . . 82
. . . . . . . . 85
. . . . . . . . . 94
. . . . . . . . . . . . . . . . 96
. . . . . . . . . . . . . 109
xxi
22
23
24
25
26
27
28
29
30
. . . . . . . . . . . . . 112
. . . . . . . . . . . . . 116
. . . . . . . . . . . . . 119
. . . . . . . . . . . . . 121
32
33
34
35
36
37
38
39
40
41
. . . . . . . . . . . . . 126
. . . . . . . . . . . . . 129
. . . . . . . . . . . . . 132
. . . . . . . . . . . . . 134
. . . . . . . . . . . . . 137
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
42
43
44
45
RQ5 - Regression results for 30-day risk adjusted mortality rate for heart attack
. . . . . . . . . . 145
xxii
46
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Attack by Owner Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
47
RQ5- Regression results for 30-day risk adjusted readmission rate for heart attack
by owner groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
48
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate for
Heart Attack by Owner Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
49
RQ5- Regression results for 30-day risk adjusted mortality rate for heart failure
by owner groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
50
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Failure by Owner Groups
51
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
RQ5- Regression results for 30-day risk adjusted readmission rate for heart failure
by owner groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
52
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate for
Heart Failure by Owner Groups
53
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
RQ5- Regression results for 30-day risk adjusted mortality rate for pneumonia by
owner groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
54
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rates for
Pneumonia by Owner Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
55
RQ5- Regression results for 30-day risk adjusted readmission rate for pneumonia
by owner groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
56
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rates
for Pneumonia by Owner Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
57
58
RQ5 - OLS Regression results for 30-day risk adjusted mortality rate for heart
attack by patient satisfaction component 1
59
. . . . . . . . . . . . . . . . . . . . . . 167
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Attack with Patient Satisfaction Component 1 . . . . . . . . . . . . . . . . . 169
xxiii
60
RQ5- Regression results for 30-day risk adjusted readmission rate for heart attack
by satisfaction component 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
61
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate for
Heart Attack with Patient Satisfaction Component 1 . . . . . . . . . . . . . . . . . 172
62
RQ5- Regression results for 30-day risk adjusted mortality rate for heart failure
by patient satisfaction component 1
63
. . . . . . . . . . . . . . . . . . . . . . . . . . 173
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Failure with Patient Satisfaction Component 1 . . . . . . . . . . . . . . . . . 175
64
RQ5- Regression results for 30-day risk adjusted readmission rate for heart failure
by patient satisfaction component 1
65
. . . . . . . . . . . . . . . . . . . . . . . . . . 176
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate for
Heart Failure with Patient Satisfaction Component 1 . . . . . . . . . . . . . . . . . 178
66
RQ5- Regression results for 30-day risk adjusted mortality rate for pneumonia by
patient satisfaction component 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
67
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Pneumonia with Patient Satisfaction Component 1 . . . . . . . . . . . . . . . . . . 181
68
RQ5- Regression results for 30-day risk adjusted readmission rate for pneumonia
by patient satisfaction component 1
69
. . . . . . . . . . . . . . . . . . . . . . . . . . 182
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate for
Pneumonia with Patient Satisfaction Component 1 . . . . . . . . . . . . . . . . . . 184
70
71
RQ5 - Regression results for 30-day risk adjusted mortality rate for heart attack
by quality components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
72
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Attack with Process of Care Quality Components . . . . . . . . . . . . . . . 188
73
RQ5 - Regression results for 30-day risk adjusted readmission rate for heart attack
by quality components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
xxiv
74
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate for
Heart Attack with Process of Care Quality Components . . . . . . . . . . . . . . . 191
75
RQ5 - Regression results for 30-day risk adjusted mortality rate for heart failure
by quality components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
76
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Failure with Process of Care Quality Components . . . . . . . . . . . . . . . 194
77
RQ5 - Regression results for 30-day risk adjusted readmission rate for heart failure
by quality components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
78
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Attack with Process of Care Quality Components . . . . . . . . . . . . . . . 198
79
RQ5 - Regression results for 30-day risk adjusted mortality rate for pneumonia
by quality components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
80
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Pneumonia with Process of Care Quality Components
81
. . . . . . . . . . . . . . . . 201
RQ5 - Regression results for 30-day risk adjusted readmission rate for pneumonia
by quality components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
82
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate for
Pneumonia with Process of Care Quality Components. . . . . . . . . . . . . . . . . 204
83
84
85
86
87
88
89
90
. . . . . . . . . . . . 245
xxv
91
92
93
94
xxvi
List of Figures
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
. . . . . . . . . . . . . . . . . . . . . . . . . 95
. . . . . . . . . . . . . . . . . . . . . . . . . 97
. . . . . . . . . . . . . . . . . . . . . . . . . 97
. . . . . . . . . . . . . . . . . . . . . . . . . 98
10
11
12
13
14
15
16
17
18
19
20
21
xxvii
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
RVF plot for Heart Attack Mortality Rate - Quality Components . . . . . . . . . . . . 187
44
45
RVF Plot for Heart Attack Readmission Rate - Quality Components . . . . . . . . . . 190
46
47
RVF Plot for Heart Failure Mortality Rate - Quality Components . . . . . . . . . . . . 193
48
RVF plot for Heart Attack Mortality Rate - Quality Components . . . . . . . . . . . . 196
. . . . . . . . . . . . . . . . . 144
. . . . . . . . 174
xxviii
49
ACPRplot for Heart Failure Readmission Rate and Quality Component 1 . . . . . . . . 197
50
ACPRplot for Heart Failure Readmission Rate and Quality Component 5 . . . . . . . . 197
51
52
53
54
Unied Field Chart - Blow-up of healthcare portion in the upper left section . . . . . . 261
55
56
57
58
59
60
61
. . . . . . . . . . . . . . . . . . 266
. . . . . . . . . . . . . . . . . . 268
xxix
Institute of Medicine
JCAHO
AHRQ
AMA
QI
Quality Indicator
HHS
HQA
AMI
CHF
PN
Pneumonia
PSI
TQM
MMA
CMS
MUM
xxx
APA
URL
ES
KMO
CLES
EFA
PCA
OLS
BLUE
ANOVA
ANalysis Of Variance
XLSTAT
Stata
VIF
UFC
RAC
Chapter 1
Study Overview
This chapter introduces some of the problems with the US healthcare system, its high cost
and the need to improve the quality consistent with the high standard of living. The purpose and
signicance of the study are brought out in subsequent sections. The theoretical framework,
research questions, limitations, delimitations and study assumptions are discussed. The chapter
ends with a brief description of how the study is organized.
Charakas concept of healthcare quartet. Charaka, the ancient Vedic physician of India
(Valiathan, 2007) considered healthcare as made of a quartet composed of four parts: 1) patient,
2) physician, 3) attendants and 4) treatment and compared them to four legs of an animal. The
four parts have to function together to enable the animal to move. Charaka considered that
healthcare is accomplished by the balanced functioning of all the four parts. The present day
healthcare in the US is not marked by such a balance and consequently is facing several problems.
Problems of healthcare in US. Healthcare in the US has become hugely expensive, but
the quality is not with a commensurate level of the high cost. Also, cases of medical errors,
infections acquired during hospital stay, incidents affecting patient safety and malpractice cases
that have been reported have caused public distrust that has forced the government and regulatory
bodies to advise hospitals to embark on performance and quality improvement activities.
High cost. Public concern over ever increasing health costs is rising in US. In 2008,
total national health expenditure in the US was expected to rise by 6.9%two times the rate of
ination. Total spending was $2.3 trillion in 2007 which translates to $7681 per person. Total
healthcare spending represented 16.2 percent of GDP, representing an increase from 15.9% in
2007 (CMS, 2008). This is the highest per capita spending on healthcare in the world. With health
insurance premiums doubling every 5 years, DoBias and Evans (2006) predicted that a familys
annual costs for health insurance would be $22,000 by the year 2010. The Money magazine
(CNN, 2012) reported that a typical family of four under an employer plan, spent more than
$20,000 on healthcare in 2012, quoting the consulting rm Milliman inc. Today (2012) reported
that the US median household income at the end of 2011 was $ 51,413. This shows that the
average family in US had to spend nearly 40% of their income to meet healthcare costs in 2012.
Inefciencies. At the same time, many patients think that the quality of healthcare
services is not with a commensurate level of the high cost. The same report (CMS, 2008)
mentions Experts agree that our health care system is riddled with inefciencies, excessive
administrative expenses, inated prices, poor management, and inappropriate care, waste and
fraud. A survey conducted by ABC News, the Kaiser Family Foundation and USA Today found
that most Americans are dissatised with the healthcare system. An overwhelming 80% think that
the costs are too high, while 54% are dissatised with the quality of healthcare (Enzi, 2007).
These evidences point to an urgent need to reduce cost and enhance the quality of healthcare. An
interview conducted among healthcare opinion leaders suggested that they had a strong belief that
comprehensive strategiesincluding nancing reform, a robust information technology
infrastructure coupled with changes to work design and culture, and alignment between nancial
and clinical accountabilitycould result in a more efcient health care system (Greiner &
Starkey, 2006).
Errors and patient safety. Kohn, Corrigan, and Donaldson (2000) estimated that
between 44,000 and 98,000 preventable deaths occur every year as a result of errors in the
health care system and preventable health care-related injuries result in costs of between $17 and
$29 billion annually . Given the preceding scenario, many policy makers have begun to question
the value that is being delivered by the U.S. health care system to the public.
IOM reports. Institute of Medicine (IOM) concluded that the American healthcare
system is in a serious state of disrepair and is in need of transformation. The full extent of the
problems with the U.S. healthcare service delivery system is outlined in a series of IOM reports
that consider the components of medical safety, quality of care, performance measurement,
quality improvement, and workforce capacity. Together these reports clearly establish that (a)
quality of care is well below the standard that the U.S. population expects and deserves, and (b)
the sources of the problems are not a lack of goodwill or right intention but rather can be found in
the fundamental construction of the healthcare system. In response, the IOM has advocated the
strategic redesign of this structure and many components of the system (Daniels, England, Page,
& Corrigan, 2005).
Joint Commission efforts. In early 1990, in response to an increasing awareness about
inefciencies in the healthcare industry, the Joint Commission, a private sector nonprot
Organization to accredit hospitals, made changes in their hospital accreditation policy, requiring
hospitals to implement performance improvement measures. Formerly, this organization was
called as the Joint Commission on Accreditation of Healthcare Organization (JCAHO).
to health.
.
1
R
Transcendental Meditation technique
, Maharishi TM-Sidhi program, Maharishi Vedic Approach to Health,
Maharishi Ayur-Veda , Science of Creative Intelligence, Maharishi Vedic Science, and Maharishi University of
Management are registered or common law trademarks licensed to Maharishi Vedic Education Development
Corporation and used with permission.
and Pneumonia (PN). The HQA data provide hospitals with performance benchmarks and
can be used to guide quality improvement. The percentage of cases a hospital treated as
recommended measures its process quality. This study uses the HQA data for
process-of-care quality data.
3. Outcomeoutcomes offer evidence of patients health status after treatment. This study
considers the following published measures for comparison: Risk adjusted 30 day mortality
rates from heart attack, heart failure and pneumonia.
Qualities of healthcare staff. Healthcare is delivered by humans and, therefore, their
qualities and ways of approaching the patients, determine the quality of healthcare. Often, this is
ignored in healthcare management research. The importance of empathy of healthcare
practitioners towards patients has been brought out by Epstein and Hundert (2002) and Larson
and Yao (2005). The role of the healthcare practitioners qualities has been emphasized by
Charaka in his monumental work Charaka Samhita : friendship, compassion, joy in serving, and
equanimity. Charaka recommended that, at the time of selecting students, only those who show
signs of possessing these qualities should be selected. The instructors should encourage their
students to develop these qualities during their education. In a similar vein, Larson and Yao
(2005) have recommended regular training during medical education in making conscious efforts
to develop their empathetic abilities. Chapter 7 of the dissertation will discuss this aspect and
connections with Maharishi Vedic Science.
Effects of ownership type. Hospitals may be owned by prot making corporations or
government or not- for-prot organizations. According to the theory developed by Hansmann
(1980), any differences among ownership types should vanish under managed care. Managed care
plans like Medicare pay hospitals on a prospective basis, and there is risk sharing between the
plan and the provider. Over time, Hansmann argued that only the most efcient hospitals will
thrive and survive. Consequently, the incentives to manage a hospital efciently provided by
managed care will reduce the nonoptimal behavior of all hospitals. However, a few studies have
reported signicant differences between ownership types (e.g., Baker et al., 2000). The current
study assessed the relationships between patient satisfaction, process quality, patient outcomes
and hospital ownership type.
Patient satisfaction.
Improving customer satisfaction is the goal of all quality management concepts. Total
Quality Management (TQM) concepts guide us by focusing on quality and customer satisfaction.
TQM concepts believe that customers ultimately dene quality, and if customers are satised with
a product or a service, it is of high quality and adds value to customers. Monitoring patient
satisfaction has become a standard operating procedure in most healthcare organizations,
especially with the implementation of public reporting of patient satisfaction ratings for hospitals.
Higher patient satisfaction is expected to lead to higher customer volume per market research
literature that suggests that customer satisfaction leads to customer loyalty, e.g.(Hallowell, 1996).
Patient satisfaction is a complex construct and its measurement and interpretation differ widely
with the demography of patients, nature of ailment, patients level of education, income and
maturity and many more such confounding variables.
Despite this, patient satisfaction ratings have become valuable for hospitals for the
following reasons.
1. HCAHPS ratings are publicly published and regularly updated on the Medicare web site.
Low ratings could affect the corporate image and funding of hospitals.
2. Hospitals use the HCAHPS ratings to set corporate goals, monitor the performance,
identify areas for improvement, and in quality assurance-type activities.
3. Marketing places a strong emphasis on customer satisfaction.
4. Many hospitals routinely outsource services such as emergency services to outside agencies
and use the patient ratings as a measure of performance in these contracts.
5. Patient satisfaction is taken as a quality measure by accreditation agencies. It can be tied to
quality metrics, including length of stay, patient safety indicators and core measures
6. Medicare reimbursements to hospitals are being linked to patient satisfaction scores from
HCAHPS surveys from scal year 2008. Hospitals need to monitor their HCAHPS ratings,
to avoid possible reduction in their Medicare payments.
7. Increased patient satisfaction may be associated with mental satisfaction and a feeling of
wellness that could help recovery. Satised patients are more likely to react positively and
subsequently benet to a greater extent from their treatment. This is supported by empirical
evidence as reported by Guldvog (1999).
Improving patient satisfaction. Because of the importance of patient satisfaction,
hospitals face the problem of how to increase patient satisfaction. The current national average
reported in HCAHPS survey report for patients giving a high overall rating of 9-10 to a hospital
on their visit is 65%. Patients who responded that they would denitely recommend the hospital
to their friends and relatives, averaged 68%. If a hospital is reported to score below the national
average, it runs the risk of a cut in the government Medicare payments with the new performance
based payments system. Also, since patient satisfaction ratings are publicly reported, patients
may opt for a hospital with better rating, and this could reduce the patient load of the hospital.
Hospitals need to identify the areas to focus on to improve patient satisfaction and this study
attempts to answer this by using a factor analysis to determine the dimensions of patient
satisfaction.
10
Process-of-care quality.
Though it is difcult to dene or quantify quality in healthcare, it was found necessary to
do so to assess and improve a hospitals performance. Technical process quality refers to the
appropriateness of the treatment. Poor quality can mean too much care with unnecessary tests,
medications or procedures, or too little care with omitting appropriate tests or procedures or
wrong care with procedures or medications that should not have been given. This would be
difcult to measure, and hospitals may have to do an evaluation in selected cases for quality
assurance.
Another way to measure process quality is to determine whether the provided care meets
professional standards. This is done using the quality indicators (QIs) prescribed by AHRQ.
HQA uses a subset in accordance with government guidelines. Using the HQA data, Jha, Orav,
Zhonghe, and Epstein (2007) found that higher quality is associated with lower risk adjusted
mortality rates. Several other studies seem to conrm this. However, Isaac and Jha (2008)
reported inconsistent and usually poor associations between the patient safety indicators and
HQA quality measures. There have been no empirical studies published on the association
between patient satisfaction and process of care quality measured by QIs. This study attempts to
nd if such a relationship exists and its nature. The current study tested the research hypothesis
that hospitals compromise quality in favor of higher patient satisfaction.
Ownership type.
Hospitals may be government or privately owned for prot or owned by not-for-prot
voluntary organizations or church groups. Mobley suggested that church-owned hospitals
consider that their chief mission is to provide indigent care and that it takes precedence over
nancial performance (1997). Thomson Reuters surveys US hospitals and lists the top hundred
hospitals every year (Reuters, 2010). From this site, I listed the hospitals that repeat more than ten
times since their rst such list in 1992. There are nine such hospitals and all of them are voluntary
11
nonprot hospitals and two of them are owned by church groups. This points to the strong
inuence of ownership on hospital performance.
A few studies have tried to study the relationships between ownership type and patient
satisfaction and quality, but these had limited scope (e.g., Baker et al., 2000; Eggleston, Shen,
Lau, Schmid, & Chan, 2008). The present study aims at nding how patient satisfaction, quality
and outcomes vary across ownership types. Particularly, if signicant differences are found in
favor of hospitals run by church groups, this would bring out the connection between spirituality
in organizations and performance. Heaton, Schmidt-Wilk, and Travis (2004) suggest that
spirituality can be used in managing change. Change management is required for carrying out
performance improvement in hospitals.
12
Third purpose is to test the relationship between patient satisfaction, quality and
outcomes. The outcome data are published by HHS as risk adjusted 30 day mortality and
readmission rates for heart attack, heart failure and pneumonia.
Fourth purpose is to test the relationships between hospital ownership type with patient
satisfaction, quality and outcomes.
13
The study will evaluate the relationships between patient satisfaction, quality, outcome
variables and ownership group. This has not been studied at length. K. White and Ozcan (1996)
showed that church owned hospitals were more efcient than secular nonprot hospitals, using a
California sample. However, Thornlow and Stukenborg (2006) reported inconsistent relationship
between ownership type and quality of care showing conicting study ndings. This study will
analyze the relationship using the comprehensive CMS data.
Denition of Terms
Operational denitions.
Patient Satisfaction.
Patient satisfaction is a construct to measure the patients perception of the healthcare
service quality. In terms of Donabedians structure-process-outcomes (1988) framework for
assessing healthcare quality, patient satisfaction measures the patient perception of structural and
process qualities of healthcare. While there are many available instruments to measure patient
satisfaction, HCAHPS is the most frequently used instrument for hospital comparisons. CMS
(2010) gives the standards used in HCAHPS.
Process of care quality.
A way to measure process quality is to determine whether the provided care meets
professional standards. This assessment is done by creating a list of quality indicators that
describe a process of care that should occur for a particular type of patient or clinical
circumstance and then evaluating whether the patients care was consistent with the indicators.
AHRQ has formulated a very large number (nearly 500) of quality indicators (QIs) concerning
various aspects of healthcare such as prevention, inpatient, safety and pediatrics (AHRQ, 2011).
Out of these, the following were adopted for hospital comparison by CMS in consultation with
hospitals and Joint Commission:
14
Constitutional denitions.
Certain technical terms and expressions used in the dissertation are described here.
CMS data.
CMS is the Centers for Medicare & Medicaid Services, a federal government organization
that manages the Medicare and Medicaid programs.
15
HCAHPS survey.
CMS and the Agency for Healthcare Research and Quality (AHRQ) developed the
HCAHPS survey questionnaire. It is a core set of questions that hospitals can combine with a
customized group of hospital-specic items if the order of the questions is not changed, and
hospital-tailored questions are added at the end. The National Quality Forum, established to
standardize healthcare-quality measurement and reporting, formally endorsed HCAHPS in May
2005.
Originally, the conceptual framework of the survey drew from the following domains of
quality health care proposed in the IOM report Crossing the Quality Chasm: A New Health
System for the 21st Century (IOM, 2001):
1. Respect for patients values
2. Attention to patients preferences and expressed needs
3. Coordination and integration of care
4. Patient information, communication and education
5. Physical comfort
6. Emotional support
7. Involvement of family and friends
8. Transition and continuity of care
9. Access to care
After pilot tests, the original set of questions was simplied. Two domains (1, and 7) were
dropped because these two are difcult to measure, and not fully under a hospitals control.
16
17
mortality and readmission outcome measures data downloaded are for the period July, 2005 to
June, 1008. The downloaded data covered the following:
1. Process of care and outcome Quality measures, 28 in number covering heart attack, heart
failure, pneumonia, surgical care improvement and childrens asthma care.
2. Mortality measures, (6 in number) cover hospital 30-day death and readmission rates for
heart attack, heart failure and pneumonia.
3. Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) Survey
results. This survey measures 29 indicators of patient satisfaction in response to 22
questions. The HCAHPS Patient Satisfaction Surveys cover several aspects of patient
perception of healthcare given to them.
Theoretical Framework
Two widely used theories in healthcare research are used to in this study. These are as
follows:
Hansmanns theory on the role of nonprot enterprise. Hansmann brought out the
theory of the role of nonprot enterprise to explain the difference between not-for-prot and for
prot hospitals. Hansmann divided nonprots loosely into two broad categories: donative
nonprots, which derive a substantial portion of their income from grants or donations, and
commercial nonprots which derive most or all of their income by selling their services directly
to consumers. However, with the trend towards managed care in the United States, Hansmann
(1980) argued that any differences between ownership types should vanish under managed care
(like Medicare). Managed care plans pay hospitals on a prospective basis, and there is risk
sharing between the health plan and the provider. Over time, only the most efcient hospitals will
thrive and survive. Consequently, the incentives to manage a hospital efciently provided by
managed care will reduce the non-optimal behavior of all types of hospitals.
18
Some studies (e.g., Sloan, Trogdon, Curtis, & Schulman, 2003; E. F. X. Shortell Stephen
M. & Hughes, 1988) reported ndings that broadly agree with this theory. Nevertheless, some
other studies have reported signicant differences between not-for-prot and for-prot hospitals
(e.g., Milcent, 2005; Yuan, Cooper, Einstadter, Cebul, & Rimm, 2000). This study tested the
applicability of Hansmanns theory over a wider range of hospital ownership groups.
Donabedians Structure-Process-Outcomes theory. It is difcult to dene quality in
healthcare because medical science is not an exact science, so no one can guarantee a fully
successful outcome for any treatment. Also, the human body is not a standardized mechanism,
and there are human variations to deal with in healthcare. Donabedian (1988) brought out a
theoretical framework to measure healthcare quality.
This structureprocessoutcome approach is used extensively in published research in
healthcare for evaluating quality. Structural characteristics such as size, ownership, staff, and
technology are considered necessary but not sufcient elements in the delivery of health services
and are typically considered indirect measures of quality. Process is the clinical service provided
to a patient, the interaction resulting in a treatment to the patient. The outcome is the result of the
process to the patients health and well-being. Within this framework, patient satisfaction is the
patients perception of the structural and process qualities. The process-of-quality measures
represent the process quality. The outcome variables measure the status of the patients after
treatment. This study uses this framework to collect data on healthcare quality and test the
research hypotheses.
Research Questions
This research is focused on probing the following research questions:
1. Which dimensions of the patient perceptions of healthcare have a major impact on patient
satisfaction? A Principal Component factor analysis on the HCAHPS data will help to
identify these dimensions.
19
2. What is the inuence of the type of hospital ownership (government, church groups etc.)?
This will identify other issues that are usually ignored such as the inuence of spirituality in
healthcare organizations that could affect patient satisfaction.
3. What is the relationship between process of care quality measures and patient satisfaction?
This will bring out the possible conicts and modications that may be needed in patient
satisfaction measurement scales.
4. What is the relationship between process of care quality measures and types of hospital
ownership? This will again bring to focus other components that may affect the quality.
5. What is the relationship between outcome of care measures (mortality and readmission
rates) and patient satisfaction and type of hospital ownership? Any positive correlation
between patient satisfaction and the outcomes would highlight the positive effect of mental
satisfaction on health outcomes. The relationship of ownership on outcomes will bring out
the inuence of organizational characteristics.
Limitations
The basic limitation of the study is its dependence on analysis of archival data maintained
by CMS. This database is used by patients to make healthcare decisions, and by government
agencies to evaluate a hospitals performance for their performance linked payments system. The
data collection methods have been carefully studied and corrected by experts in the eld. So, the
data is highly reliable and is expected to yield reliable results. Keller et al. (2005) reported that
the scales used in HCAHPS had acceptable levels of reliability to discriminate between hospitals
(> 0.7), using standard psychometric methods to assess the reliability and internal consistency.
The content validity and criterion-related validity of the instrument are also evident from
this reports nding that twenty-four of the survey items were signicantly and uniquely related to
patients general hospital experience at p < 0.0001. Only four items had nonsignicant
relationships with the global hospital rating. The major potential threats to the study are validity
20
issues associated with the use of archival data. These are discussed under Cook and Campbells
validity topology (Shadish, Cook, & Campbell, 2002).
Statistical conclusion validity issues
Selection - Potentially, dissatised patients are more likely to respond than satised patients.
This is a general threat to all satisfaction surveys. This violates the assumption of statistical
tests that both satised and dissatised patients are equally likely to be in the sample.
HCAHPS has tried to control this bias by randomizing the selection of patients to be
surveyed, but, the bias is still possible because the response is voluntary.
Experimenter bias - Surveys are compiled by hospitals themselves and reported to CMS. This
violates the assumptions of statistical tests that the survey results were not inuenced by the
partisan behavior of the survey administrator and this makes the results unreliable.
HCAHPS has controlled for this bias by imposing certain program restrictions and
compliance requirements in the HCAHPS quality assurance guidelines. HCAHPS also
undertakes random audits in hospitals to make sure that they comply with the guidelines.
Subject bias - Response could be affected by gender, race, education level, motives and
physical, mental or medical condition or expectations of the patients. This heterogeneity of
units could be a source of increased variance. OMalley, Zaslavsky, Elliott, Zaborski, and
Cleary (2005) have studied the need for case-mix adjustments. They found that the most
important case-mix variables are: hospital service (surgery, obstetric, medical), age, race
(non-Hispanic black), education, general health status, speaking Spanish at home, having a
circulatory disorder, and interactions of each of these variables with service. They
concluded that while case-mix adjustment can lead to important reductions in the bias, in
comparisons between hospitals, its impact on hospital ratings is small. To control this,
HCAHPS makes case-mix corrections in the reported results.
21
Attitude bias - Patients personal bias because of gender, race, or color of the hospital staff
could potentially bias the response. The randomized selection of the survey respondents
controls this.
Survey mode bias - Differences in the modes of surveymail, telephone, mail combined with
telephone follow-up (mixed mode), and active interactive voice response (IVR)may
cause extraneous variance in the survey results. Elliott et al. (2009) have studied Effects of
survey Mode, patient Mix, and non-response on HCAHPS and found that mode effects are
consistent across hospitals and are generally larger than total patient-mix effects. HCAHPS
partly controls this bias by a process that randomizes the mode. In addition, survey mode
corrections are applied on the reported results.
Communication bias - Differences in understanding the survey questions are a potential threat
that could lead to extraneous variance in the experimental setting. HCAHPS procedure is
controlling this by various measures such as simplifying the questions, having a Spanish
version, and having an explanation for each question.
Internal validity issues
Internal validity is related to the question Did the experiment make a signicant
difference to the result? Even though, the respondents were randomly selected and were assured
of anonymity, it may be possible that some respondents may have felt elated by the selection and
given higher ratings than their true feelings. This is controlled by HCAHPS by only taking the
results for a hospital with more than 100 completed surveys.
External validity issues
Some hospitals are not reporting the results. Also, some hospitals (especially smaller
sized) are not covered if there are fewer than 100 completed surveys. External validity issues are
minimal because all US hospitals are subject to nearly identical rules and protocols. Nevertheless,
22
extending the results to other countries may not be valid because of cultural differences and
different operating conditions.
Construct validity issues
Both content and criterion related validity of HCAHPS has been established by several
studies E.g. (Keller et al., 2005) cited above. Both patient satisfaction and quality are well
established constructs in the healthcare industry and has been recognized by government bodies.
Delimitations
There are three delimitations of this study.
Patient safety indicators were omitted from the analysis motivated by a desire to get a better
understanding of patient satisfaction, quality and outcomes. Also, Isaac and Jha (2008)
reported inconsistent and usually poor associations between the patient safety indicators
and HQA quality measures.
While there are other programs that capture patient satisfaction and quality data, this study
uses only CMS data. This delimitation is enforced because CMS data is reliable,
consistently collected and covers almost all US hospitals. The CMS data is publicly
available and can be used without restrictions. Several published studies have used CMS
data.
The study is delimited to in-patient data. This is because out-patients are not covered with
the CMS data. HQA quality data, also shared with the Joint Commission, covers all
in-patients including Medicare recipients.
Assumptions
HCAHPS results are adjusted for the effects of both mode of survey administration and
patient-mix on the basis of national averages on a nationwide random sample. An exact
23
replication of patient-mix adjustment at hospital level is not possible and so, the study
assumed that these adjustments do not cause any bias.
HCAHPS survey is administered to a random sample of adult patients between 48 hours
and six weeks after discharge; the survey is not restricted to Medicare beneciaries. This
study assumed that patient ratings are not inuenced by the time that lapsed after discharge.
In contrast, Jackson and Chamberlin (2001) have shown that patient satisfaction surveys
need to take into account the sampling time frame.
While HQA quality data cover all in-patients, the risk adjusted mortality and readmission
rates are limited to Medicare beneciaries. CMS processes the enrollment and claims data
to produce these statistics. While relating outcome variables with patient satisfaction and
quality, the study assumed that no signicant differences between the general population
and Medicare beneciaries in quality and patient satisfaction. It is not possible to test the
difference between the Medicare group and the general population as the respondent details
are not published.
24
25
Chapter 2
Literature Review
Chapter Overview
Healthcare cost is high in the US, making the citizens spend annually $ 7600, the highest
per capita spending on healthcare in the world. However, the quality of medical care is not at a
commensurate level with the cost. The average life expectancy in the US is lower than many
industrialized countries, pointing to the need for improving the quality of healthcare. There are
also problems with medical errors, safety incidents, medical malpractice and infections acquired
during hospital stay. Therefore, many researchers have studied US healthcare issues at length. In
this chapter, a review of literature signicant to this study is given to gain a comprehensive
understanding on the aspects that the present study can contribute to this eld. The body of this
chapter is structured in nine sections:
The second section gives details of the strategy adopted for searching.
Third section focuses on problems with US healthcare in general, particularly the high cost
coupled with the need to improve quality.
Fourth section deals with healthcare errors and incidents affecting patient safety in US
healthcare.
Fifth section deals with the reliability, validity and adjustments of HCAHPS survey data on
patient satisfaction.
Sixth section reviews selected studies on quality data.
26
The seventh section reviews selected studies on the effects of hospital ownership type.
The eighth section reviews studies on relationships between patient satisfaction, quality and
outcomes.
The last section concludes the chapter with the discernible gaps in the knowledge that are
taken up in this study.
27
conrmed this. Palosky and Singh (2011) reported a bench employer survey and reported that the
average annual premiums for family health benets exceeded $15,000 in 2011. Since 2001,
family premiums increased by 113% while the growth in wages was only 34% and 27% in
ination.
Healthcare errors
Several studies have conrmed the existence of a signicant number of preventable
adverse events in hospitals (e.g. Phillips et al. (2004), Romano et al. (2003)). Patient safety
indicators (PSIs) formulated by AHRQ include the adverse events that might compromise the
patient safety. Further, the Institute of medicine estimated that between 44,000 and 98,000
preventable deaths occur every year as a result of errors in the healthcare system and
preventable health care related injuries result in costs of between $17 and $29 billion annually
(Kohn et al., 2000). This estimate was questioned by Hayward and Hofer (2001), who reviewed
111 hospital deaths and estimated that only 0.5% of these patients could have lived for 30 days or
more if the care had been optimal. From their ndings, they felt that the IOM gure might have
been exaggerated. However, Leape (2000) argued that the IOM error gures are not exaggerated
quoting that the Centers for disease control and prevention estimates that 500,000 surgical
infections occur every year suggesting 20,000 deaths from this cause alone assuming a mortality
rate of 4.3%.
Besides outright errors, adverse incidents have been experienced by patients. These are
dened as unintended injuries caused by medical management rather than the disease process.
The Harvard medical practice study found that 3.7% of hospital admissions (1133 cases) led to
adverse events, in a sample of 30,195 randomly selected hospital records (Leape et al., 1991). A
similar study of two acute hospitals in Greater London area reported that 10.8 % (in 110 cases)
experienced an adverse event. About half these events were judged preventable, and a third of
these adverse events led to disability or even death (Vincent, Neale, & Woloshynowych, 2001).
Thus, the economic consequence of these preventable adverse events is enormous. For example,
28
Bates et al. (1997) estimated the annual costs of all adverse drug events for a 700-bed teaching
hospital to be $ 5.6 million and of preventable adverse drug events to be $ 2.8 million. However,
patient safety and healthcare errors are not in the scope of the present study.
Year
et 2009
Findings
Elliott and others evaluated the need for survey mode adjustments and
concluded that valid comparisons of hospital performance require that
reported hospital scores be adjusted for survey mode and patient mix.
table continued on the next page
29
Year
Findings
liott, Hep-
ner, Keller,
and Hays
Hargraves,
2003
Hays, and
measures and concluded that the plan level reliability and internal
Cleary
Goldstein
and Fyock
2001
30
Year
2005
Findings
OMalley et al analyzed HCAHPS Hospital survey data to assess the
Zaslavsky,
Elliott,
et
al.
OMalley,
2005
Zaslavsky,
Hays, et al.
structures for the HCAHPS data collected in 2003 and concluded that six
factors provided the best description of inter-item covariance at the patient
level. Hospital-level factor structures also differed across services as much
variation in quality reports was explained by service as by composite.
Marshall,
2001
Elliot,
Morales,
Spritzer,
and Hays
31
Year
2001
Findings
Morales et al examined racial and ethnic group differences using a sample
Elliott,
Weech-
data and found that racial/ethnic minority groups other than Asians/Pacic
Maldonado,
Spritzer,
and Hays
Rothman,
2008
Park,
Hays,
Edwards,
They concluded that the additional discharge information questions and the
and
Dudley
Teleki
al.
et 2007
Many
32
Year
et 2008
al.
Findings
Davies et al evaluated the use of a modied HCAHPS survey to support
quality improvement in a collaborative focused on patient-centered care,
assess subsequent changes in patient experiences, and identify factors that
promoted or impeded data use. They concluded that small measurable
improvements in patient experience may be achieved over short projects.
Sustaining more substantial change is likely to require organizational
strategies, engaged leadership, cultural change, regular measurement and
performance feedback and experience of interpreting and using survey data.
end of table
Year
Findings
Fine conducted a retrospective cohort analysis using data from the Medicare
Galusha,
Petrillo,
race, fever, nurse-bed ratio, hospital size, teaching status, and southern
and
location are among the major patient and hospital characteristics associated,
Meehan
33
Year
2004
et al.
Findings
Havranek et al studied Medicare patients hospitalized with heart failure.
They used 2 quality measures across the United States, using Bayesian
technique and nonlinear hierarchical models to assess for associations
between the quality indicators and provider and hospital characteristics
independent of patient characteristics. They concluded that characteristics
of providers and hospitals explain in part the geographic variation in
guideline-based care for elderly patients with heart failure.
Weiner et 2006
al.
Vos et al.
2009
34
35
Duggan (2000) found that decision makers in private not-for-prot hospitals are no more
altruistic than their counterparts in prot-maximizing facilities and that public medical spending
had not improved health outcomes.
Kessler and McClellan (2002) found that the effects of ownership type on medical
productivity are quantitatively important. Areas with a presence of for-prot hospitals have about
2.4% lower levels of hospital expenditures, but virtually the same patient health outcomes. They
concluded that for-prot hospitals have important spill-over benets for medical productivity.
Harrington, Woolhandler, Mullan, Carrillo, and Himmelstein (2001) concluded that
investor owned nursing homes compromise the quality of care as a result of a multivariate
analysis of 13,693 facilities.
36
satisfaction data, HQA quality data and HHS outcome data. This study explores if such
differences exist among other ownership types.
37
Gillies (2006) carried out a multivariate regression cross-sectional analysis of 272 health
plans and concluded that the type of delivery system used by health plans is related to many
clinical performance measures but is not related to patient perception of care.
Alazri and Neal (2003) carried out a study to determine whether there is an association
between satisfaction in patients with Type 2 diabetes and the outcome of their diabetic care, and
to determine the contribution of different aspects of satisfaction with the primary care. They
concluded that processes that can act to increase patient satisfaction may be contributing to
improved clinical outcomes.
Guldvog (1999) explored association of patient satisfaction with outcome among patients
with angina pectoris and found that patient satisfaction with medical treatment (P=0.002) and
with information (P=0.03) were associated with improved physical and mental health-related
quality of life.
38
The search of academic databases was conducted using a combination of search terms
such as healthcare studies, HCAHPS survey, CMS, hospital quality analysis, patient
satisfaction, ownership type effects in peer-reviewed articles published after 2000. If some
relevant sources published before 2000 were found, these were obtained by subsequent searches.
In all, about 400 references were collected and used. JabRef Ver. 2.6 was used as the reference
management software. It was found to be particularly useful because of its support to BibTeX
used by LaTeX for generating bibliography.
Conclusion
Some gaps in the present studies were identied as follows:
Interaction of ownership type with patient satisfaction, quality and outcomes has not been
studied using CMS data.
The main dimensions of the HCAHPS patient satisfaction have not been identied. Some
studies (E.g. Jha et al. (2008)) focused primarily on the fraction of patients who rated the
hospital in the highest category (9 or 10 on a scale of 0 to 10) as the primary indicator of
patient satisfaction. This may not be accurate. Also, hospitals need to know the principal
dimensions of patient satisfaction in order to nd the areas to focus in order to improve it.
Similarly, principal components of quality have not been addressed by current research.
Relationships between quality of care and patient satisfaction have not been explored.
Relationships between outcome variables and quality of care have been studied using a
subset of the data. In contrast, this study will use the full data.
Relationships between patient satisfaction and outcome variables have been studied, using
limited data but this study will use comprehensive data to analyze the relationships.
Details of the methodology used in this study are discussed in Chapter 3 of the report.
39
Chapter 3
Methodology
Chapter Overview
In this chapter, the research design and methodology that was used for data collection and
analysis are discussed. The primary goal of this study was to test the research questions relating
to patient satisfaction, quality, outcome and ownership of hospitals. Archival data from the
Hospital compare database were used to this end. The methodology employed to test the research
questions is presented in this chapter. The chapter is organized into ve sections: (a) research
design and approach, (b) description of data, (c) analysis of data, (d) research questions and the
hypotheses that were tested, and (e) summary.
40
Description of Data
Archival data retrieved.
The Hospital Compare database was downloaded from the following Uniform Resource
Locater or Universal Resource Locater (URL):
http://www.hospitalcompare.hhs.gov. The download covers patient satisfaction data
from HCAHPS surveys, process of care quality data from HQA and outcome measure data from
HHS. The aims of including the HCAHPS survey data in the hospital compare website was to
help patients to make informed health care decisions, help healthcare providers for monitoring
their quality of care, and help government agencies to launch a pay for performance scheme
under the Decit Reduction Act of 2005. The study analyzed the available data to identify
concrete steps for hospitals and providers in their performance improvement.
Participants.
The HCAHPS survey is administered to a random sample of adult patients across medical
conditions between 48 hours and six weeks following discharge. The survey is not restricted to
Medicare beneciaries. Participating hospitals may either use an approved survey vendor, or
collect their own HCAHPS data (if approved by CMS to do so). To accommodate the needs of
hospitals, HCAHPS can be implemented in four different survey modes: mail, telephone, mail
with telephone follow-up, or active interactive voice recognition (IVR). Hospitals may either
integrate HCAHPS with their own patient survey, or use HCAHPS by itself. Hospitals must
survey patients throughout each month of the year. The survey is available in ofcial English,
Spanish, Chinese, Russian and Vietnamese versions. The survey itself, as well as detailed
information on sampling, data collection and coding, and le submission are contained in the
HCAHPS Quality Assurance Guidelines found at the ofcial HCAHPS web site
www.hcahpsonline.org.
41
42
each hospital is shown in the following three categories: 300, between 100 to 299, and
100. The survey response rate is also available for each hospital. It ranges from 0 % to 80
%. The mean is about 33 %. A sample listing of HCAHPS patient satisfaction survey data
is shown in Appendix A (Tables 87, and 88).
2. Outcome measures, (6 in number) cover hospital 30-day death and readmission rates for
heart attack, heart failure and pneumonia. These rates are risk adjusted percentages for each
hospital. Comparison to US national average rates are also available for each category. The
values for US comparison are: Better, not different, worse, and number of cases too small
to compare. The listing also gives for each hospital: Lower and upper estimates and number
of patients treated. A sample listing of outcome data is shown in Appendix B (Table 89).
3. Process of care and outcome quality measures, 28 in number covering heart attack, heart
failure, pneumonia, surgical care improvement and childrens asthma care. These are
quality measure scores for each hospital. For each measure, percentage of patients given the
required treatment is the score. The patient sample size for each measure that the hospital
submitted is also given. The process of care quality measures for children are also given.
However, the childrens measures were excluded from analysis in this study because only
144 hospitals (3 % of the total) reported the three measures for children. Moreover, these
process measures are not related to outcome and satisfaction measures.
A sample listing of process of care quality data is shown in Appendix C (Tables 90 and
91).
Data preparation.
To maintain referential integrity and reliability of the database, the data maintained in
various at les and Access database was loaded into MySQL database tables designed for this
purpose. The software SQL Maestro for MySQL was used as the interface for ease of use. The
data was checked and reformatted using SQL queries into one Excel workbook that was then
43
loaded into PASW Statistics 18 (Predictive Analytics SoftWare software (formerly known as
SPSS and renamed after acquisition by IBM) for statistical analysis and reporting (e.g., Norusis &
SPSS, 2010; IBM SPSS Inc., 2011). This spreadsheet was later used to load the data in other
statistical software such as Stata.
Threats to validity.
Shadish et al. (2002) have discussed experiments and causation in depth. Following their
terminology, the research design adopted is nonexperimental, because even though presumed
cause and effects are identied, other structural features of experiments are missing such as
random assignment and such design elements as pretests and control groups. The possible threats
to validity are briey examined here using their framework. Later, the threats to validity were
examined in Chapter 6 in the light of empirical evidence.
Statistical conclusion validity issues.
Selection bias Potentially, dissatised patients are more likely to respond than satised patients.
This is a general threat to all satisfaction surveys. HCAHPS has tried to control this bias by
randomizing the selection of patients to be surveyed.
Experimenter bias Surveys are compiled by hospitals themselves and reported to CMS.
HCAHPS has controlled for this bias by imposing certain program restrictions and
compliance requirements in the HCAHPS quality assurance guidelines and by requiring
prior approval. HCAHPS also undertakes random audits in hospitals to make sure that they
comply with the guidelines.
Subject bias Responses could be affected by gender, race, education level, motives and physical,
mental or medical condition or expectations of the patients. This heterogeneity of units
could be a source of increased variance. OMalley, Zaslavsky, Elliott, et al. (2005) have
studied the need for case-mix adjustments. They found the following inuential case-mix
44
variables: hospital service (surgery, obstetric, medical), age, race (non-Hispanic black),
education, general health status, speaking Spanish at home, having a circulatory disorder,
and interactions of each of these variables with service. They concluded that while,
case-mix adjustment can lead to signicant reductions in the bias in comparisons between
hospitals, its impact on hospital ratings is small. To control this, HCAHPS makes case-mix
corrections in the reported results.
Attitude bias Patients personal bias because of gender, race, or color of the hospital staff could
potentially affect the response. HCAHPS controls this by randomized selection of the
survey respondents.
Survey mode bias Differences in survey results by mail, telephone, mail combined with
telephone follow-up (mixed mode), and active interactive voice response (IVR) are likely to
result in extraneous variance in the experimental setting. Elliott et al. (2009) have studied
effects of survey mode, patient mix, and non-response on HCAHPS and found that mode
effects are consistent across hospitals and are generally larger than total patient-mix effects.
HCAHPS partly controls this bias by a process that randomizes the mode. In addition,
survey mode corrections have been applied by HCAHPS on the published results.
Communication bias Differences in understanding the survey questions is a potential threat that
could lead to extraneous variance in the experimental setting. HCAHPS survey procedure is
controlling this by various measures such as simplifying the questions, having a Spanish
version, and having an explanation for each question.
Internal validity issues.
Internal validity is related to the question Did the experiment make a signicant
difference to the result? Even though the respondents were randomly selected and were assured
of anonymity, it may be possible that some respondents may have given lower or higher ratings
45
than their true feelings. It is also possible that some responses might have been affected by prior
experience or knowledge.
This is controlled by HCAHPS to a large extent by requiring hospitals to hold the survey
continuously on a monthly basis and by restricting the response to one per family in a month.
Also, the survey is given two days after the patient is released from the hospital.
External validity issues.
The survey is restricted to in-patients and, therefore, the results may not be extended for
out-patients satisfaction. Threats to external validity concerning in-patients may not be high
although some US hospitals are not covered, because all US hospitals are governed by uniform
standards and protocols. Nevertheless, extending the results to other countries may not be valid
because of cultural differences and different operating conditions.
Construct validity issues.
Both content and criterion related validity of HCAHPS have been established by several
studies e.g. (Keller et al., 2005) cited above. Both patient satisfaction and quality are well
established constructs in the healthcare industry and has been recognized by government bodies.
Analysis of Data
Data analysis answered the research questions covered in the Introduction. This study is
concerned with the structure of the constructs patient satisfaction and quality of care to test
their relationships with outcome variables and hospital ownership. Factor analysis provided the
means to undertake such a structural analysis. HCAHPS surveys involve as many as 29 variables
and process of care quality data involves another 28 variables. The purpose of factor analysis is to
nd out if the observed variables can be explained largely or entirely in terms of a much smaller
number of variables called factors. A factor is a linear combination or cluster of related observed
variables representing a specic underlying dimension as a construct. Dealing with a smaller
46
number of factors makes it easier to analyze the research questions and test the research
hypotheses. Also, these factors could help in identifying the main dimensions which would help a
hospital in taking steps to improve patient satisfaction and quality.
In this study, Exploratory Factor Analysis (EFA) was implemented using the method of
Principal Component Analysis (PCA) because the number of factors necessary to explain
interrelationships of the variables was not known.
The assumptions of PCA are as follows:
Within the variables, there exists a set of underlying factors smaller in number than the
observed variables that can explain the interrelationships among the variables (Kim &
Mueller as cited in Pett et al., 2003).
Initial steps of PCA are performed using Pearson product moment correlations and,
therefore, many of the assumptions relevant to this parametric statistic are applicable e.g.
large sample sizes, continuous distributions and linear relationships among items.
The response categories in HCAHPS are constructed using the trichotomy of permitted
responses: Always, Usually, Sometimes or never. This violates the assumption that
the variables should not be discrete. However, this is compensated by the fact that HCAHPS
values are cumulative percentage scores and can take continuous values between 0 and 100.
Assumption of normality of distribution is not critical if the researchers intent is to
describe the relationships among the variables. On the other hand, if the goal is to identify
the number of factors that underlie the items being examined, multivariate normality is an
issue to be considered (Tabachnick and Fidell as cited in Pett et al., 2003).
No selection bias should be present in the sample. Also, the model needs to be properly
specied. The exclusion of relevant variables or the inclusion of irrelevant variables in the
correlation matrix will affect the factors substantially (Garson, 2008). This assumption is
met by the HCAHPS.
47
Outliers can impact the correlations heavily and so inuential outliers need to be identied
and corrected.
Homoscedasticity is not considered to be a critical assumption of PCA (Garson, 2008).
Absence of high multicollinearity is required. KMO statistics is used to detect
multicollinearity in factor analysis (Garson, 2008).
No perfect multicollinearitySingularity in the input matrix, also called an ill-conditioned
matrix, arises when two or more variables are perfectly redundant. Singularity prevents the
matrix from being inverted and prevents a solution.
Moderate to high inter-correlations are needed to provide conditioned correlation matrix for
PCA.
Adequate sample size is required. At a minimum, there must be more cases than factors.
This is satisfactorily met with the data in this study.
48
HCAHPS data.
A PCA was carried out to identify the main dimensions on which the hospitals need to
focus to achieve higher patient satisfaction. Researchers assumed that patient satisfaction is
associated with overall rating given by the patient and the response to the question: Would you
recommend the hospital to family and friends: Denitely yes? Probably yes? Or no? Jha et al.
(2008) focused on the fraction of patients who rated the hospital in the highest category (9 or 10
on a scale of 0 to 10) as the primary indicator of patient satisfaction. However, this turned out to
be inaccurate. PCA results showed that this question is associated with one of the principal
components (Component 4) accounting for only 12% of the variance. Problems encountered and
taken care of during the PCA on HCAHPS and process-of-care quality data are described in the
following paragraphs.
Multivariate normality. One problem found while doing PCA on HCAHPS was the
violation of the normality assumption. Both the Shapiro-Wilk test and Kolmogorov-Smirnov tests
(Norusis & SPSS, 2010) rejected the null hypothesis of normality. Normality is not believed to be
a critical assumption of factor analysis as discussed by Garson (2008). In addition, the large
sample size (N > 3,500) supports normality assumption. However, normality was carefully tested
and analyzed during hypothesis testing.
Multivariate outliers. For proper PCA, multivariate outliers in the raw data have to be
identied and removed. Otherwise, these outliers can impact correlations heavily (Garson, 2008).
Garson (2008) has given details of identifying multivariate outliers using the Mahalanobis
distance. This procedure is outlined here and was used to identify the multivariate outliers in
HCAHPS data:
1. An Ordinary Least Squares (OLS) regression was done with all 29 HCAHPS variables as
dependent variables and ownership type code as independent variable. OLS assumptions
are discussed in (Wooldridge, 2006, p. 29). Any numerical variable with non-missing
49
values can be used as independent variable because the regression results are not relevant.
This regression is only for nding Mahalanobis distance option in regression to identify
multivariate outliers. The Mahalanobis distance option was used and saved.
2. A new variable was created to get the probability of the chi-square distribution function.
3. All cases with p-value < 0.001 were identied as outliers. 108 such outliers were identied.
4. A binary 0-1 variable was created with value of 1 for outliers with high Mahalanobis
distance and value of 0 for all other cases.
5. A regression was run using this dummy variable as the response variable and the other
variables as explanatory variables.
6. The R2 from this regression would help to check if these outliers are random errors or not.
The value of R2 was 0.15 conrming that the outliers were random errors and would not
affect the PCA. The values were also small, showing that the non-normality may not
affect the PCA results.
Multivariate outliers in HCAHPS data were identied with Mahalanobis distance (MD)
using Predictive Analytics SoftWare (PASW). Out of 4,460 observations, 108 had high MD.
Closer examination of the outliers could lead to insights and possible improvements in HCAHPS
survey questions. These outlier cases were dropped from the data.
Linearity assumption. For testing of the research hypotheses, linear regression based
ANalysis Of Variance (ANOVA) was used. Ramseys Reset test was used to test the linearity
assumption. In cases where the test was signicant, further testing was done using the
Box-Tidwell test. Box-Tidwell transformation linearizes the relationship by nding
transformations of the explanatory variables (Cohen, Cohen, West, & Aiken, 2003). Box-Tidwell
regression procedure is implemented in Stata 11 which gives the signicance of the nonlinearity
in the model (ATS, 2012).
50
51
Research question 1.
Which dimensions of the patient perceptions of healthcare have the greatest impact on
patient satisfaction?
The results of this analysis brought out the principal dimensions for hospitals to focus on
for improving their patient satisfaction ratings. Though the HCAHPS surveys cover seven
dimensions out of the nine mentioned in the IOM report, some of these dimensions may account
for bulk of the variance in patient satisfaction. Once the principal components are identied,
performance improvement activities for improving patient satisfaction can focus on these areas.
Andaleeb (2001) obtained evaluations from patients on several dimensions of perceived
service quality including responsiveness, assurance, communication, discipline, and baksheesh.
Using factor analysis and multiple regression, he found signicant associations between the ve
dimensions and patient satisfaction. The current study used the more complete HCAHPS data that
have been collected using standard survey methods and empirically determined the dimensions
that heavily inuence patient satisfaction and thus highlight the low hanging fruits for the
attention of hospitals. A PCA was carried out and ve principal components of patient
satisfaction were identied.
Research question 2.
What is the inuence of hospital ownership (government, churches etc.) on patient
satisfaction?
The research hypothesis that was tested for this question was:
Hospitals owned by churches have higher mean patient satisfaction than the other 7
groups.
The ancient Vedic physician Charaka taught that four guiding principles in healthcare
ought to be : friendship, compassion, joy in serving and equanimity. One of the objectives in this
research was to test this in current practice. Church groups run hospitals as a service and are more
likely to apply Charakas principles in practice than others.
52
The ve satisfaction components together account for 84.3% of the explained variance
while the rst component alone accounts for 59.3% of the variance. Therefore, the rst
component is by itself sufcient to test the research hypothesis. This is reasonable because all the
principal components are correlated because of oblique rotation used in PCA. A similar approach
was taken by Webster (2001).
For testing this hypothesis, an OLS regression was run after creating binary variables for
eight ownership groups. The dependent variable was the rst principal component of patient
satisfaction and is dimensionless, being the standardized component score. The church group was
the base group. The regression equation used, was:
Patient satisfaction = 0 +1 GF+2 GH+3 GL+4 GS+5 PR+6 VO+7 VP+u (1)
where GF, GH, GL, GS, PR, VO, and VP are the binary indicator variables for each of the
ownership groups: government - federal, government - hospital district or authority, government local, government - state, proprietary, voluntary - nonprot - other, and voluntary - nonprot private, respectively.
The estimated intercept term in the regression (0 ) gives the mean dissatisfaction score for
the omitted (baseline) ownership group church owned hospitals. The estimated regression
coefcient for each other group gives the difference in mean dissatisfaction between that group
and the mean for church owned hospitals.
The term u is the error or disturbance term and contains unobserved factors such as
hospital characteristics, costs, patient characteristics and errors in measuring patient satisfaction.
In fact, these omitted explanatory variables in the error term are likely to have a much larger
contribution to R2 , the explained variance in the response variable.
In terms of the regression equation (1), the hypothesis to be tested is H0 : 1 , 7 0
and H1 : 1 , 7 > 0. For conceptual clarity, these hypotheses are stated as one-sided
hypotheses to make expected signs clear. To be conservative, however, statistical tests were based
on two sided p values. The signs of 1 , 7 were expected to be positive because P1 is the rst
53
patient satisfaction component is associated with poor satisfaction. General comparison was
rst be made using the estimated marginal means plots for each satisfaction component obtained
by using PASW (IBM SPSS Inc., 2011).
Assumptions to be satised in testing. The following assumptions in ANOVA and linear
regression (which is basically the same as ANOVA) are required to be met and the test was run
after the checks showed reasonable compliance:
Observations on the dependent variable should be normally distributed with respect to the
ownership groups. The normality assumption is not valid with HCAHPS data as seen in the
paragraph 3 on page 48, because the distribution is skewed to the left with relatively fewer
hospitals scoring low on patient satisfaction. This was expected because patient
dissatisfaction is a serious economic deterrent to hospitals.
The large sample size (N 4000) compensates for lack of normality because of Central
Limit Theorem. Bock(1975), was quoted by Stevens (2002) as stating, even for
distributions which depart markedly from normality, sums of 50 or more observations
approximate to normality. Stevens (2002) further stated that F statistic is robust with
respect to normality assumption, as are the asymptotic t-statistics for regression
coefcients. Wooldridge (2006) has also discussed the large sample properties of estimators
and test statistics. He showed that normality plays no role in the unbiasedness of OLS. He
used the central limit theorem to conclude that the OLS estimators satisfy asymptotic
normality, which means they are approximately normally distributed in large samples.
The homogeneity of variance assumption requires that the population variances of the
ownership groups should be equal. Levenes test or BrownForsythe test using the median,
which is a modied form of Levenes test was used to test homoscedasticity. In case, a
violation of this assumption was found, statistical tests were based on standard errors for
the regression coefcients that are robust to such heteroscedasticity (H. White, 1980;
Wooldridge, 2006).
54
Research question 3.
What is the relationship between hospital ownership and process of care quality
measures?
To analyze this research question, a PCA was rst done on the process of care quality
measures, to identify the main dimensions; otherwise dealing with 24 quality variables would
have been impractical. The data concern 29 clinical conditions covering heart attack, heart failure,
pneumonia, surgical care and childrens asthma care. Three variables concerning childrens health
care are excluded from this study since these are not related to patient satisfaction and outcome
variables. The population has very few children hospitals that provide the data (20 out of 4,530
hospitals). The PCA was done after cleaning the data similar to HCAHPS variables in research
question 1.
Research hypotheses under research question 3.
Research question 3 was expressed in terms of research hypothesis that can be tested:
Hospitals owned by churches have higher mean process of care quality than the other 7 owner
groups.
The ancient Vedic physician Charaka taught that the four guiding principles in healthcare
ought to be friendship, compassion, joy in serving and equanimity. The researcher wanted to
test this in current practice. Church groups started hospitals as a service and hence are more likely
to apply Charakas principles in practice than others.
55
Here, testing needs to be done for all the 5 quality components independently, because
these components are uncorrelated, having been obtained after orthogonal rotation. For testing the
hypotheses, an OLS regression was run after creating binary variables for the eight ownership
groups. The dependent variables was the principal components of process of care quality. The
church group was the base group. The regression equation used was as follows:
Qual1 = 0 + 1 GF + 2 GH + 3 GL + 4 GS + 5 PR + 6 VO + 7 VP + u
(2)
where Qual1 is the rst quality component score and GF, GH, GL, GS, PR, VO, and VP
are the binary indicator variables for each of the ownership groups: government-federal,
government-hospital district or authority, government-local, government-state, proprietary,
voluntary nonprot-other, and voluntary nonprot-private, respectively.
The estimated intercept term in the regression (0 ) gives the mean quality component
score for the omitted (baseline) ownership group church owned hospitals. The estimated
regression coefcient for each other group gives the difference in mean quality component score
between that group and the mean for church owned hospitals.
The term u is the error or disturbance term and contains unobserved factors such as
hospital characteristics, workload on the medical staff and errors in measuring the process-of-care
quality. In fact, these omitted explanatory variables in the error term are likely to have a much
larger contribution to R2 from the regression equation than the ownership explanatory variables
alone.
The regression approach is equivalent to one-way ANOVA with pairwise contrasts of
group means. Equation (2) was not used to predict quality components. The testing was carried
out for each quality component. The regression/ANOVA assumptions already stated under
Research Question 2 were checked for compliance before testing.
56
Research question 4.
What is the relationship between process of care quality measures and patient
satisfaction?
The research hypotheses under this question assume that hospitals are compromising on
quality to gain patient satisfaction. Thus, the hypothesis to be tested for each quality component is
Patient satisfaction components have a negative association with quality components. To test
the research hypotheses, the quality and HCAHPS components found earlier were used. All the
ve HCAHPS components were used because these are correlated; otherwise, the regression
model would become misspecied with omitted variables. Assumptions of linear regression were
evaluated for compliance before testing.
An OLS regression was carried out for each principal component of quality as the
response variable and all the patient satisfaction components as explanatory variables. The
regression equation is:
Qual1 = 0 + 1 P1 + 2 P2 + 3 P3 + 4 P4 + 5 P5 + u
(3)
where, Qual1 is the rst principal component score of quality and P1 , P2 , P3 , P4 , and P5 are the
ve component scores of patient satisfaction.
The term u is the error or disturbance term and contains unobserved factors such as
hospital characteristics, patient characteristics and errors in process-of-care quality and patient
satisfaction. In fact, these omitted explanatory variables in the error term are likely to have a
much larger contribution to R2 than the patient satisfaction components.
The research hypothesis assumes that hospitals are compromising on quality to gain
patient satisfaction. So, the coefcients are expected to be negative. Research hypothesis is:
each 1,2, < 0. Test each 1,2, , H0 : 1,2, 0 and HA : 1,2, < 0. This was repeated for each
principal component of quality.
57
The research hypotheses tested assumed that hospitals may be sacricing quality in favor
of patient satisfaction. These tests were expected to bring out possible conicts and modications
that may be needed in patient satisfaction measurement scales. The impact of public reporting of
patient satisfaction and QIs has been studied to some extent. Pham, Coughlan, and OMalley
(2006) reported that these play complementary roles in encouraging quality improvement but are
poorly coordinated and command sizable resources, in large part because of inadequate
information technology. Policy should be directed at encouraging formal assessments of how
individual and combinations of programs affect quality outcomes, and the development of
adaptable information systems.
Jha et al. (2008) compared the hospitals performance on the HCAHPS survey for to
performance on the indicators of the quality of clinical care and found that hospitals which
provided a higher quality of clinical care had a higher level of patient satisfaction. They used the
fraction of patients who rated the hospital in the highest category (9 or 10 on a scale of 0 to 10) as
the primary indicator of patient satisfaction. They examined bivariate relationship between three
hospital characteristics (nurse/patient ratio, prot or nonprot, teaching or nonteaching) and
HCAHPS rating. Subsequently, they constructed multivariate linear regression models that
adjusted for the other two as well as the other hospital characteristics. The dependent variable was
the proportion of patients who gave a high rating (9 or 10). They categorized all hospitals into
quartiles of HCAHPS rating and examined the mean score for clinical quality within each
quartile, using a test for trend to determine if a higher rating on the HCAHPS survey was
associated with better clinical quality score.
As shown earlier, using high rating as the primary indicator of patient satisfaction is
questionable as the high rating does not gure in the list of principal components.
In contrast to the approach of Jha et al. (2008), the present study carried out a PCA on the
process-of-care quality data as was done with HCAHPS scores to reduce the number of variables
to a manageable number, after testing the statistical assumptions as before.
58
Research question 5.
What is the relationship between outcome of care measures (mortality and readmission
rates) and patient satisfaction, hospital ownership and process of care quality measures?
The outcome variables are separately given for the three conditions: heart attack (HA),
heart failure(HF) and pneumonia(PN). In the absence of a single measure for outcomes, RQ-5
was tested separately for each health condition. This question comprises six different research
hypotheses for the outcome variables:
5.1 The mean 30-day mortality rates are lower for hospitals owned by churches than the other 7
owner groups.
5.2 The mean 30-day readmission rates are lower for hospitals owned by churches than the other
7 owner groups..
5.3 Lower 30-day mortality rates are associated with higher patient satisfaction.
5.4 Lower 30-day readmission rates are associated with higher patient satisfaction.
5.5 Lower 30-day mortality rates are associated with higher quality.
5.6 Lower 30-day readmission rates are associated with higher quality.
Relationship between outcomes and hospital ownership.
The relationship between outcome variables, quality and hospital ownership has been
subjected to some studies but the results are conicting. Some studies that showed signicant
differences are summarized as follows:
1. Shen (2002) examined the effect of ownership choice on patient outcomes after treatment
for acute myocardial infarction and found that for-prot and government hospitals have a
higher incidence of adverse outcomes than not-for-prot hospitals by 34%. In addition, the
incidence of adverse outcomes increases by 79% after a not-for-prot hospital converts to
59
for-prot ownership, but there is little change in patient outcomes in other forms of
ownership conversion.
2. McClellan (2000) studied the risk adjusted 90-day mortality rates for AMI for three years
and found that
For-prot hospitals have signicantly higher mortality rates than not-for prot
hospitals.
Government hospitals have higher mortality rates.
Teaching hospitals have lower mortality rates than not-for prot hospitals.
3. Milcent (2005) studied the effect of hospital ownership on quality in France and found that
the private sector hospitals perform more innovative procedures providing a better quality
of care, measured by the probability of dying. Nevertheless, heterogeneity within hospitals
was greater in for-prot hospitals than in other types of hospital.
4. Yuan et al. (2000) examined the association between hospital type, mortality, and length of
stay using hospitalized Medicare beneciaries for a 10-year period. They studied six
hospital types: for prot (FP), not-for-prot (NFP), osteopathic (OSTEO), public (PUB),
teaching not for prot (TNFP) and teaching public (TPUB). They found that as measured
by the risk-adjusted 30-day mortality, TNFP hospitals had an overall better performance
than other hospital types. However, patients at TNFP hospitals had relatively longer length
of stay than patients at other hospital types, perhaps reecting the medical education and
research activities found at teaching institutions.
On the other hand, some studies have concluded that hospital ownership has little or no
effect on performance. Some are summarized as follows:
1. Sloan et al. (2003) found that 30-day mortality rates in AMI in for-prot hospitals were not
different from those in public and private nonprot hospitals.
60
2. E. F. X. Shortell Stephen M. & Hughes (1988) analyzed the variation among 981 hospitals
in mortality rates for Medicare inpatients in a group of 16 diagnostic categories as a
function of regulatory constraints and the competitiveness of local markets and found no
signicant differences in mortality rates among Investor-Owned hospitals in multihospital
systems, not-for-prot hospitals in multihospital systems, or public hospitals, as compared
with independent nongovernmental hospitals.
3. Baker et al. (2000) reported that the evidence is mixed or inconclusive regarding hospital
ownership and access to care, morbidity, and mortality. The association of hospital
ownership and adverse events is consistently supported.
Eggleston et al. (2008) carried out a meta-study of 31 observational studies that used multivariate
analysis to examine the quality of care and found that ownership does appear to be systematically
related to differences in quality among hospitals in several contexts. The effect of ownership
depended on the institutional context, including differences across regions, markets, and over
time.
All these studies have restricted ownership types to a few categories. The current study
was more complete with eight different ownership groups that are covered in the CMS database.
RQ 5.1 and RQ 5.2 - relationship between outcome variables and ownership type. The
regression equations used for testing outcomes for heart attack were:
hamort1 = 0 + 1 GF + 2 GH + 3 GL + 4 GS + 5 PR + 6 VO + 7 VP + u (4)
hamort2 = 0 + 1 GF + 2 GH + 3 GL + 4 GS + 5 PR + 6 VO + 7 VP + u (5)
where: hamort1 is the 30-day risk adjusted mortality rate for heart attack,
hamort2 is the 30-day risk adjusted readmission rate for heart attack, and GF, GH, GL,
GS, PR, VO, and VP are the binary indicator variables for each of the ownership groups:
government - federal, government - hospital district or authority, government - local, government
61
- state, proprietary, voluntary - nonprot - other, and voluntary - nonprot - private, respectively
other than the church group which is the base group for comparison.
The estimated intercept term in the regression (0 ) gives the mean mortality and
readmission rates for the omitted (baseline) ownership group church owned hospitals. The
estimated regression coefcient for each other group gives the difference in mean mortality and
readmission rates between that group and the mean for church owned hospitals.
The term u is the error or disturbance term and contains unobserved factors such as
patients individual risk characteristics and errors in measuring outcomes. The contribution of the
omitted explanatory variables to R2 from the regression would be far more than the explained
variance from the owner group indicator variables, included in the equations (4) and (5). The
regression was used as a substitute for ANOVA, rather than for prediction.
The research hypothesis assumed that hospitals owned by churches have lower outcomes
than other groups and so the regression coefcients are expected to be positive. In terms of the
regression equations (4) and (5), the hypotheses tested were H0 : 1 , 7 0 and
H1 : 1 , 7 > 0. The coefcients were tested for signicant differences with the base group.
For conceptual clarity, these hypotheses are stated as one-sided hypotheses to make expected
signs clear. To be conservative, however, statistical tests were based on two sided p values.
OLS assumptions that were veried. For testing the difference in group means, the
following OLS assumptions were checked:
The normality assumption is invalid for outcome variables, but the large sample size
( 2, 000) compensates for nonnormality because of Central Limit Theorem.
Independence of observations assumption was valid, because of the nature of the data.
Heteroscedasticity was tested and if necessary corrected using t-tests robust to
heteroscedasticity, as discussed earlier.
These tests were repeated for the heart failure and pneumonia conditions.
62
(6)
hamort2 = 0 + 1 P1 + u
(7)
hfmort1 = 0 + 1 P1 + u
(8)
hfmort2 = 0 + 1 P1 + u
(9)
heart failure
63
pneumonia
pnmort1 = 0 + 1 P1 + u
(10)
pnmort2 = 0 + 1 P1 + u
(11)
where hamort1, hfmort1 and pnmort1 are the 30-day risk adjusted mortality rates for heart attack,
heart failure and pneumonia respectively, hamort2, hfmort2 and pnmort2 are the 30-day risk
adjusted readmission rates for heart attack, heart failure and pneumonia respectively, and P1 is the
rst principal component score of patient satisfaction (poor satisfaction related). The term u is
the error or disturbance term and contains unobserved explanatory variables such as hospital
characteristics, patient risk factors such as age, and errors in measuring patient satisfaction and in
risk adjustment. These are likely to have a major contribution to R2 from the regression. These
equations (6), (7), (8), (9), (10), and (11) was used only to test the hypothesized relationships
between patient satisfaction and outcome variables.
The research hypothesis assumed that greater patient satisfaction is associated with
decreased mortality, and therefor, the regression coefcient was expected to be positive for the
poor satisfaction component. The research hypothesis tested, was: H0 : 1 0 and H1 :
1 > 0. For conceptual clarity, these hypotheses are stated as one-sided hypotheses to make
expected signs clear. To be conservative, however, statistical tests were based on two sided p
values. If the other components of patient satisfaction were tested, these coefcients would have
been tested for negative sign. Normality, linearity and heteroscedasticity assumptions were
validated after each regression.
Relationship between outcomes and quality.
The motivation for the development of quality indicators was to help improve patient
outcomes. The Joint Commission requires reporting of these measures for accreditation. (Werner,
Bradlow, & Asch, 2008). The current study tested the relationship between outcome measures
64
and process-of-care quality data. Jha, Orav, Zhonghe, and Epstein (2007) studied the relationship
between mortality rates and quality and found that higher performance on the AMI, CHF, and
pneumonia indicators was each associated with lower risk-adjusted mortality. They used data
from the 1, December 2005 release of CMS data concerning care provided from 1 April 2004
through 31 March 2005. They limited their analysis to those who were sixty-ve or older. They
chose three subgroups of patients from MedPAR database. The analysis was done at a hospital
level, mapping process-of-care quality indicators from CMS database Their purpose was to gauge
the importance of the process-of-care measures. Werner et al. (2008) compared ten
process-of-care quality measures with mortality rates for 3657 acute care hospitals and concluded
that performance measures predict small differences in mortality rates,
The current study empirically tested the relationship as follows:
5.5 and 5.6 Relationships of Outcome Variables with process-of-care Quality
Components.
The research hypothesis tested, was: Lower mean 30-day mortality and
readmission rates are associated with higher process-of-care quality. The regression equations
were:
heart attack
hamort1 = 0 + 1 Q1 + 2 Q2 + 3 Q3 + 4 Q4 + 5 Q5 + u
(12)
hamort2 = 0 + 1 Q1 + 2 Q2 + 3 Q3 + 4 Q4 + 5 Q5 + u
(13)
hfmort1 = 0 + 1 Q1 + 2 Q2 + 3 Q3 + 4 Q4 + 5 Q5 + u
(14)
hfmort2 = 0 + 1 Q1 + 2 Q2 + 3 Q3 + 4 Q4 + 5 Q5 + u
(15)
heart failure
65
pneumonia
pnmort1 = 0 + 1 Q1 + 2 Q2 + 3 Q3 + 4 Q4 + 5 Q5 + u
(16)
pnmort2 = 0 + 1 Q1 + 2 Q2 + 3 Q3 + 4 Q4 + 5 Q5 + u
(17)
where hamort1, hfmort1 and pnmort1 are the 30-day risk adjusted mortality rates for heart attack,
heart failure and pneumonia respectively, hamort2, hfmort2 and pnmort2 are the 30-day risk
adjusted readmission rates for heart attack, heart failure and pneumonia respectively, and Q1 , Q2 ,
Q3 , Q4 , and Q5 are the component scores of the ve principal components score of care quality.
The term u is the error or disturbance term and contains unobserved factors such as hospital
characteristics, patient risk factors such as age, and errors in measuring process-of-care quality
and in risk adjustment. Many of the process-of-care quality measures are not covered in the data.
The Agency for Healthcare Research and Quality (AHRQ) have established many quality
indicators (AHRQ, 2011) but only a small subset had been adopted by the Hospital Quality
Alliance and publicly reported. The quality indicators that are not covered in HQA, such as
patient safety indicators and prevention quality indicators are likely to be included in the error
term.
Under OLS assumptions, the error term u has zero expected value and is uncorrelated
with the included explanatory variables (assumption of no omitted variables bias) (Wooldridge,
2006, p. 29). The research hypothesis is that greater quality is associated with decreased
mortality. H0 : 1 0 and H1 : 1 < 0. For conceptual clarity, these hypotheses are stated as
one-sided hypotheses to make the expected signs clear. To be conservative, however, statistical
tests were based on two sided p values.
It was found that OLS regression assumptions were adequately met and no transformation
was needed. Only in the cases of readmission rate for heart attack and mortality rate for heart
failure, robust standard errors were necessary because of heteroscedasticity.
66
Summary
In this chapter, the methodology used in this study was described. The statistical
technique Principal Component Analysis was used for patient satisfaction and process of care
quality data to identify the main dimensions of patient satisfaction and process-of-care quality.
The ve research questions were analyzed by statistically testing the research hypotheses
formulated for each question, using regression analysis. In the next chapter, the results of the
analysis carried out using the methodology are given.
67
Chapter 4
Chapter Overview
The ve research questions formulated in Chapter 3 are analyzed by statistically testing
the research hypotheses formulated for each question in this and the next chapter. The empirical
knowledge gained as a result of the analysis will be useful to hospitals in their efforts to improve
performance. The purpose of the study was achieved by examining and analyzing the archival
data in the CMS database concerning HCAHPS patient satisfaction survey ndings, hospital
quality performance indicators reported by hospitals and the patient outcome data on mortality
and readmission rates of Medicare patients. The archival data for analysis was downloaded from
publicly available CMS databases. This chapter presents the results of the data analysis for the
rst four of the ve research questions that were raised in Chapter 3.
Research Question 1
Which dimensions of the patient perceptions of healthcare have the greatest impact on
patient satisfaction?
The main dimensions of patient satisfaction were identied by doing a principal
component analysis (PCA) of the HCAHPS data that was downloaded from CMS. PCA provides
the means to undertake such a structural analysis. Out of 4,530 cases, hospitals with no HCAHPS
data or fewer than 100 responses were dropped in line with the HCAHPS recommendation that
the results are unreliable in such cases. An outlier analysis was done using Mahalanobis distance,
68
69
maternity, medical and surgical line categories are identied as eligible for the survey. Surveying
must be done between 48 hours and six weeks after discharge.
From those eligible for the survey, some patients are removed as per exclusions:
Patients who request that they are not to be contacted
Court/Law enforcement patients (i.e. prisoners)
Patients discharged to hospice care
Patients with a foreign home address
Patients who are excluded because of state regulations
Hospitals are required to retain documentation that veries all exclusions. A
de-duplication process is also applied to make sure that only one adult member per household is
included in the sample for a given month. The nal sample drawn each month must reect a
random sample of eligible patients after applying exclusions and de-duplication. Three options
are available for sampling: Simple random sampling, Proportionate stratied random sampling
and Disproportionate random sampling. The standard method is simple random sample.
Assuming a 40% response rate, a 17% ineligibility rate and a target to get at least 300 completed
surveys, CMS estimates 909 patient discharges in a 12-month reporting period. Smaller hospitals
having a lesser number of discharges are allowed to sample all eligible discharges to meet the
target of 300 completed responses. For ease of sampling, CMS recommends that hospitals sample
an equal number of discharges each month (CMS, 2010). The published HCAHPS data
incorporate adjustments for survey mode and patient mix to minimize bias.
Survey questions. The HCAHPS survey uses 10 forced-response questions. To seven
questions, patients respond selecting one of three options (sometimes or never, usually and
always). To the question on Were you given information about what to do during recovery at
home?, patients can select between yes and no. To the question on overall rating to the
70
hospitals, patients can select a rating out of 3 options: 6 or lower (low), 7 or 8 (medium), 9 or 10
(high). To the question Would you recommend the hospital to friends and family?, the patients
should select one out of three options: no, denitely yes, probably yes. Thus, there are 29
variables. Including all 29 variables in the PCA would cause multicollinearity because the total
scores would add up to 100 for each question. But dropping one variable in every question to
avoid multicollinearity would exclude this variable from the factor scores. It was preferred to
keep all the 29 variables in the PCA so that the resulting factor scores would account for all
variables. The problem in this approach would be the resulting singular correlations matrix.
However, statistical software packages use the technique of generalized matrix inverse to tackle
this (Basilevsky, 1981).
Summary statistics of HCAHPS variables. Summary statistics for HCAHPS variables
are given in Table 1.
The last column in this table, excess kurtosis is obtained by subtracting from kurtosis, 3
which is the value of kurtosis for standard normal distribution. From this table, the following are
the key ndings:
The skewness statistic is observed to be less than 1 in absolute value for all but four
variables.
The excess kurtosis statistic is greater than 2 for three variables: Nurses sometimes or never
communicated well, Pain was sometimes or never controlled, and Patients would not
recommend the hospital. For these three variables, the skewness statistic is also greater than
1.
The skewness and kurtosis values show moderate deviation from normal distribution. The
signicance of the deviation needs to be tested statistically. However, because of the large
sample size (> 3500), normality tests are also signicant.
Label
Nurses sometimes or never communicated well
Nurses usually communicated well
Nurses always communicated well
Doctors sometimes or never communicated well
Doctors usually communicated well
Doctors always communicated well
Patients sometimes or never received help as soon as they wanted
Patients usually received help as soon as they wanted
Patients always received help as soon as they wanted
Pain was sometimes or never well controlled
Pain was usually well controlled
Pain was always well controlled
Staff sometimes or never explained about medicines to patients
Staff usually explained about medicines to patients
Staff always explained about medicines to patients
Rooms/bath rooms were sometimes or never clean
Rooms/bath rooms were usually clean
Rooms/bath rooms were always clean
Area around patients room were sometimes or never kept quiet
Area around patients room were usually kept quiet
Area around patients room were always kept quiet
Patients were given info. on what to do during recovery at home
Patients were not given info. on what to do during recovery at home
Patients gave a rating of 6 or lower (low)
Patients who gave a rating of 7 or 8 (medium)
Patients who gave a rating of 9 or 10 (high)
No, patients would not recommend the hospital
Yes, patients would probably recommend the hospital
Yes, patients would denitely recommend the hospital
N
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
3678
Min.
0
3
48
0
0
60
0
4
31
0
2
46
0
2
34
0
4
39
0
4
27
54
4
0
2
30
0
2
31
Max.
22
34
97
17
29
100
40
40
96
25
40
95
44
35
98
30
36
94
36
56
96
96
46
34
43
98
27
53
98
SD
3.26
4.3
6.71
2.43
3.83
5.7
5.9
5.08
9.45
3.27
4
5.84
5.85
3.02
6.67
4.59
4.72
8.32
6.1
6.14
10.74
5.03
5.03
4.83
5.79
9.6
3.55
7.71
10.22
RSE - Relative Standard Error is the standard error of the mean expressed as a fraction of the mean and is displayed as a percentage.
Excess kurtosis = kurtosis - 3 indicating how much larger than the coefcient associated with a normal distribution, which is around 3.
Variable
h comp 1 s p
h comp 1 u p
h comp 1 a p
h comp 2 s p
h comp 2 u p
h comp 2 a p
h comp 3 sn p
h comp 3 u p
h comp 3 a p
h comp 4 sn p
h comp 4 u p
h comp 4 a p
h comp 5 sn p
h comp 5 u p
h comp 5 a p
h clean hsp sn p
h clean hsp u p
h clean hsp a p
h quiet hsp sn p
h quiet hsp u p
h quiet hsp a p
h comp 6 y p
h comp 6 n p
h hsp rating 0 6
h hsp rating 6 8
h hsp rating 9 10
h recmnd dn
h recmnd py
h recmnd dy
Table 1
X
5.8
19.9
74.3
4.76
15.59
79.66
11.59
26
62.41
7.56
24.12
68.32
22.51
18.62
58.87
10.04
20.48
69.48
12.73
31.11
56.16
80.57
19.43
9.87
25.19
64.94
5.73
26.07
68.21
SEM
0.05
0.07
0.11
0.04
0.06
0.09
0.1
0.08
0.16
0.05
0.07
0.1
0.1
0.05
0.11
0.08
0.08
0.14
0.1
0.1
0.18
0.08
0.08
0.08
0.1
0.16
0.06
0.13
0.17
RSE1
0.86%
0.35%
0.15%
0.84%
0.38%
0.11%
0.86%
0.31%
0.26%
0.66%
0.29%
0.15%
0.44%
0.27%
0.19%
0.80%
0.39%
0.20%
0.79%
0.32%
0.32%
0.10%
0.41%
0.81%
0.40%
0.25%
1.05%
0.50%
0.25%
skewness
1.37
-0.19
-0.33
0.88
-0.23
-0.11
0.88
-0.83
0.22
1.07
-0.41
0.06
0.14
-0.14
0.39
0.54
-0.48
0.18
0.45
-0.85
0.48
-0.38
0.38
1.02
-0.66
0.09
1.42
-0.11
-0.13
Excess kurtosis 2
2.64
0.34
0.64
1.67
0.15
0.36
1.39
1.22
0.18
2.37
1
0.9
0.54
2.2
1.2
0.3
0.14
-0.22
-0.09
1.24
0.23
0.59
0.59
1.61
1.21
0.6
2.83
0.11
0.13
71
72
74 % feel that nurses always communicated well (relative standard error 0.15%). Relative
standard error (RSE) is a measure of the means reliability and is obtained by dividing the
standard error of the mean (SEM) by the mean; then multiplied by 100 to be expressed as a
percentage.
Compared to the nurses performance, about 80% feel that doctors always communicated
well (RSE = 0.11%)
Only 62% (RSE = 0.26%) feel that they always received help as soon as they wanted. This
low value seems to bring into focus Charakas premise that compassion should be the
healthcare personnels prime quality (Valiathan, 2007). This does not seem to be the case in
US hospitals.
Only about two-thirds of the patients (mean 68 with RSE of 0.15%) feel that pain was
always well controlled. This again brings into focus Charakas insistence of four qualities
that should be the guiding principles of healthcare personnel: friendship, compassion, joy
in serving, equanimity.
Only about 59% (RSE = 0.19%) feel that staff always explained about medicines before
giving to patients. This low value could result in patient safety incidents if patients take
medications higher or lower doses than prescribed or adequate precautions are not taken.
About 65% (with an RSE of 0.2%) gave the top over all rating of 9 or 10 and 68% (RSE =
0.17%) would denitively recommend their hospital. About one third of the patients gave
medium or low over-all rating and may or may not recommend their hospital to friends and
relatives, showing the widespread feeling of dissatisfaction among the patients in US.
Validation of PCA assumptions.
The assumptions of PCA are discussed and multivariate analysis are discussed by e.g.,
Stevens (2002, p. 237). Garson (2008) has also discussed the key assumptions. In particular, Pett
73
et al. (2003) have devoted an entire book on factor analysis. Before applying factor analysis, it is
necessary to test if the assumptions of PCA are valid for HCAHPS data.
For valid PCA, all relevant variables should be included. Otherwise, correlations could be
distorted. The HCAHPS questionnaire has been developed after careful evaluation and
testing and has been accepted by healthcare professionals as a valid indicator of patient
satisfaction. The data has already been subjected to several factor analyses and so it is
reasonable to conclude that all relevant variables have been included.
Outliers can impact correlations heavily and so, before factor analysis, HCAHPS data was
tested for presence of inuential outliers. For the multivariate analysis, the study used
Mahalanobis distance to identify cases which are multivariate outliers, as discussed in
(Garson, 2008). The following procedure was used:
1. An OLS regression was done with all 29 HCAHPS variables as dependent variables
and ownership type code as independent variable. The term OLS stands for Ordinary
Least Squares. OLS assumptions are discussed in (Wooldridge, 2006, p. 29).
Actually any numerical variable with non-missing values can be used as independent
variable as the regression results are not important. This regression is only for nding
Mahalanobis distance option in regression to identify outliers. The Mahalanobis
distance option was used and saved.
2. A new variable was created to get the probability of 2 distribution function.
3. All cases with p-value < 0.001 were identied as outliers. 108 such outliers were
identied.
4. A binary 0-1 variable was created with value of 1 for outliers and value of 0 for all
other cases. A dummy Boolean variable (0-1) was dened and set to 1 for cases with
high Mahalanobis distance. A regression was run using this dummy variable as the
response variable and the other variables as explanatory variables. Garson (2008)
discussed this procedure. The R2 is about 0.175. Since R2 is low, it was reasonable to
74
conclude that these outliers are random errors that do not affect the outcome of the
factor analysis. Only 88 out of more than 4,000 cases were outliers and these cases
were omitted from PCA.
Checking validity of PCA assumptions.
HCAHPS data after removing multivariate outliers and missing data, meets with the basic
requirements of PCA. Other possible violations are explored below:
1. The observations need to be independent. Stevens (1999, p. 77) considered that any
violation may result in a serious bias in the testing. For HCAHPS data, independence
assumption is valid because possible interactions between patients are minimal. The survey
is independently administered to patients between 48 hours and six weeks after they are
released from the hospital. Therefore, it is reasonable to assume that subjects are
responding independently.
2. For hypothesis testing after PCA, the observations on the dependent variables (patient
satisfaction here) need to follow a multivariate normal distribution. Univariate normal
distribution for each variable is a necessary but not sufcient condition for multivariate
normality. Univariate normality of HCAHPS data is moderately valid as seen from Table 1.
The data are skewed mostly to the left signifying relatively few low values. The null
hypothesis of normality for all univariates was tested with Shapiro-Wilk normality test
(Park, 2008). This ruled out normality assumption for all univariates. A saving grace for
HCAHPS variables is the size of the sample for each hospital. The central limit theorem
states that the sum of independent observations having any distribution whatsoever
approaches a normal distribution as the number of observations increases (Stevens, 2002,
p. 243). Sums of 50 or more observations are considered to approximate to normality. For
HCAHPS data, each hospital was required to have at least 100 respondents for inclusion in
the database. The large sample size of 3,678 hospitals shows robustness to deviation from
75
normality. Therefore, it is reasonably assumed that the deviation of HCAHPS data from
multivariate normality may not pose problems.
3. PCA requires high correlations among the variables. The HCAHPS variables are highly
correlated as seen from the correlations matrix in tables 92, 93, and 94 in Appendix - A.
Correlations among the variables are signicant either at 0.01 or at 0.05 level. Also, the
matrix is neither null nor close to an identity matrix. The data are suitable for PCA, even
though Bartletts sphericity test could not be used because of multicollinearity in the data.
PCA results.
Excel based statistical software from Addinsoft, inc. (XLSTAT) 2011 software was used
to carry out PCA on HCAHPS data (Addinsoft, 2011) and (Amza, Paris, Tarcolea, Carp, &
Nimigean, 2009). Five principal components were identied using Kaisers criterion of taking
only the eigenvalues above 1. This is justied because Stevens (2002, p. 367) showed that this
criterion is accurate when the number of variables > 30 and when sample size > 250. Stevens
(2002, p. 367) has also discussed a graphical method known as scree test. In this method, the
magnitude of the eigenvalues (vertical axis) are plotted against their ordinal numbers (rst
eigenvalue, second eigenvalue, and so on). The magnitude of successive eigenvalues drops off
sharply. The expert recommendation is retaining all the eigenvalues (and hence components) in
the sharp descent before leveling off. Accordingly, a scree plot was produced and is given in
Figure 1. The scree plot also shows that after ve components, the inection of the curve reduces.
76
F2
F3
F4
F5
Eigenvalue
17.197
Variability (%) 59.300
Cumulative (%) 59.300
3.288
11.337
70.637
1.698
5.856
76.493
1.255
4.327
80.820
1.008
3.476
84.295
As recommended by Field (2005, p. 625), an oblique rotation was selected rst because
the underlying components are likely to be correlated because all are measures of an aspect of the
construct patient satisfaction. Promax rotation was selected following Stevens (2002, p. 370).
After oblique rotation, the correlation among the component scores were tested for signicant
correlation by Spearman test and found to be signicantly correlated. The results are given in
77
Table 3. This table lists the p-values from the test. The component loadings of the variables after
promax rotation are given in Table 4.
Table 3
HCAHPS: Spearmann Test - p Values
Variables[1]
HCAHPS D1
HCAHPS D2
HCAHPS D3
HCAHPS D4
HCAHPS D5
HCAHPS
HCAHPS
HCAHPS
HCAHPS
HCAHPS
0
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0
D1
D2
D3
D4
D5
Correlation (or contribution) biplots are useful to interpret the components as explained in
(Greenacre, 2010) because they give an immediate idea of which variables are most responsible
for the given display. In the correlation biplots given in Appendix - B, correlations of the
HCAHPS variables with the component axes are given by arrows (with ball head).
The rst chart shows that several variables are correlated with the component axes 1 and
2. Many variables associated with responses usually are seen to be close to the second
component (Y-axis). Thus this component is interpreted as Required level of performance.
Many variables associated with responses sometimes or never are seen to fall near the rst
component (X-axis). Hence this component is interpreted as Poor satisfaction.
From the second chart, the 3rd component is interpreted as overall because variables
associated with Overall and recommendation map on this axis. From the third chart, the
fourth component is interpreted as Cleanliness because responses for the question How clean
do you nd the hospital? map on this axis.
From the fourth chart, the fth component is interpreted as Post hospitalization care
because responses associated with Were you given information on what to do during their
recovery at home? map on this axis.
Text
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
comp 1 sn p
comp 1 u p
comp 1 a p
comp 2 sn p
comp 2 u p
comp 2 a p
comp 3 sn p
comp 3 u p
comp 3 a p
comp 4 sn p
comp 4 u p
comp 4 a p
comp 5 sn p
comp 5 u p
comp 5 a p
clean hsp sn p
clean hsp u p
clean hsp a p
quiet hsp sn p
quiet hsp u p
quiet hsp a p
comp 6 y p
comp 6 n p
hsp rating 0 6
hsp rating 7 8
hsp rating 9 10
recmnd dn
recmnd py
recmnd dy
Variable
Table 4
0.949
0.357
-0.690
0.966
0.466
-0.724
0.823
-0.017
-0.505
0.916
0.136
-0.606
0.668
-0.236
-0.479
0.455
-0.119
-0.184
0.184
-0.488
0.174
-0.080
0.080
0.698
-0.095
-0.294
0.862
-0.019
-0.285
D1
-0.230
0.626
-0.289
0.150
0.722
-0.548
-0.138
0.686
-0.283
-0.207
0.818
-0.444
0.039
0.911
-0.447
-0.364
0.271
0.047
0.306
0.870
-0.671
-0.016
0.016
-0.080
0.158
-0.055
-0.234
0.071
0.027
D2
0.049
0.024
-0.039
-0.076
-0.131
0.120
-0.090
-0.064
0.091
0.070
0.063
-0.082
-0.054
-0.016
0.055
-0.026
-0.036
0.035
0.084
0.204
-0.164
0.000
0.000
0.365
0.826
-0.682
0.315
1.005
-0.868
D3
0.101
0.069
-0.093
-0.166
-0.044
0.100
0.280
0.312
-0.343
0.021
-0.167
0.103
0.217
-0.320
-0.045
0.759
0.829
-0.890
0.417
0.203
-0.353
0.025
-0.025
0.012
0.145
-0.094
-0.052
-0.122
0.110
D4
0.017
-0.029
0.011
-0.054
-0.161
0.131
0.060
0.036
-0.057
0.039
0.014
-0.032
0.142
0.117
-0.178
-0.050
0.006
0.024
-0.008
-0.022
0.017
-0.950
0.950
0.029
-0.009
-0.009
-0.014
-0.001
0.006
D5
78
79
The bipolar nature of the variables is brought out in the correlation biplots. For example,
the binary response Yes/No to question 6 Were you given information on what to do during
their recovery at home? plot on opposite side of the origin on the vertical axis. Many of the
component loadings are bipolar because of the nature of the variables and these are relatively
more difcult to interpret. Correlation biplots make the interpretation simpler.
Interpretation of principal components of HCAHPS scores.
From Table 4 and the correlation biplots, the following interpretation is offered for the
principal components:
Component 1 Poor satisfaction accounts for 59.3% of variance and is associated with
responses relating to dissatisfaction because the main loadings come from responses
associated with sometimes or never. It is seen that responses associated with always
also load to a lesser extent. Their loadings are negatively signed. We are dealing with
polarity because these variables are collinear.
Component 2 Required level of performance accounts for 11.3% of the variance and is
associated with responses relating to an expected level of performance (usually). This is
the minimum level of performance expected by the patients.
Component 3 Overall accounts for 6% of the variance and is associated with responses to
over-all rating and recommendation
Component 4 Cleanliness accounts for 4.3% of the variance and is associated with
response to cleanliness
Component 5 Post hospitalization care accounts for 3.5% of the variance and is
associated with response to how well instructions were given to patients at discharge. Here
also, the polarity of the variable comes through.
80
The contributions of the variables (%) to the rst ve principal components accounting for 84%
of the standardized variance in patient satisfaction (as measured by HCAHPS survey) are
summarized in Table 5.
Table 5
Contribution of the HCAHPS Variables (%) after Promax Rotation
Variable
Description
D1
D2
D3
D4
D5
h comp 1 sn p
11.190
0.960
0.074
0.338
0.015
h comp 1 u p
1.582
7.138
0.017
0.158
0.045
h comp 1 a p
5.918
1.526
0.047
0.289
0.006
h comp 2 sn p
11.609
0.409
0.173
0.912
0.150
h comp 2 u p
2.699
9.497
0.520
0.063
1.342
h comp 2 a p
6.517
5.473
0.437
0.331
0.887
h comp 3 sn p
8.432
0.347
0.243
2.617
0.184
h comp 3 u p
0.003
8.574
0.126
3.238
0.067
h comp 3 a p
3.175
1.455
0.249
3.911
0.166
h comp 4 sn p
10.426
0.783
0.146
0.014
0.080
h comp 4 u p
0.230
12.206
0.120
0.932
0.011
h comp 4 a p
4.569
3.589
0.204
0.353
0.053
h comp 5 sn p
5.545
0.028
0.088
1.571
1.048
to patients
h comp 5 u p
0.690
15.122
0.008
3.409
0.703
h comp 5 a p
2.847
3.642
0.090
0.069
1.631
h clean hsp sn p
2.575
2.414
0.020
19.181
0.128
h clean hsp u p
0.176
1.339
0.040
22.886
0.002
h clean hsp a p
0.420
0.040
0.037
26.376
0.030
h quiet hsp sn p
0.422
1.704
0.212
5.795
0.004
night
h quiet hsp u p
2.959
13.804
1.263
1.375
0.025
h quiet hsp a p
0.377
8.207
0.817
4.152
0.016
h comp 6 y p
0.080
0.005
0.000
0.021
46.673
0.080
0.005
0.000
0.021
46.673
home
h comp 6 n p
h hsp rating 0 6
6.051
0.118
4.034
0.005
0.042
h hsp rating 7 8
0.111
0.453
20.637
0.699
0.004
h hsp rating 9 10
1.073
0.054
14.050
0.292
0.004
h recmnd dn
9.235
1.002
3.010
0.090
0.011
h recmnd py
0.004
0.093
30.576
0.497
0.000
81
Variable
Description
D1
D2
D3
D4
D5
h recmnd dy
1.007
0.014
22.762
0.404
0.002
82
Research Question 2
What is the relationship between hospital ownership and patient satisfaction?
Hospital ownership groups for HCAHPS data are listed in Table 6.
Table 6
HCAHPS Data by Hospital Ownership Groups
Categories
Government - Federal
Count
f1
SD
48
1.305
-0.14
1.04
357
9.706
-0.15
1.02
Government - Local
260
7.069
-0.3
0.91
Government - State
56
1.523
0.37
1.09
Proprietary
646
17.564
0.44
1.28
458
12.452
0.01
0.82
628
17.074
-0.12
0.83
1225
33.306
-0.08
0.9
Total
3678
100
83
P1 is the rst patient satisfaction component is associated with poor satisfaction. General
comparison will rst be made using the estimated marginal means plots for each satisfaction
component obtained by using PASW software (PASW, 2007). For the other components, the
84
85
Dependent variable observations in each group need to follow normal distribution. Testing
normality with PASW results in Table 7 which shows that except in the state government
group, Shapiro-Wilk statistic is signicant at 95% condence level.
As a visual check, Q-Q plots for the rst principal component of patient satisfaction were
created and these are shown in Appendix - D. The Q-Q plots also showed some deviation
from normality. Marked deviation from normality are seen in three groups: govt.- Hospital
District or Authority, govt.-local and voluntary not-for-prot-private. Other groups do not
show gross deviation from normality.
The group size ranged from 48 to 1,225, and using central limit theorem, deviation from
normality is not likely to increase Type I error (Stevens, 2002, p. 243). Lumley, Diehr,
Emerson, and Chen (2002) have also shown, after a simulation study with extreme
non-normal data, that nonnormality or even heteroscedasticity do not affect the validity of
linear regression in large data sets used in public health research. Accordingly,
non-normality was not considered as a serious problem for doing an OLS regression.
Nevertheless, normality of the residuals and multicollinearity were tested after regression,
before hypothesis testing.
Table 7
Normality Testing of HCAHPS Component 1 by Ownership Groups
Group
Government - Federal
Government - Hospital District or Authority
Government - Local
Government - State
Proprietary
Voluntary non-prot - Church
Voluntary non-prot - Other
Voluntary non-prot - Private
Shapiro-Wilk
Statistic
0.947
0.938
0.95
0.968
0.986
0.969
0.97
0.933
df
48
357
260
56
646
458
628
1225
signicance
0.029
0.000
0.000
0.143
0.000
0.000
0.000
0.000
Homogeneity of variance across groups is required for OLS regression. Normality robust
Levenes test was run using Stata 11 (Baum, 2006). The results show evidence of
86
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
Owner Group
Government - federal
Government - hosp. area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
Voluntary non-pr-other
Voluntary non-pr-private
Constant term
48
357
260
56
646
458
628
1225
-0.15
-0.16
-0.31
0.36
0.43
-0.13
-0.09
0.01
0.15
0.07
0.07
0.15
0.06
0.05
0.05
0.04
-0.97
-2.44
-4.52
2.41
6.78
-2.61
-1.99
0.31
0.33
0.02
0.00
0.02
0.00
0.01
0.05
0.76
-0.45
-0.29
-0.44
0.07
0.30
-0.23
-0.18
-0.06
0.15
-0.03
-0.17
0.65
0.55
-0.03
0.001
0.09
0.18
0.18
0.36
0.42
0.39
0.16
0.10
0.57
0.56
0.64
0.67
0.66
0.56
0.54
freedom is 3,670. The R2 for the estimated equation is 0.0493 showing that the regression only
explains 5% of the response variable, patient satisfaction. The adjusted R2 is obtained by using
the mean sum of squares rather than total sum of squares to get unbiased estimates for the
population variance and is only slightly lower, 0.0475. The RMSE, the estimated standard
deviation of the residuals was 0.976 units in estimating patient satisfaction. This shows that the
model t is not high. Predicted R2 is used to show how well the model predicts responses for new
observations. Its value of 0.045 is close to R2 = 0.0493 showing cross-validation. This lack of
model t was expected because only ownership variable is included in the prediction equation for
group mean comparison and not other potentially important explanatory variables such as hospital
and patient characteristics and nature of ailment.
87
The overall F-statistic is signicant suggesting that patient satisfaction is related to the
hospital ownership. On the basis of two-sided tests of signicance, individual coefcient
estimates show that church group hospitals have signicant differences in patient satisfaction
from hospitals owned by district or area, state government, local government, proprietary,
voluntary nonprot-other, and voluntary nonprot-private groups. Only federal government
hospitals do not differ signicantly. Nevertheless, the signs were positive as expected only for
local government and proprietary groups while the other groups had negative sign showing that
they have lesser dissatisfaction. For proprietary hospital, is positive as expected, high (0.43) and
signicant (t = 6.78, P > |t| = 0) and this means that on the average, the mean patient
dissatisfaction is 0.43 units higher in proprietary hospitals than in church group hospitals, while
the population mean dissatisfaction level in any hospital is 0 units, considering that these are
standardized scores. The 95% CI is [0.30, 0.55]. The constant term is the mean satisfaction
component 1 score for the omitted base owner group (church).
Effect sizes. Effect sizes for comparison of independent group means were computed
from the table given by (Cohen, 1992). Cohens d is the uncorrected standardized mean difference
between two groups based on the pooled standard deviation (Ellis, 2010) and is computed for
each individual group by dividing the difference between the group mean and the church group
mean by the pooled standard deviation calculated (Cohen, 1988, p. 67) by the equation (18).
sdpooled
2
2
XA XA + XB XB
(18)
=
(NA + NB 2)
where the subscript B refers to the base group and A refers to the group being tested. Cohen
(1992) gives the size index for small effect as 0.20, for medium as 0.50 and for large as 0.80. The
effects are seen to be small except comparisons with proprietary, state and local government
hospitals with indexes of 0.39, 0.42 and 0.36,considerably larger than the low effect index of 0.2.
as expected
88
89
90
Table 9
RQ2 : Robust Regression Results
Regression Results
SE
P > |t|
[95%
CI]
Owner Group
Government - federal
Government - hosp. area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
Voluntary non-pr-other
Voluntary non-pr-private
Constant term
-0.23
-0.24
-0.36
0.31
0.356
-0.13
-0.14
-.039
0.13
0.062
0.069
0.13
0.054
0.054
0.048
.041
-1.71
-3.92
-5.17
2.46
6.62
-2.47
-2.87
-0.94
0.087
0.000
0.000
0.014
0.000
0.014
0.004
0.346
-0.49
-0.37
-0.49
0.062
0.252
-0.241
-0.234
-0.12
0.03
-0.12
-0.22
0.55
0.46
-0.027
-0.044
0.042
which those records with Cooks distance greater than 1 are given zero weight. From this model,
weights are assigned to records according to the absolute difference between the predicted and
actual values (the absolute residual). The records with small absolute residuals are weighted more
heavily than the records with large absolute residuals. Then, another regression is run using these
newly assigned weights, and then new weights are generated from this regression. This process of
regressing and re-weighting is iterated until the differences in weights before and after a
regression is sufciently close to zero (Bruin, 2011).
A robust regression with Stata 11 (using the command rreg) on the Equation (1)
identied 13 potentially inuential outliers (with absolute value of standardized residuals 3.8)
and weighted them by 0, thus effectively leaving these cases out. These cases had patient
satisfaction component 1 higher than 3.7 units, hospitals with which patients have expressed very
high dissatisfaction. F (7,3670) increased to 30.88 from 19.17, and values were changed
slightly, but again, church group hospitals had signicant differences in patient satisfaction from
hospitals owned by district or area, state government, local government, proprietary, voluntary
nonprot-other, and voluntary nonprot-private groups. The signs of the coefcients were the
same as before. This shows that the empirical results are not seriously affected by the presence of
outliers. In summary, robust regression conrmed that OLS results were not affected by the
presence of inuential outliers and the OLS test results are reasonable. The robust regression
results are given in Table 9.
91
Research Question 3
What is the relationship between hospital ownership and process of care quality
measures?
To analyze this research question, a PCA was rst done on the process of care quality
measures, to identify the main dimensions; otherwise dealing with 24 quality variables will be
very time-consuming. The data concern 29 clinical conditions covering heart attack, heart failure,
pneumonia, surgical care and childrens asthma care. Three variables concerning childrens health
care are excluded from this study since these are not related to patient satisfaction and outcome
variables. The population has very few childrens hospitals that provide the data (20 out of 4,530
hospitals).
Description
Surgery Patients whose antibiotics are stopped within 24 hours after surgery
Variable
AMI1
AMI2
AMI3
AMI5
AMI4
AMI8A
HF2
HF3
HF1
HF4
PN2
PN5C
PN3B
PN4
PN6
PN7
SCIPINF1
SCIPINF3
SCIPINF2
SCIPVTE2
SCIPVTE1
SCIPINF4
SCIPINF6
SCIPCARD2
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
615
Observations
Table 10
Minimum
38
82
60
65
63
88
61
72
49
68
75
68
70
57
84
45
76
88
18
92
91
75
90
92
Maximum
100
100
100
100
100
100
99
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
Mean
89.621
98.556
91.309
93.272
90.457
97.393
91.272
95.042
89.459
91.007
97.628
94.143
94.133
90.974
99.044
85.437
93.989
98.28
84.589
99.447
98.53
94.961
98.488
98.576
Std. deviation
9.868
2.455
6.452
5.562
6.513
2.004
5.557
3.517
7.96
5.107
3.664
4.357
4.051
7.169
2.015
10.979
5.149
2.082
12.141
1.146
1.535
4.873
1.626
1.39
92
93
94
of less than 1 in accordance with Kaiser criterion. These ve components together explain about
Table 11
Identifying Principal Components of Process of Care Quality Data
Eigenvalue
Variability (%)
Cumulative %
F1
F2
F3
F4
F5
6.56
27.335
27.335
2.222
9.258
36.593
1.752
7.299
43.891
1.497
6.238
50.129
1.09
4.542
54.671
55% of the total variance. Identication of ve principal components is conrmed by the scree
plot which shows that after the fth component, the curve tapers off.
Rotating component axes of process of care quality data. Spearman rank correlation
test showed that the quality principal components are not signicantly correlated. Hence, an
orthogonal rotation will be needed. After doing an orthogonal rotation, varimax rotation was
95
chosen to make the components become more easily interpretable. After varimax rotation, the
percentage of the explained variance by the components is given in Table 12. After rotation, the
Table 12
Explained Variance of Quality Components after Varimax Rotation
Variability (%)
Cumulative %
D1
D2
D3
D4
D5
15.797
15.797
10.655
26.453
11.166
37.619
8.119
45.738
8.932
54.671
From the component loading Table 13 and the correlation biplots, the following
interpretation is offered for the quality principal components:
Quality component 1 heart attack, failure related accounts for 15.8 % of the variance.
Quality component 2 pneumonia related accounts for 10.4 % of the variance.
D1
0.689
0.761
0.752
0.814
0.154
0.474
0.561
0.705
0.378
0.086
0.130
0.172
0.121
0.036
0.125
0.112
0.166
0.238
0.095
0.201
0.164
0.199
0.061
0.070
Variable
AMI1
AMI2
AMI3
AMI5
AMI4
AMI8A
HF2
HF3
HF1
HF4
PN2
PN5C
PN3B
PN4
PN6
PN7
SCIPINF1
SCIPINF3
SCIPINF2
SCIPVTE2
SCIPVTE1
SCIPINF4
SCIPINF6
SCIPCARD2
0.018
0.059
0.161
0.112
0.065
0.351
0.151
0.162
0.122
0.123
0.618
0.730
0.727
0.132
0.581
0.601
0.193
0.156
0.165
0.001
0.011
0.216
0.110
-0.028
D2
0.132
0.171
0.082
0.137
0.016
0.206
0.216
0.084
0.005
0.023
0.006
0.113
0.021
0.039
0.273
-0.078
0.337
0.578
0.530
0.881
0.891
0.304
0.087
0.137
D3
-0.003
0.051
0.105
0.097
0.785
-0.016
0.125
0.088
0.085
0.786
0.112
0.108
0.087
0.732
0.073
0.170
0.092
0.027
-0.038
0.078
0.091
-0.114
0.100
0.107
D4
0.178
0.072
0.131
0.002
0.011
-0.050
0.148
0.273
0.535
0.192
0.479
0.002
0.138
0.201
0.058
0.486
0.575
0.304
0.314
0.061
0.029
0.020
0.394
0.676
D5
Patients given aspirin at arrival
Patients given aspirin at discharge
Patients given ACE inhibitor or ARB
Patients given beta blocker at discharge
Patients given smoking cessation advice
Patients given PCI within 90 minutes of arrival
Patients given evaluation of LVS function
Patients given ACE inhibitor or ARB for LVSD
Patients given discharge instructions
Patients given smoking cessation advice
Patients given pneumococcal vaccine
Patients given antibiotics within 6 hours of arrival
Patients given ER blood culture before antibiotics
Patients given smoking cessation advice
Patients given most appropriate initial antibiotics
Patients given inuenza vaccine
Surgery Patients given antibiotics one hour before incision
Surgery Patients whose antibiotics are stopped within 24 hours after surgery
Surgery Patients given antibiotics for their surgery
Surgery Patients given treatment for blood clots
Surgery Patients given treatment for preventive blood clots
Cardiac surgery patients with controlled 6 AM post-operative blood glucose
Surgery Patients with appropriate hair removal
Surgery Patients continued on beta blockers during and after surgery
Description
Table 13
96
97
98
99
Table 14
Quality Component 1 Data by Ownership Groups
Categories
Count
f1
SD
Government - Federal
Government - Hospital District or Authority
Government - Local
Government - State
Proprietary
Voluntary non-prot - Church
Voluntary non-prot - Other
Voluntary non-prot - Private
Total
2
45
23
8
82
126
124
205
615
0.325
7.317
3.740
1.301
13.333
20.488
20.163
33.333
100.00
-0.296
-0.299
0.083
0.487
0.256
0.091
-0.031
-0.007
0
0.687
1.155
1.298
1.525
1.003
1.003
1.016
0.874
1
Table 15
Quality Component 2 Data by Ownership Groups
Categories
Count
f1
SD
Government - Federal
Government - Hospital District or Authority
Government - Local
Government - State
Proprietary
Voluntary non-prot - Church
Voluntary non-prot - Other
Voluntary non-prot - Private
Total
2
45
23
8
82
126
124
205
615
0.325
7.317
3.740
1.301
13.333
20.488
20.163
33.333
100.00
-0.759
-0.068
-0.319
-0.961
0.286
-0.01
0.194
-.130
0
0.474
1.179
1.180
0.630
0.880
1.025
0.860
1.013
1
Table 16
Quality Component 3 Data by Ownership Groups
Categories
Count
f1
SD
Government - Federal
Government - Hospital District or Authority
Government - Local
Government - State
Proprietary
Voluntary non-prot - Church
Voluntary non-prot - Other
Voluntary non-prot - Private
Total
2
45
23
8
82
126
124
205
615
0.325
7.317
3.740
1.301
13.333
20.488
20.163
33.333
100.00
0.680
-0.874
-0.309
-0.409
-0.108
-0.161
0.230
0.239
0
0.310
1.474
1.095
0.765
0.947
1.068
0.844
0.782
1
100
Table 17
Quality Component 4 Data by Ownership Groups
Categories
Count
f1
SD
Government - Federal
Government - Hospital District or Authority
Government - Local
Government - State
Proprietary
Voluntary non-prot - Church
Voluntary non-prot - Other
Voluntary non-prot - Private
Total
2
45
23
8
82
126
124
205
615
0.325
7.317
3.740
1.301
13.333
20.488
20.163
33.333
100.00
-0.764
-0.057
-0.294
-0.2423
0.221
0.0668
-0.036
-0.045
0
1.927
1.120
1.165
1.499
0.897
0.953
0.989
0.996
1
Table 18
Quality Component 5 Data by Ownership Groups
Categories
Count
f1
SD
Government - Federal
Government - Hospital District or Authority
Government - Local
Government - State
Proprietary
Voluntary non-prot - Church
Voluntary non-prot - Other
Voluntary non-prot - Private
Total
2
45
23
8
82
126
124
205
615
0.325
7.317
3.740
1.301
13.333
20.488
20.163
33.333
100.00
0.809
-0.098
-0.154
-0.889
0.403
0.122
-0.107
-0.106
0
1.913
1.006
1.662
1.366
0.898
0.893
0.926
0.977
1
101
hence are more likely to apply Charakas principles in practice than others. For testing the
hypotheses, an OLS regression was run after creating binary variables for the eight ownership
groups. The dependent variables was the principal components of process of care quality. The
church group was the base group. The regression equation (2) in Chapter 3 was used. The
regression approach is equivalent to one-way ANOVA with pairwise contrasts of group means.
Equation (2) will not be used to predict quality components.
The signs of the regression coefcients for all the ownership binary variables (i=1...7 are
expected to be negative by the research hypothesis because the church group hospitals are
expected to have higher quality components than other owner groups. The hypothesis to be tested:
H0 : i=1...7 0 and H1 : i=1...7 < 0. For conceptual clarity, these hypotheses are stated as
one-sided hypotheses to make expected signs clear. To be conservative, however, statistical tests
will be based on two sided p values. This test was repeated for all the ve quality components.
Before testing the hypotheses, a general comparison will rst be made using the estimated
marginal means plots for each satisfaction component obtained by using PASW software are
given in Figures 9, 10, 11, 12, and 13.
102
103
104
District and other government hospitals score below the mean. It is seen that district hospitals
score almost 1 standard deviation below the mean on this quality component.
Quality component 4 smoking cessation related is highest in proprietary, followed by
those run by church and other voluntary nonprots, as seen in Figure 12. Government hospitals
rank below mean. Federal hospitals score almost 0.75 standard deviations below the mean. This
reects general government ambivalence towards controlling the smoking habit.
In the prevention related quality component 5, federal hospitals rank high followed by
proprietary hospitals, as seen in Figure 13. Church group run hospitals rank above other
voluntary nonprots which operate at the mean level. Government hospitals rank below the mean
in general It is seen from these charts that government hospitals, in general are below par in
quality components while church group hospitals run well above par. However, these are only
indicative. Statistical testing was done using multiple regression.
Validating OLS regression assumptions for quality data
1. Independence of observations assumption is valid because the quality observations are truly
independent of each other.
2. Dependent variable observations in each group need to follow normal distribution, but the
test results given in Table 19 show evidence of nonnormality.
3. OLS requires absence of perfect collinearity between explanatory variables. This is not
likely to be the case because the ownership groups are mutually exclusive. Still,
multicollinearity was tested after regression for conrmation.
4. The groups should have homogeneity of variance in the dependent variables. This was
tested using Levenes test which showed some evidence of heteroscedasticity and so robust
standard errors are needed.
Testing normality with PASW resulted in Table 19.
105
Table 19
Normality Testing Process of Care Quality Variables in Ownership Groups
Component
Owner Group
Shapiro-
df
Sig.
Wilk
Statistic
X qual D1
X qual D2
X qual D3
X qual D4
X qual D5
Government - Federal
0.934
45
0.014
Government - Local
0.768
23
0.000
Government - State
0.805
0.032
Proprietary
0.921
82
0.000
0.902
126
0.000
0.937
124
0.000
0.95
205
0.000
Government - Federal
0.918
45
0.004
Government - Local
0.927
23
0.093
Government - State
0.91
0.352
Proprietary
0.929
82
0.000
0.873
126
0.000
0.894
124
0.000
0.91
205
0.000
Government - Federal
0.92
45
0.004
Government - Local
0.969
23
0.655
Government - State
0.942
0.635
Proprietary
0.955
82
0.006
0.93
126
0.000
0.898
124
0.000
0.967
205
0.000
Government - Federal
0.791
45
0.000
Government - Local
0.821
23
0.001
Government - State
0.871
0.154
Proprietary
0.594
82
0.000
0.773
126
0.000
0.784
124
0.000
0.802
205
0.000
Government - Federal
0.981
45
0.68
106
Component
Owner Group
Shapiro-
df
Sig.
Wilk
Statistic
Government - Local
0.933
23
0.127
Government - State
0.929
0.511
Proprietary
0.915
82
0.000
0.959
126
0.001
0.94
124
0.000
0.943
205
0.000
107
Table 20
RQ3: Regression Results for Quality Component 1 by Ownership Group
DV: Quality component 1
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
2
45
23
8
82
458
124
205
-0.39
-0.39
-0.01
0.40
-0.07
-0.12
-0.10
0.09
0.36
0.19
0.28
0.52
0.14
0.13
0.11
0.09
-1.08
-2.01
-0.03
0.77
-0.45
-0.94
-0.89
0.99
0.28
0.05
0.98
0.44
0.65
0.35
0.38
0.32
-1.09
-0.77
-0.56
-0.62
-0.35
-0.38
-0.31
-0.09
0.32
-0.01
0.55
1.41
0.22
0.13
0.12
0.27
0.38
0.37
0.01
0.38
0.06
0.12
0.10
0.65
0.64
0.50
0.65
0.52
0.55
0.54
Own code
Government - federal
Government - hospital area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
voluntary non-pr-other
Voluntary non-pr-private
Constant term
108
high. The normality robust Levenes test failed to reject homoscedasticity (Pr > F = 0.47).
Figure 14. RQ3: Kernel Density Plot of Residuals for Quality Component 1
Multicollinearity is not a problem, as shown by the low variable ination factor of 1.32.
Multicollinearity was not expected to be a problem because the predictor variables (ownership
group) are denitely not correlated. Ramsey reset test for model specication errors was not run
because all explanatory variables are dummy variables. In summary, the regression diagnostic
tests showed the adequacy of OLS regression.
Sensitivity analysis.
Predicted R2 is within 0.2 of R2 = 0.0121 shows cross-validation. The low values of
Predicted R2 and R2 show that the present publicly reported quality indicators may be too broad.
R2 being small, we need to rule out the possibility of a few inuential outliers affecting the model
predictions. A procedure to identify the inuential outliers as was used earlier in RQ2 was used
and this identied only one inuential outlier, a hospital (owned by Local government group)
109
Table 21
RQ3 : Robust Regression Results for Quality Component 1
Robust Regression Results
SE
P > |t|
[95%
CI]
Owner Group
Government - federal
Government - hosp. area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
Voluntary non-pr-other
Voluntary non-pr-private
Constant term
-0.53
-0.39
0.129
0.72
-0.1
-0.15
-0.18
0.234
0.65
0.16
0.21
0.33
0.13
0.116
0.104
0.082
-0.81
-2.44
0.62
2.15
-0.77
-1.31
-1.72
2.86
0.418
0.015
0.536
0.032
0.443
0.192
0.085
0.004
-1.816
-0.702
-0.28
0.063
-0.356
-0.38
-0.38
0.074
0.76
-0.076
0.538
1.378
0.156
0.076
0.0249
0.39
which scored -0.33 in the quality component 1 and a residual of -4.67. Robust regression after
weighting this case by 0 and other cases appropriately, gave an F(7,607) = 2.26 which was
signicant at 5% condence level (P > |t| = 0.0279). On the basis of two-sided tests of
signicance, church group hospitals had signicant differences in quality component 1 from
hospitals owned by district or area ownership group. However, robust regression showed that state
government hospitals have signicantly higher score on quality component 1 than church group
hospitals (95% CI [0.06, 1.38]). This is surprising because only a local government hospital was
identied as an inuential outlier. On closer inspection of the data, it was found that among the 8
state government hospitals that have a component score, six have a weight close to 1 and one has
a weight close to 0. Seven of the residuals have a cooks d < 0.04 while only one is > 3.6. Robust
regression result with respect to state government group is attributed to many causes such as low
group count, low variability of the scores within the group and one possible inuential outlier
close to being assigned zero weight. The signs of the coefcients were the same as before. This
shows that the test results are not seriously affected by the presence of outlier, except the level of
signicance. In summary, robust regression conrmed OLS regression results. The research
hypothesis is supported in one out of seven owner groups (government-hospital area/authority).
The robust regression results are given in Table 21.
Regression results for quality component 2 (pneumonia related)
results are given in Table 22.
The regression
110
Table 22
RQ3: Regression Results for Quality Component 2 by Ownership Group
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
-0.75
0.26
-2.93
0.00
-1.25
-0.25
0.74
0.77
45
-0.06
0.20
-0.30
0.77
-0.45
0.33
0.05
0.52
government - local
23
-0.31
0.26
-1.20
0.23
-0.82
0.20
0.30
0.62
government - state
-0.95
0.23
-4.16
0.00
-1.40
-0.50
0.95
0.83
82
0.30
0.13
2.21
0.03
0.03
0.56
0.31
0.62
458
voluntary non-pr-other
124
0.20
0.12
1.70
0.09
-0.03
0.44
0.22
0.59
voluntary non-pr-private
205
-0.12
0.12
-1.04
0.30
-0.35
0.11
0.12
0.55
-0.01
0.09
-0.10
0.92
-0.19
0.17
proprietary
Constant term
Summary of regression results:
R2
= 0.04, Adjusted
R2
Regression diagnostics:
Model degrees of freedom = 7, Residual degrees of freedom = 607, F(7,607) = 7.316, Prob. > F = 0.0000
RMSE = 0.985, Model sum of squares = 25.98, Residual sum of squares = 589.0
Levenes robust test for heteroscedasticity using median: W50 = 1.579, df (7, 607), Pr > F = 0.139,
Shapiro-Wilk test for normality of residuals: W=0.915, Z= 8.6, Prob > Z = 0.000,
R2 has a very low value of 0.04 showing that the regression only explains about 4% of the
variation in response variable. This was expected because hospitals have to maintain a minimum
level of quality; otherwise, face shut down by accreditation and government agencies. The
adjusted R2 takes into account the number of regression parameters, and is lower = 0.0312.
RMSE is 0.985, showing that the model t is low. Predicted R2 is 0.02 within 0.2 of R2 , showing
that similar results may be expected from new sample. F(7,607) = 7.316 is signicant (Prob. > F
= 0.0000). On the basis of two-sided tests of signicance, three of the regression coefcients are
signicant, showing that federal and state government hospitals score signicantly less than
church run hospitals on quality component 2 (heart attack/failure related), while proprietary
hospitals score signicantly higher. The effect size for federal government hospitals is between
Cohens limits for medium and high effects, while the effect size is high for state government
hospitals. The effect size is low to medium for proprietary hospitals. The Common Language
Effect Size (CLES) shows that church hospitals have higher mean pneumonia quality score than
111
77% of federal government and 83% of state government hospitals. In summary, the research
hypothesis is supported in respect of two out of seven ownership groups. Church group hospitals
score at least as much as other groups, while only the proprietary group scores higher than church
group in quality component 2.
Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (2) are reported below Table 22. Shapiro-Wilk normality test showed that the
residuals are not normally distributed, but this is not critical because of the large sample size (N >
600). Also, a kernel density plot (given in Figure 15) shows that the deviation from normal is not
high. The normality robust Levenes test failed to reject homoscedasticity (Pr > F = 0.14).
Figure 15. RQ3: Kernel Density Plot of Residuals for Quality Component 2
Multicollinearity is not a problem, as shown by the low variable ination factor of 1.32.
Multicollinearity was not expected to be a problem because the independent variables (ownership
112
Table 23
RQ3 : Robust Regression Results for Quality Component 2
Robust Regression Results
SE
P > |t|
[95%
CI]
Owner Group
Government - federal
Government - hosp. area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
Voluntary non-pr-other
Voluntary non-pr-private
Constant term
-0.91
0.021
-0.263
-1.111
0.232
0.158
-0.128
0.154
0.60
0.15
0.19
0.30
0.12
0.106
0.095
0.075
-1.52
0.14
-1.38
-3.63
1.94
1.48
-1.34
2.06
0.128
0.89
0.168
0.000
0.052
0.138
0.179
0.040
-2.09
-0.27
-0.64
-1.71
-0.002
-0.051
-0.315
0.007
0.264
0.308
0.111
-0.51
0.47
0.367
0.059
0.301
group) are denitely not correlated. Ramsey reset test for model specication errors was not run
because all explanatory variables are dummy variables. In summary, the regression diagnostics
show that OLS regression is valid here.
Sensitivity analysis.
Predicted R2 having 0.02 value, is within 0.2 of R2 = 0.04. This shows cross-validation,
because similar results are expected for new cases as well. R2 being small, we need to rule out the
possibility of a few inuential outliers affecting the model predictions. A procedure to identify
the inuential outliers as was used earlier in RQ2 was used and this identied only four inuential
outliers, two hospitals owned by Nonprot Private group and two owned by church group. They
had standardized residuals less than -3.65. Robust regression, gave an F(7,607) = 4.85 which was
signicant (P > |t| = 0.0000). However, only one regression coefcient was found to be
signicant. State government hospitals score signicantly less (95% CI [-1.7 -0.5]) than church
run hospitals on quality component 2. This shows that the test results are affected by the presence
of outliers. Two of the hospitals owned by church group turned out to be inuential outliers and
this has caused the difference in the results. However, the research hypothesis is still supported
because church group hospitals turned out to have higher quality than state government owned
hospitals and not lower quality than other six groups. The robust regression results are given in
Table 23.
113
Regression results for quality component 3 (surgical care related) The regression
results are given in Table 24.
Table 24
RQ3: Regression Results for Quality Component 3 by Ownership Group
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
0.84
0.18
4.59
0.00
0.48
1.20
0.80
0.79
45
-0.71
0.24
-2.99
0.00
-1.18
-0.24
0.60
0.73
Government - local
23
-0.15
0.24
-0.61
0.54
-0.63
0.33
0.14
0.56
Government - state
-0.25
0.27
-0.91
0.36
-0.78
0.29
0.24
0.59
82
0.05
0.14
0.37
0.71
-0.23
0.33
0.05
0.52
458
Voluntary non-pr-other
124
0.39
0.12
3.20
0.00
0.15
0.63
0.41
0.66
Voluntary non-pr-private
205
0.40
0.11
3.63
0.00
0.18
0.62
0.44
0.67
-0.16
0.10
-1.68
0.09
-0.35
0.03
Proprietary
Constant term
Summary of regression results:
R2
= 0.0996, Adjusted
Regression diagnostics:
R2
Model degrees of freedom = 7, Residual degrees of freedom = 607, F (7.607) = 8.793, Prob. > F = 0.0000,
RMSE = 0.955, Model sum of squares = 61.24, Residual sum of squares = 553.8
Levenes robust test for heteroscedasticity using median: W50 = 4.63, df (7, 607), Pr > F = 0.000,
Shapiro-Wilk test for normality of residuals: W = 0.951, Z = 7.27, Prob > Z = 0.000,
R2 has a very low value of 0.0996 showing that the regression only explains 10% of the
variation in response variable. This was expected because hospitals have to maintain a minimum
level of quality; otherwise, face shut down by accreditation and government agencies. The
adjusted R2 takes into account the number of regression parameters, and is lower = 0.0892. The
RMSE is high at 0.955, showing that the model t is low. Predicted R2 is 0.075, is low but close
to R2 within 0.2. The overall statistic F(7,607) = 8.793 is signicant (Prob. > F = 0.0000). On
the basis of two-sided tests of signicance, four of the regression coefcients are signicant,
showing that local government hospitals score signicantly less than church run hospitals on
quality component 3. 95% CI for the difference was [-1.18, -0.24] and the effect size was between
medium and high. However, federal government hospitals (95% CI [0.48, 1.20], high effect size),
114
nonprot other hospitals (95% CI [0.15, 0.63], low to medium effect size) and nonprot private
hospitals (95% CI [0.18, 0.62], low to medium effect size) scored better than church hospitals .
CLES showed that 79% of federal hospitals, 66% of nonprot other and 67% of nonprot private
hospitals score higher on this quality component than church hospitals. In summary, the research
hypothesis is supported in one group (District/area) but three groups (federal government,
nonprot-other and non-prot-private) have scored higher than church group in quality
component 3.
Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (2) are reported below Table 24. Shapiro-Wilk normality test showed that the
residuals are not normally distributed, but, this is not critical because of the large sample size (N
> 600). Also, a kernel density plot (given in Figure 16) shows that the deviation from normal is
not high. The normality robust Levenes test rejected homoscedasticity (Pr > F = 0.000),
Figure 16. RQ3: Kernel Density Plot of Residuals for Quality Component 3
115
making it necessary to run the regression with heteroscedasticity robust standard errors.
Multicollinearity is not a problem, as shown by the low variable ination factor of 1.32.
Multicollinearity was not expected to be a problem because the independent variables (ownership
group) are denitely not correlated. Ramsey reset test for model specication errors was not run
because all explanatory variables are dummy variables. In summary, the regression diagnostics
showed that OLS regression results are valid. Sensitivity analysis.
Predicted R2 has a low value of 0.075, close to R2 = 0.1 shows cross-validation for the
model, because new cases may also behave similarly. R2 being small, we need to rule out the
possibility of a few inuential outliers affecting the model predictions. A procedure to identify
the inuential outliers as was used earlier in RQ2 was used and this identied three inuential
outliers, a hospital (owned by church group, another owned by district or authority and a third
owned by nonprot other group. Robust regression after weighting this case by 0 and other cases
appropriately, gave an F(7,607) = 5.85 which was signicant (P > |t| = 0.0000). On the basis of
two-sided tests of signicance, three regression coefcients were signicant. Church group
hospitals had signicantly higher scores than in quality component 3 than hospitals owned by
government hospital area/authority group (95% CI [-0.73, -0.14]) and had signicantly lower
scores than nonprot - other (95% CI [0.84, 0.52]) and nonprot - private (95% CI [0.075,
0.465]). This shows that the test results are affected by the presence of inuential outliers,
particularly when one of them was a church-owned hospital. In summary, the research hypothesis
is still supported in one group (District/area) but, two owner groups (nonprot-other and
nonprot-private) scored signicantly higher than church group owned hospitals in quality
component 3. The robust regression results are given in Table 25.
Regression results for quality component 4 smoking cessation related
regression results are given in Table 26.
The
116
Table 25
RQ3 : Robust Regression Results for Quality Component 3
Robust Regression Results
SE
P > |t|
[95%
CI]
Owner Group
Government - federal
Government - hosp. area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
Voluntary non-pr-other
Voluntary non-pr-private
Constant term
0.66
-0.44
-0.21
-0.38
-0.036
0.302
.27
0.0176
0.62
0.152
0.199
0.319
0.124
0.110
0.099
0.078
1.06
-2.87
-1.07
-1.20
-0.29
2.73
2.72
0.23
0.289
0.004
0.285
0.232
0.774
0.007
0.007
0.821
-0.564
-0.735
-0.603
-1.010
-0.28
0.084
0.075
-0.136
1.89
-0.137
0.178
0.245
0.208
0.52
0.465
0.171
Table 26
RQ3: Regression Results for Quality Component 4 by Ownership Group
DV: Quality component 4
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
Own code
Government - federal
-0.83
0.97
-0.85
0.39
-2.74
1.08
0.87
0.81
45
-0.12
0.19
-0.67
0.51
-0.49
0.24
0.13
0.55
Government - local
23
-0.36
0.25
-1.42
0.16
-0.86
0.14
0.37
0.64
Government - state
-0.31
0.51
-0.61
0.54
-1.30
0.69
0.31
0.62
82
0.15
0.13
1.18
0.24
-0.10
0.41
0.17
0.57
458
Voluntary non-pr-other
124
-0.10
0.12
-0.83
0.41
-0.34
0.14
0.11
0.54
Voluntary non-pr-private
205
-0.11
0.11
-1.02
0.31
-0.33
0.10
0.11
0.54
0.07
0.09
0.78
0.43
-0.10
0.23
Proprietary
Constant term
Summary of regression results:
Model degrees of freedom = 7, Residual degrees of freedom = 607, F (7,607) = 1.22, Prob. > F = 0.2927
R2 = 0.0145, Adjusted R2 = 0.00313, Predicted R2 = 0,
RMSE = 0.999, Model sum of squares = 8.912, Residual sum of squares = 606.1
Regression diagnostics:
Levenes robust test for heteroscedasticity using median: W50 = 1.62, df (7, 607), Pr > F = 0.127,
Shapiro-Wilk test for normality of residuals: W = 0.79, Z = 10.78, Prob Z = 0.000,
Multicollinearity test: Mean VIF = 1.32
R has a very low value of 0.015 showing that the regression only explains 1.5% of the variance.
This was expected because hospitals have to maintain a minimum level of quality; otherwise, face
117
shut down by accreditation and government agencies. The adjusted R2 takes into account the
number of regression parameters, and is lower = 0.003.The RMSE has a high value of 0.999,
showing that the model t is low. Predicted R2 is 0, is low, but within 0.2 of R2 . The overall F
statistic F (7,607) = 1.22 is not signicant at 5% level (Prob. > F = 0.2927). On the basis of
two-sided tests of signicance, none of the coefcients is signicant. Nevertheless. there is
wide variation in each group and in effect size. In summary, the research hypothesis is not
supported for quality component 4 with no signicant difference among the different owner
groups.
Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (2) are reported below Table 26. Shapiro-Wilk normality test showed that the
residuals are not normally distributed (Prob > Z = 0.000) , but this is not critical because of the
large sample size (N > 600). The kernel density plot (given in Figure 17) shows considerable
deviation from normal. The normality robust Levenes test failed to reject homoscedasticity
(Pr > F = 0.127). Multicollinearity is not a problem, as shown by the low variable ination
factor of 1.32. Multicollinearity was not expected to be a problem because the independent
variables (ownership group) are denitely not correlated. Ramsey reset test for model
specication errors was not run because all explanatory variables are dummy variables. In
summary, the regression diagnostics show that OLS regression is valid.
Sensitivity analysis.
Predicted R2 has 0 value, but within 0.2 of R2 , showing that similar results are expected
for new samples. R2 being small, we need to rule out the possibility of a few inuential outliers
affecting the model predictions. A procedure to identify the inuential outliers as was used earlier
in RQ2 was used and this identied 29 inuential outliers. Robust regression after weighting
these cases by 0 and other cases appropriately, gave an F(7,607) = 3.82 which was signicant (P
> |t| = 0.0004). On the basis of two-sided tests of signicance, three regression coefcients were
signicant and two of them were negative as expected. Church group hospitals had signicantly
118
Figure 17. RQ3: Kernel Density Plot of Residuals for Quality Component 4
higher scores in quality component 4 than hospitals owned by federal government (95% CI
[-1.91, -0.28], and hospitals owned by local governments (95% CI [-0.57, -0.051]. However, they
had signicantly lower scores than state government hospitals (95% CI [0.05, 0.88]). This shows
that the test results are greatly affected by the inuential outliers. In summary, the robust
regression test results support the research hypothesis in two out of seven owner groups (federal
and local governments) , and not support in only one owner group (state government). The robust
regression results are given in Table 27.
Regression results for quality component 5 prevention related
are given in Table 28. . R2 has a very low value of 0.045 showing that the regression only
explains about 4.5% of the variance. This was expected because hospitals have to maintain a
minimum level of quality; otherwise, face shut down by accreditation and government agencies.
The adjusted R2 takes into account the number of regression parameters, and is lower = 0.034.
119
Table 27
RQ3 : Robust Regression Results for Quality Component 4
Robust Regression Results
SE
P > |t|
[95%
CI]
Owner Group
Government - federal
Government - hosp. area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
Voluntary non-pr-other
Voluntary non-pr-private
Constant term
-1.10
-0.04
-0.31
0.463
0.047
-0.092
-0.109
0.337
0.415
0.101
0.132
0.212
0.083
0.074
0.066
0.052
-2.65
-0.36
-2.35
2.18
0.57
-1.25
-1.65
6.49
0.008
0.716
0.019
0.030
0.570
0.212
0.100
0.000
-1.92
-0.235
-0.57
0.0459
-0.115
-0.237
-.238
0.235
-0.286
0.1619
-0.052
0.880
0.209
0.053
0.021
0.439
Table 28
RQ3: Regression Results for Quality Component 5 by Ownership Group
DV: Quality component 5
Own code
Government - federal
Government - hospital area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
Voluntary non-pr-other
Voluntary non-pr-private
Constant term
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
2
45
23
8
82
458
124
205
0.69
-0.22
-0.28
-1.01
0.28
-0.23
-0.23
0.12
0.97
0.17
0.35
0.46
0.13
0.12
0.11
0.08
0.71
-1.30
-0.79
-2.19
2.21
-1.99
-2.17
1.53
0.48
0.19
0.43
0.03
0.03
0.05
0.03
0.13
-1.21
-0.55
-0.96
-1.92
0.03
-0.46
-0.43
-0.03
2.58
0.11
0.41
-0.10
0.53
-0.003
-0.02
0.28
0.76
0.24
0.27
1.10
0.32
0.25
0.24
0.78
0.59
0.61
0.86
0.63
0.60
0.59
The RMSE is high at 0.984, showing that the model t is low. Predicted R2 is 0, but within 0.2 of
R2 . The overall F statistic (F(7,607) = 3.975) is signicant (Prob. > F = 0.0003). On the basis of
two-sided tests of signicance, four of the regression coefcients are signicant, and three of
them are negative as expected: state government hospitals score signicantly less (95% CI [-1.92,
-0.10], high effect size); nonprot other hospitals score signicantly less (95% CI [-0.46, -0.003],
low effect size) and nonprot private hospitals scoring signicantly less than church run hospitals
(95% CI [-0.43, -0.02], low effect size). Only proprietary hospitals score higher than church
120
hospitals (95% CI [0.03, 0.53], effect size between low and medium). In summary, the research
hypothesis is supported in three out of seven owner groups (state government, nonprot-other,
and nonprot-private), while one group (proprietary) scored signicantly higher than church
group on quality component 5.
Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (2) are reported below Table 28. Shapiro-Wilk normality test showed that the
residuals are not normally distributed, but this is not critical because of the large sample size (N >
600). The kernel density plot also (given in Figure 18) does not show considerable deviation
from normal. The normality robust Levenes test rejected homoscedasticity (Pr > F = 0.0025),
Figure 18. RQ3: Kernel Density Plot of Residuals for Quality Component 5
justifying regression with robust errors. Multicollinearity is not a problem, as shown by the low
variable ination factor of 1.32. Multicollinearity was not expected to be a problem because the
121
Table 29
RQ3 : Robust Regression Results for Quality Component 5
Robust Regression Results
SE
P > |t|
[95%
CI]
Owner Group
Government - federal
Government - hosp. area/authority
Government - local
Government - state
Proprietary
Voluntary non-pr-church (base group)
Voluntary non-pr-other
Voluntary non-pr-private
Constant term
0.608
-0.251
0.13
-0.925
0.299
-0.205
-0.186
0.2015
0.636
0.155
0.202
0.325
0.127
0.1129
0.101
0.079
0.96
-1.62
0.64
-2.84
2.36
-1.81
-1.84
2.53
0.340
0.105
0.521
0.005
0.018
0.070
0.066
0.011
-0.64
-0.56
-0.268
-1.564
0.051
-0.427
-0.384
.0454
1.857
0.053
0.527
-0.286
0.548
0.0168
0.0126
0.358
independent variables (ownership group) are denitely not correlated. Ramsey reset test for
model specication errors was not run because all explanatory variables are dummy variables. In
summary, regression diagnostics support the use of OLS regression.
Sensitivity analysis.
Predicted R2 having 0 value, is within 0.2 of R2 = 0.04, shows cross-validation. R2 being
small, we need to rule out the possibility of a few inuential outliers affecting the model
predictions. A procedure to identify the inuential outliers as was used earlier in RQ2 was used
and this identied three inuential outliers, two hospitals owned by local governments and one
owned by proprietary group. They had standardized residuals less than -3.6. Robust regression,
gave an F(7,607) = 4.63 which was signicant (P > |t| = 0.0000). On the basis of two-sided tests
of signicance, two regression coefcients were found to be signicant. State government
hospitals score signicantly less (95% CI [-1.6 -0.3]) than church run hospitals on quality
component 5. Proprietary hospitals scored higher (95% CI [0.05, 0.55]). This shows that the test
results are affected by the presence of outliers. In summary, robust regression supports the
research hypothesis in one of the seven owner groups (state government) while proprietary group
scored signicantly higher than church group in quality component 5. The robust regression
results are given in Table 29.
Summary of Regression results for RQ-3. The OLS regression coefcients for RQ-3
are summarized in Table 30. The coefcients that are not statistically signicant are shown with
122
ns in the table. The effect sizes (Cohens d) are given within parentheses, where the
coefcients are signicant.
Table 30
RQ3 - Summary of OLS Regression Coefcients on Quality Components by Ownership Groups
Ownership group
Government - Federal
Government - Hospital District or
Authority
Government - Local
Government - State
Proprietary
Voluntary
non-prot
((Base
group) - Church
Voluntary non-prot - Other
Voluntary non-prot - Private
Heart attack/failure
Quality components
Pneumonia
Surgical care
Smoking cessation
Prevention
ns
-0.39 (0.37)
-0.75 (0.74)
ns
0.84 (0.80)
-0.71 (0.60)
ns
ns
ns
ns
ns
ns
ns
ns
-0.95 (0.95)
0.3 (0.31)
ns
ns
ns
ns
ns
ns
ns
-1 (1.10)
0.28 (0.32)
ns
ns
ns
ns
0.39 (0.41)
0.4 (0.44)
ns
ns
-0.23 (0.25)
-0.23 (0.24)
Research Question 4.
What is the relationship between process of care quality measures and patient
satisfaction?
To analyze this research question, the quality and HCAHPS components found earlier
were used. All the ve HCAHPS components have to be used because these are correlated;
otherwise, the regression model would become misspecied. An OLS regression will be carried
out for each principal component of quality as the response variable and all the patient satisfaction
components as explanatory variables. The regression equation (3) in Chapter 3 was used.
The term u is the error or disturbance term and contains unobserved factors such as
hospital characteristics, patient characteristics and errors in process of care quality and patient
satisfaction. In fact, these omitted explanatory variables in the error term are likely to have a
much larger contribution to R2 than the patient satisfaction components.
The research hypothesis assumes that hospitals may be compromising on quality to gain
patient satisfaction. Therefore, the signs of the regression coefcients should be negative for all
quality components except quality component 1, being poor satisfaction related, will have
positive coefcients.
123
124
Table 31
RQ4 - Regression Results for Quality Component 1 by Satisfaction
Components
DV: Quality component 1
Robust SE
P > |t|
[95%
CI]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
0.10
0.20
-0.40
0.30
-0.17
-0.37
0.08
0.07
0.07
0.07
0.06
0.08
1.34
3.05
-5.62
4.03
-2.62
-4.91
0.18
0.00
0.00
0.00
0.01
0.00
-0.05
0.07
-0.54
0.15
-0.29
-0.52
0.25
0.33
-0.26
0.44
-0.04
-0.22
parameters, and is lower = 0.08. The RMSE is high (=0.96), showing that the model t is low.
Predicted R2 is 0.07 and is within 0.2 of R2 . The overall F statistic (F(5,607) = 11.57) is
signicant (Prob > F = 0.000). On the basis of two-sided tests of signicance, four regression
coefcients are signicant, showing that only the rst component of quality is not signicantly
affected by the poor satisfaction component. While the satisfaction components 3 (95ES was
between low and medium limits and the power is seen to be high (= 0.995). In summary, the
research hypothesis is supported in two satisfaction components (3 and 5) but not in the other 3
satisfaction components. Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (3) are reported below Table 31. Shapiro-Wilk normality test showed that the
residuals are not normally distributed, but this is not critical because of large sample size (N >
600). The kernel density plot (given in Figure 19), also does not show considerable deviation
from normal. The Whites test rejected homoscedasticity with p = 0.0042, justifying the use of
robust standard errors. Multicollinearity was not a problem, as shown by the low variable ination
factor of 1.83. This was surprising because the independent variables () were correlated. as shown
125
Figure 19. RQ4: Kernel Density Plot of Residuals for Quality Component 1
earlier. Ramsey reset test using powers of tted values for model specication errors was not
signicant (Prob > F = 0.61). In summary, the regression diagnostics support OLS assumptions.
Sensitivity analysis.
Predicted R2 0.069 is within 0.2 of R2 , and this show cross-validation. New cases are
likely to show similar results. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in RQ2 was used and this identied only two inuential outliers. They had
standardized residuals < -4. Robust regression, gave an F( 5, 607) = 10.51 which was signicant
(P > |t| = 0.0000). All the ve coefcient were found to be signicant. Components 3 and 5
have negative coefcients as found earlier with OLS regression. Poor satisfaction component
which was not signicant earlier, has now become signicant and positive (95% CI [0.03, 0.29])
This shows that the test results are affected by the presence of inuential outliers. In summary,
126
Table 32
RQ4 : Robust Regression Results for Quality Component 1
DV: Quality component 1
Robust SE
P > |t|
[95%
CI]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
0.16
0.12
-0.35
0.26
-0.198
-0.217
0.066
0.059
0.069
0.069
0.0575
0.0631
2.45
2.07
-5.16
3.71
-3.45
-3.45
0.015
0.039
0.000
0.000
0.001
0.001
0.032
0.0064
-0.488
0.121
-0.311
-.341
0.29
0.24
-0.219
0.394
-0.085
-0.0935
Table 33
RQ4 - Regression Results for Quality Component 2 by Satisfaction Components
Robust SE.
P > |t|
[95% Conf.
Interval]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
-0.06
0.09
0.03
-0.26
0.05
0.14
0.08
0.07
0.07
0.07
0.07
0.06
-0.79
1.28
0.38
-3.59
0.74
2.26
0.43
0.20
0.71
0.00
0.46
0.02
-0.22
-0.05
-0.12
-0.41
-0.09
0.02
0.09
0.23
0.17
-0.12
0.19
0.26
Summary of regression results: Model degrees of freedom = 5, Residual degrees of freedom = 607, F (5,607) = 3.379,
RMSE = 0.992, Model sum of
Prob. > F = 0.0051
R2 = 0.0268, Adjusted R2 = 0.0188, Predicted R2 = 0.0188,
squares = 16.45, Residual sum of squares = 597.6
Effect size (ES:
f 2 = 0.03; Cohen (1988) considers f 2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 16.88, Power = 0.930 (From Table 9.3.2 of Cohen (1988) for a=0.05)
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 26.31, df = 20, p = 0.1558,
Shapiro-Wilk test
for normality of residuals: W = 0.927, Z= 8.21, Prob > Z = 0.00,
Multicollinearity test: Mean VIF = 1.83,
Ramsey RESET test using powers of tted values: F(3,604) = 0.84, Prob > F = 0.4724
robust regression did not alter the earlier conclusion that the research hypothesis is supported in
two satisfaction components (3 and 5) but in not in the other three. Table 32 shows the robust
regression results.
Regression results for quality component 2 (pneumonia related) The regression
results are given in Table 33.
R2 has a very low value of 0.0268 showing that the regression only explains about 2.8%
of the response variable. This was expected because hospitals have to maintain a minimum level
of quality; otherwise, face shut down by accreditation and government agencies. The adjusted R2
127
takes into account the number of regression parameters, and is lower and = 0.02. The RMSE is
high (=0.992), showing that the model t is low. Predicted R2 is 0.02 and is within 0.2 of R2 . The
overall F statistic (F(5,607) = 3.38) is signicant at 5% level (Prob > F = 0.0051). On the basis
of two-sided tests of signicance, only one regression coefcients is signicant, showing that
only the satisfaction component 4 (Cleanliness) is signicantly related to quality component 2
(95% CI [-0.41, -0.12]) and the sign is negative as expected. One unit increase in Cleanliness is
associated with a decrease of 0.26 units in pneumonia related quality. ES was between small
and medium limits and the power is seen to be high (= 0.930). In summary, the research
hypothesis is supported in one satisfaction component (4). The other satisfaction components are
not signicantly related.
Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (3) are reported below Table 33. Shapiro-Wilk normality test showed that the
residuals are not normally distributed (p = 0.00), but this is not critical because of the large
sample size (N > 600). The kernel density plot (given in Figure 20), also does not show
considerable deviation from normal. The Whites test failed to reject homoscedasticity with p =
0.16. Multicollinearity was not a problem, as shown by the low variable ination factor of 1.83.
This was surprising because the independent variables (satisfaction component) were correlated,
as shown earlier . Ramsey reset test using powers of tted values for model specication errors
was not signicant (Prob > F = 0.47). in summary, the regression diagnostics support the OLS
regression carried out.
Sensitivity analysis.
Predicted R2 0.02 is within R2 = 0.027, show cross-validation. New cases are likely to
show similar test results. R2 being small, we need to rule out the possibility of a few inuential
outliers affecting the model predictions. A procedure to identify the inuential outliers, used
earlier in RQ2 was used and this identied three inuential outliers. They had standardized
residuals < -3.8. Robust regression, gave an F( 5, 607) = 2.45 which was signicant at 5% level
128
Figure 20. RQ4: Kernel Density Plot of Residuals for Quality Component 2
(P > |t| = 0.0326). Only one coefcient was found to be signicant. Component 4
(Cleanliness) was found to be signicant (95% CI [-0.32, -06]) with a negative sign as found
earlier with OLS regression. A unit increase in Cleanliness is associated with a decrease of 0.19
units units in pneumonia related quality. This is close to the previous nding of 0.26. In
summary, the test results are not affected by the presence of inuential outliers. As before, the
research hypothesis is supported in patient satisfaction component 4, but not in the other four
components. The robust regression results are given in Table 34
Regression results for quality component 3 (surgical care related) The regression
results are given in Table 35.
R2 has a very low value of 0.056 showing that the regression only explains about 5.6% of
the response variable. This was expected because hospitals have to maintain a minimum level of
quality; otherwise, face shut down by accreditation and government agencies. The adjusted R2
129
Table 34
RQ4 : Robust Regression Results for Quality Component 2
DV: Quality component 1
Robust SE
P > |t|
[95%
CI]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
0.02
0.08
-0.044
-0.194
0.031
0.201
0.064
0.0577
0.0667
0.0674
0.0559
0.0614
0.33
1.45
-0.67
-2.88
0.56
3.28
0.743
0.148
0.503
0.004
0.576
0.001
-0.105
-0.03
-0.176
-0.327
-0.078
0.0806
0.147
0.197
0.086
-0.062
0.141
0.322
Table 35
RQ4 - Regression Results for Quality Component 3 by Satisfaction Components
Robust SE.
P > |t|
[95% Conf.
Interval]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
-0.21
0.21
0.04
-0.01
-0.14
-0.03
0.07
0.07
0.08
0.07
0.06
0.07
-2.93
2.94
0.50
-0.16
-2.22
-0.42
0.00
0.00
0.62
0.88
0.03
0.68
-0.35
0.07
-0.11
-0.16
-0.27
-0.16
-0.07
0.35
0.19
0.14
-0.02
0.10
Summary of regression results: Model degrees of freedom = 5, Residual degrees of freedom = 607, F (5,607) = 8.168,
Prob. > F = 0.0000
R2 = 0.056, Adjusted R2 = 0.049, Predicted R2 = 0.037,
RMSE = 0.976, Model sum of squares = 34.53, Residual sum of squares = 578.2,
Effect size ES:
f 2 = 0.06; Cohen (1988) considers R2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 36.36, Power = 0.995 (From Table 9.3.2 of Cohen (1988) for a=0.05),
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 41.82, df = 20, p = 0.0029,
Shapiro-Wilk test
for normality of residuals: W = 0.929, Z= 8.16, Prob > Z = 0.00,
Multicollinearity test: Mean VIF = 1.83,
Ramsey RESET test using powers of tted values: F(3,604) = 1.04, Prob > F = 0.3744
130
takes into account the number of regression parameters, and is lower = 0.049. The RMSE is high
(=0.976), showing that the model t is low. Predicted R2 is 0.037 and is within 0.2 of R2 . The
overall F statistic (F(5,607) = 8.168) is signicant (Prob > F = 0.0000). On the basis of
two-sided tests of signicance, three regression coefcients (satisfaction components 1,2 and 5)
are signicant, showing that the satisfaction components 5 (Post hospitalization care) is
signicantly related to quality component 3 (95% CI [-0.27, -0.02]) and the sign is negative as
expected. One unit increase in Post hospitalization care is associated with a decrease of 0.14
units in surgical care related quality. The coefcient for Poor satisfaction component
(95expected. Similarly, expected service component was also signicant (95% CI [0.07, 0.35])
but with a positive sign while expected to be negative. ES was between small and medium limits
and the power is seen to be high (= 0.995). In summary, the research hypothesis is supported in
one satisfaction component (Post hospitalization care), but not in two of the other four
components.
Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (3) are reported below Table 35. Shapiro-Wilk normality test showed that the
residuals are not normally distributed (p = 0.00), but this is not critical because of the large
sample size (N > 600). The kernel density plot (given in Figure 21), also does not show
considerable deviation from normal. The Whites test rejected homoscedasticity with p = 0.0029,
requiring the use of robust standard errors in regression. Multicollinearity was not a problem, as
shown by the low variable ination factor of 1.83. This was surprising because the independent
variables (satisfaction component) were correlated, as shown earlier . Ramsey reset test using
powers of tted values for model specication errors was not signicant (Prob > F = 0.37). In
summary, the regression diagnostics support the way, the OLS regression was carried out.
Sensitivity analysis.
Predicted R2 0.037 is within 0.2 of R2 = 0.056, showing cross-validation. New cases are
likely to give similar test results. R2 being small, we need to rule out the possibility of a few
131
Figure 21. RQ4: Kernel Density Plot of Residuals for Quality Component 3
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in RQ2 was used and this identied ve inuential outliers. They had standardized
residuals < -3.96. Robust regression, gave an F( 5, 607) = 7.88 which was signicant at 5% level
(P > |t| = 0.0000). Three coefcients (satisfaction components 1,2 and 5) were found to be
signicant. Component 5 (Post hospitalization care) was found to be signicant (95% CI
[-0.22, -0.004]) with a negative sign as found earlier with OLS regression. A unit increase in
Post hospitalization care is associated with a decrease of 0.11 units units in surgical care
related quality. This is close to the previous nding of 0.14. As before,Poor satisfaction (95%
CI [-0.34, -0.09]) and expected service (95% CI [0.063, 0.286]) were also signicant with signs
that were not in line with the research hypothesis. This shows that the test results are not affected
by the presence of inuential outliers and the research hypothesis is supported in one satisfaction
132
Table 36
RQ4 : Robust Regression Results for Quality Component 3
DV: Quality component 1
Robust SE
P > |t|
[95%
CI]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
-0.215
0.175
-0.0165
0.102
-0.112
0.036
0.0632
0.0568
0.0656
0.0664
0.0550
0.0604
-3.40
3.09
-0.25
1.54
-2.03
0.61
0.001
0.002
0.801
0.123
0.042
0.545
-0.34
0.064
-0.145
-.0279
-0.22
-0.082
-0.09
0.286
0.112
0.232
-0.004
0.155
component (Post hospitalization care) but not in two components (1, and 2). The robust
regression results are given in Table 36
Regression results for quality component 4 (smoking cessation related) The
regression results are given in Table 37. R2 has a very low value of 0.012 showing that the
Table 37
RQ4 - Regression Results for Quality Component 4 by Satisfaction Components
Robust SE.
P > |t|
[95% Conf.
Interval]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
0.10
-0.02
-0.07
0.06
0.07
-0.06
0.07
0.07
0.08
0.07
0.07
0.07
1.34
-0.33
-0.89
0.85
0.99
-0.86
0.18
0.74
0.38
0.40
0.32
0.39
-0.05
-0.16
-0.24
-0.08
-0.07
-0.21
0.24
0.12
0.09
0.20
0.21
0.08
Summary of regression results: Model degrees of freedom = 5, Residual degrees of freedom = 607, F (5,607) = 1.985,
Prob. > F = 0.0791
R2 = 0.012, Adjusted R2 = 0.00383, Predicted R2 = 0,
RMSE = 0.999, Model sum of squares = 7.338, Residual sum of squares = 605.8
Effect size (ES:
f 2 = 0.01; Cohen (1988) considers f 2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 7.45, Power = 0.430 (From Table 9.3.2 of Cohen (1988) for a=0.05)
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 19.38, df = 20, p = 0.4975,
Shapiro-Wilk test
for normality of residuals: W = 0.783, Z = 10.86, Prob > Z = 0.00,
Multicollinearity test: Mean VIF = 1.83,
Ramsey RESET test using powers of tted values: F(3,604) = 1.75, Prob > F = 0.1557
regression only explains about 1.2% of the response variable. This was expected because
hospitals have to maintain a minimum level of quality; otherwise, face shut down by accreditation
and government agencies. The adjusted R2 takes into account the number of regression
parameters, and is lower = 0.00383. The RMSE has a high value of 0.999, showing that the model
133
t is low. Predicted R2 is 0 and within 0.2 of R2 . The overall F statistic (F(5,607) = 1.985) is not
signicant at 5% level (Prob > F = 0.0791). On the basis of two-sided tests of signicance, none
of the coefcients are signicant, showing that the satisfaction components do not signicantly
affect the smoking cessation related quality. ES was below small and the power was high
(0.999). In summary, the robust regression conrmed the OLS regression result that the research
hypothesis is not supported. The smoking cessation related quality component is not
signicantly associated with any of the ve satisfaction components.
Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (3) are reported below Table 37. Shapiro-Wilk normality test showed that the
residuals are not normally distributed (p = 0.00), but this is not critical because of the large
sample size (N > 600). The kernel density plot (given in Figure 22), shows considerable
deviation from normal. The Whites test failed to reject homoscedasticity with p = 0.4975.
Figure 22. RQ4: Kernel Density Plot of Residuals for Quality Component 4
134
Table 38
RQ4 : Robust Regression Results for Quality Component 4
DV: Quality component 1
Robust SE
P > |t|
[95%
CI]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
0.054
-.0164
-0.049
0.060
0.033
0.224
0.0430
0.0387
0.0447
0.0452
0.0374
0.0411
1.26
-0.42
-1.10
1.33
0.89
5.44
0.207
0.671
0.274
0.183
0.374
0.000
-0.030
-0.092
-0.137
-0.029
-.0402
0.1428
0.139
0.059
0.038
0.149
0.107
0.3043
Multicollinearity was not a problem, as shown by the low variable ination factor of 1.83. This
was surprising because the independent variables (satisfaction component) were correlated, as
shown earlier . Ramsey reset test using powers of tted values for model specication errors was
not signicant (Prob > F = 0.37). In summary, the regression diagnostics support OLS regression
if the non-normality of the residuals is excused on the basis of sample size (N > 600).
Sensitivity analysis.
Predicted R2 has 0 value and is within 0.2 of R2 , shows cross-validation. New samples are
likely to give similar test results. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in RQ2 was used and this identied 27 inuential outliers. Robust regression, gave an
F( 5, 607) = 1.60 which was not signicant at 5% level (P > |t| = 0.1588). None of the
coefcient was found to be signicant. The results are not affected by the presence of inuential
outliers. In summary, robust regression conrmed the OLS regression test result that the research
hypothesis is not supported and that the smoking cessation related quality component is not
signicantly associated with any of the ve satisfaction components. The robust regression results
are given in Table 38.
Regression results for quality component 5 (prevention related) The regression
results are given in Table 39. R2 has a very low value of 0.0234 showing that the regression only
explains about 2.3% of the variation in the response variable. This was expected because hospitals
have to maintain a minimum level of quality; otherwise, face shut down by accreditation and
135
Table 39
RQ4 - Regression Results for Quality Component 5 by Satisfaction Components
Robust SE.
P > |t|
[95% Conf.
Interval]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
0.21
-0.16
-0.02
-0.06
0.00
0.04
0.07
0.06
0.07
0.07
0.06
0.07
2.92
-2.48
-0.28
-0.81
0.06
0.63
0.00
0.01
0.78
0.42
0.95
0.53
0.07
-0.28
-0.16
-0.21
-0.12
-0.09
0.35
-0.03
0.12
0.09
0.13
0.18
Summary of regression results: Model degrees of freedom = 5, Residual degrees of freedom = 607, F (5,607) = 3.334,
RMSE = 0.994, Model sum of
Prob > F = 0.0056
R2 = 0.0234, Adjusted R2 = 0.0153, Predicted R2 = 0.0045,
squares = 14.37, Residual sum of squares = 600.2,
Effect size:
f 2 = 0.02; Cohen (1988) considers R2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 14.69, Power = 0.770 (From Table 9.3.2 of Cohen (1988) for a=0.05)
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 16.19, df = 20, p = 0.7045,
Shapiro-Wilk test
for normality of residuals: W = 0.948, Z = 7.41, Prob > Z = 0.00,
Multicollinearity test: Mean VIF = 1.83,
Ramsey RESET test using powers of tted values: F(3,604) = 0.15, Prob > F = 0.9328
government agencies. The adjusted R2 takes into account the number of regression parameters,
and is lower and = 0.0153. The RMSE has a high value (0.994), showing that the model t is low.
Predicted R2 is 0.0045 and is within 0.2 of R2 . The overall F statistic (F(5,607) = 3.334) is
signicant at 5% level (Prob > F = 0.0056). On the basis of two-sided tests of signicance, two of
the coefcients (satisfaction components 1 and 2)are signicant. Poor satisfaction component
(95% CI [0.069, 0.35]) has a positive as expected. The expected service satisfaction
component (95% CI [-0.28, -0.03]) has a negative as was expected. A unit increase in Poor
satisfaction component is associated with an increase of 0.21 units, while a unit increase in
expected service component is associated with a decrease of 0.16 units in prevention related
quality. ES was below small and the power was high (0.999). In summary, the research hypothesis
is supported in two satisfaction components (1 and 2), but not in the other three.
Regression diagnostic tests.
The results of a battery of diagnostic tests for evaluating the adequacy of the estimated
regression equation (3) are reported below Table 39. Shapiro-Wilk normality test showed that the
residuals are not normally distributed (p = 0.00), but this is not critical because of the large
136
sample size (> 600). The kernel density plot (given in Figure 23), also does not show
considerable deviation from normal. The Whites test failed to reject homoscedasticity with p =
Figure 23. RQ4: Kernel Density Plot of Residuals for Quality Component 5
137
Table 40
RQ4 : Robust Regression Results for Quality Component 5
DV: Quality component 1
Robust SE
P > |t|
[95%
CI]
Independent variables
Patient satisfaction component 1
Patient satisfaction component 2
Patient satisfaction component 3
Patient satisfaction component 4
Patient satisfaction component 5
Constant term
0.208
-0.153
-0.007
-0.053
0.059
0.155
0.0657
0.0590
0.0682
0.0690
0.0572
0.0628
3.16
-2.60
-0.10
-0.77
1.03
2.47
0.002
0.010
0.924
0.439
0.303
0.014
0.0788
-0.269
-0.141
-0.189
-0.053
0.0316
0.337
-0.037
0.1274
0.0821
0.1713
0.2784
Table 41
RQ4 - OLS Regression Coefcients on Satisfaction Components by Quality Components
Patient satisfaction components
Poor satisfaction
Expected level
Overall
Cleanliness
Post hospitalization care
Effect size f 2
Power
Quality components
Heart attack/failure
ns
0.2
-0.4
0.3
-0.17
0.1
0.995
Pneumonia
ns
ns
ns
-0.26
ns
0.03
0.93
Surgical care
-0.21
0.21
ns
ns
-0.14
0.06
0.995
Smoking cessation
ns
ns
ns
ns
ns
0.01
0.43
Prevention
0.21
-0.16
ns
ns
ns
0.02
0.77
residual < -4.0. Robust regression, gave an F( 5, 607) = 4.90 which was signicant (P > |t| =
0.0002). As before, Poor satisfaction component (95% CI [0.07, 0.34])and expected service
(95% CI [-0.27,-0.04]) components had signicant coefcients with positive and negative signs
as expected. This shows that the test results are not much affected by the presence of the lone
inuential outlier. In summary, robust regression conrmed the OLS regression results. The
research hypothesis was supported in two satisfaction components (1 and 2) but not in the other
three. The robust regression results are given in Table 40
Summary of Regression results for RQ-4. The regression coefcients for RQ-4 are
summarized in Table 41. The coefcients that are not signicant are indicated by ns in this
table.
138
Summary
In this chapter, a brief description of the archival data retrieved for analysis was given and
the statistical tests used for the research questions 1 though 4 were discussed. This was followed
by the subsections on each research hypothesis including a discussion of the test results for with a
brief description of the descriptives for the data. To establish statistical conclusion validity, the
assumptions of each statistical test are also tested. Results for the rst research question revealed
ve principal components for patient satisfaction:
1. Component 1 Poor satisfaction (accounting for 59.3% of variance).
2. Component 2 Required level of performance (accounting for 11.3% of variance).
3. Component 3 Overall rating (accounting for 6% of variance).
4. Component 4 Cleanliness (accounting for 4.3% of variance).
5. Component 5 Post hospitalization care (accounting for 3.5% of variance).
Previously, studies have been using the overall satisfaction as the primary indicator of
patient satisfaction. This study, however found that this approach is questionable because this
accounts for only 6% of the variance. The loadings of each HCAHPS question on these ve
components were also obtained and discussed, highlighting on the areas, a hospital needs to focus
to improve patient satisfaction. This would help hospitals in improving performance.
Results for the second research hypothesis showed that church hospitals have signicantly
better patient satisfaction than state government and proprietary hospitals with a small to medium
effect size. Proprietary hospitals are run by corporations for prot and yet are signicantly below
the church hospitals. This is likely to be the result of church hospitals displaying more
friendship, compassion, joy in serving and equanimity, that were emphasized by Charaka, the
ancient Vedic physician as essential for medical practitioners.
Results of the third research question showed ve principal components for quality:
139
1. Quality component 1 heart attack, failure related accounts for 15.8 % of the variance.
2. Quality component 2 pneumonia related accounts for 10.4 % of the variance.
3. Quality component 3 surgical care related accounts for 10.4 % of the variance.
4. Quality component 4 smoking cessation related accounts for 8.1% of the variance.
5. Quality component 5 prevention related accounts for 8.9% of the variance.
Under each quality component, differences between hospital owner groups were explored and the
results show that church hospitals come out mostly better, again conrming the need for
friendship, compassion, joy in serving and equanimity in healthcare.
Results for the fourth research question show that in two of the ve satisfaction
components, the research hypothesis that hospitals may be sacricing quality for patient
satisfaction was supported by statistical evidence. But, the effect size was small, and On the
whole, the evidence is not clear cut.
The fth research question will be covered in the next chapter.
140
Chapter 5
Chapter overview
The fth research question was to test the relationship of outcome variables with patient
satisfaction, quality and ownership groups. The question was subdivided into three sub-questions.
This chapter presents the results of the data analysis for this research question: What is the
relationship between outcome measures and patient satisfaction, ownership type and process of
care quality measures?
This question is divided into six sub-questions for each type of outcome: heart attack,
heart failure and pneumonia. The six sub-questions are:
5.1 Relationship between mortality rates and ownership type.
5.2 Relationship between readmission rates and ownership type.
5.3 Relationship between mortality rates and patient satisfaction.
5.4 Relationship between readmission rates and patient satisfaction.
5.5 Relationship between mortality rates and process of care quality
5.6 Relationship between readmission rates and process of care quality
141
142
Description
Count
sd
kurtosis
skewness
HAMORT1
30day-mortality-rate-heart attack
2902
16.5001
1.8041
3.6005
0.3012
HFMORT1
30day-mortality-rate-heart failure
3891
11.1680
1.5544
3.6686
0.4421
PNMORT1
30day-mortality-rate-pneumonia
4074
11.6406
1.9110
3.7093
0.5952
HAMORT2
30day-readmission-rate-heart attack
2461
19.931
1.3257
3.7116
0.2734
HFMORT2
30day-readmission-rate-heart failure
3904
24.5335
2.0573
3.8629
0.3853
PNMORT2
30day-readmission-rate-pneumonia
4041
18.1972
1.6936
4.1136
0.6207
The assumptions of the ANOVA analysis based on OLS regression are as follows:
1. Independence of observations: Independence assumption is supported because the outcome
variables are truly independent of each other. The outcome variables can be reasonably
considered as independent because people do not die just because someone else with the
same health condition dies.
2. Normality of observations: Testing was done using Shapiro-Francia test for normality (in
preference to Shapiro-Wilk test because N > 2000) and the results are shown in Table 43.
Table 43
Results of Shapiro-Francia W Test for Normality Outcome Variables Variable
Description
Obs.
Prob.>z
HAMORT1
3.885
2902
.00005
HFMORT1
3.940
3891
.00004
PNMORT1
4.213
4074
.00001
HAMORT2
4.346
2461
.00001
HFMORT2
3.930
3904
.00004
PNMORT2
4.272
4041
.00001
143
From this table, it is seen that the data does not follow normal distribution. Q-Q plots were
also created to check the extent of nonnormality and the charts are given in Figure 24. The
large sample size (N 2,000) and near-normality exhibited by the Q-Q plots shows that
normality of the data may be reasonably assumed for statistical testing. After regression,
the residuals will again be tested for normality, as a double check. Technically, this latter
test is equivalent to testing the normality of the raw data for ANOVA via OLS regression.
144
3. Homogeneity of variance: Heteroscedasticity was tested with Levenes test which is robust
to nonnormality. The variant proposed by Brown and Forsythe uses the more robust
estimator (with median rather than the mean) (Baum, 2006) and was adopted here. The
results are given in the Table 44. The results failed to reject homoscedasticity except in one
variable (HAMORT2 readmission rate in heart attack) in owner groups. The other ve
145
Table 44
Levenes Homoscedasticity Test Results for Outcome Variables Variable
Description
W50
df
HAMORT1
0.734
(7, 2894)
0.64
HFMORT1
1.716
(7, 3883)
0.100
PNMORT1
30day-mortality-rate in pneumonia
0.406
(7, 4066)
0.899
HAMORT2
2.790
(7, 2453)
0.007
HFMORT2
1.559
(7, 3896)
0.143
PNMORT2
30day-readmission-rate in pneumonia
1.671
(7, 4033)
0.111
Figure 25. Marginal Means Plot for Heart Attack Mortality Rate
From these plots, the following conclusions may be drawn:
146
Figure 26. Marginal Means Plot for Heart Failure Mortality Rate
147
Figure 28. Marginal Means Plot for Heart Attack Readmission Rate
Figure 29. Marginal Means Plot for Heart Failure Readmission Rate
Church hospitals have the lowest readmission rate for heart attack.
Church hospitals are at the lowest rate in readmission rates for heart failure.
It is again seen that church hospitals have the lowest readmission rate for pneumonia.
148
149
Table 45
RQ5 - Regression results for 30-day risk adjusted mortality rate for heart attack by owner groups
DV: Mortality (heart attack)
IV: Own code
Government - federal
Government - hospital area/authority
Government - local
Government - state
Proprietary
Voluntary nonprot-church (base group)
Voluntary nonprot-other
Voluntary nonprot-private
Constant term
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
20
240
189
40
496
410
495
1012
0.17
0.49
0.64
-0.29
0.41
0.09
0.004
16.33
0.32
0.15
0.16
0.30
0.12
0.12
0.11
0.09
0.52
3.27
4.04
-0.97
3.36
0.77
0.04
179.97
0.61
0.00
0.00
0.33
0.00
0.44
0.97
0.00
-0.47
0.20
0.33
-0.88
0.17
-0.14
-0.20
16.15
0.80
0.79
0.95
0.30
0.65
0.33
0.21
16.51
0.09
0.27
0.35
0.16
0.22
0.05
0.00
0.54
0.61
0.64
0.56
0.59
0.52
0.50
0.27), local government (95% CI is [0.33, 0.95], ES = 0.35) and proprietary (95% CI is [0.17,
0.65], ES = 0.22) in heart attack. Cohens effect size was calculated as described in Chapter 4. In
summary, the research hypothesis is supported in three of the seven owner groups with effect
sizes between low and medium.
Regression diagnostic tests.
adequacy of the estimated regression equation (4) are reported below Table 45. Shapiro-Francia
normality test showed that the residuals are not normally distributed, but this is not critical
because of the sample size (N > 2900). The normality robust Levenes test failed to reject
homoscedasticity (Pr. > F = 0.643). Multicollinearity is not a problem, as shown by the low
Variance Ination Factor (VIF) of 1.55. Multicollinearity was not expected to be a problem
because the independent variables (ownership group) are denitely not correlated. Ramsey reset
test for model specication errors was not run because all explanatory variables are dummy
variables. In summary, the regression diagnostics support the adequacy of the regression.
Sensitivity analysis. Predicted R2 is 0, and within 0.2 of R2 , showing cross-validation.
New samples are likely to give similar results. R2 being small, we need to rule out the possibility
150
of a few inuential outliers affecting the model predictions. A procedure to identify the inuential
outliers, used earlier in Chapter 4, was used and this identied two inuential outliers: a hospital
(owned by local government group) with a mortality rate of 24.9 and a residual of 4.43 and
another owned by voluntary nonprot - private with a mortality rate of 24.5 and a residual of 4.6.
Robust regression, gave an overall F statistic (F(7,2894) = 5.33) which was signicant; (P > |t| =
0.0000). Again, church group hospitals had signicant lower mortality rates from hospitals owned
by district or area, local government, and proprietary groups. The signs of the coefcients were
the same as before. This shows that the test results are not seriously affected by the presence of
the two inuential outliers. In summary, the research hypothesis is supported in the same three
groups found with OLS regression. The robust regression results are given in Table 46.
Table 46
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality
Rate for Heart Attack by Owner Groups
Robust Regression Results
SE
P > |t|
[95%
CI]
Owner Group
Government - federal
0.279
0.41
0.68
0.497
-0.525
1.082
0.391
0.145
2.69
0.007
0.106
0.677
Government - local
0.532
0.157
3.38
0.001
0.223
0.840
Government - state
-0.38
0.296
-1.28
0.199
-0.962
0.200
Proprietary
0.438
0.119
3.67
0.000
0.204
0.672
Voluntary non-pr-other
0.113
0.119
0.95
0.344
-0.121
0.347
Voluntary non-pr-private
0.024
0.105
0.23
0.815
-0.181
0.229
Constant term
16.28
0.088
184.25
0.000
16.109
16.455
151
Table 47
RQ5- Regression results for 30-day risk adjusted readmission rate for heart attack
by owner groups
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
Government - federal
14
0.71
0.32
2.24
0.03
0.09
1.34
0.52
0.70
Government - hospital
192
0.09
0.11
0.81
0.42
-0.13
0.31
0.07
0.53
Government - local
135
0.32
0.14
2.34
0.02
0.05
0.58
0.23
0.59
Government - state
34
0.54
0.28
1.91
0.06
-0.02
1.10
0.38
0.65
434
0.02
0.09
0.17
0.87
-0.16
0.19
0.01
0.50
370
non-pr-
415
0.02
0.10
0.15
0.88
-0.18
0.21
0.01
0.50
non-pr-
867
0.11
0.09
1.25
0.21
-0.06
0.28
0.08
0.53
19.85
0.07
274.12
0.00
19.71
19.99
DV:
Readmission
(heart attack)
IV: Own code
area/authority
Proprietary
Voluntary
non-pr-
R2
Regression diagnostics:
0.0049,
Levenes robust test for heteroscedasticity using median: W50 = 2.79, df (7, 2453), Pr > F =
Shapiro-Francia test for normality of residuals: W = 0.992, Z = 4.334, Prob. > Z = 0.00001,
very low value of 0.006 showing that the regression only explains only 0.6% of the variation in
the response variable. This was expected because of the strong inuence of omitted explanatory
variables such as patients risk factors and nature of ailment that play a large part on readmission
rate, thus making a major contribution to R2 . The adjusted R2 is lower and = 0.003. The RMSE
has a high value of 1.32, showing that the model t is low. Predicted R2 is 0, and within 0.2 of R2 ,
152
showing cross validation. New samples are likely to yield similar results. The overall F statistic
(F(7,4066) = 2.162) is signicant at p of 0.0000. On the basis of two-sided tests of signicance,
two of the coefcients are signicant, showing that church group hospitals have signicantly
less readmission rate than federal government (95% CI is [0.09, 1.34], ES = 0.52), local
government (95% CI is [0.05, 0.58], ES = 0.23). Cohens effect size was calculated as described
in Chapter 4. In summary, the research hypothesis is supported in two of the seven owner groups
with effect sizes of 0.52 (between medium and high) and 0.23 (between low and medium).
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated regression equation (5) are reported below Table 47. Shapiro-Francia
normality test showed that the residuals are not normally distributed, but this is not critical
because of the large sample size (N > 2,450). The normality robust Levenes test rejected
homoscedasticity (Pr > F = 0.0049); necessitating regression with heteroscedasticity robust
errors. Multicollinearity is not a problem, as shown by the low VIF of 1.55. Multicollinearity was
not expected to be a problem because the independent variables (ownership group) are denitely
not correlated. Ramsey reset test for model specication errors was not run because all
explanatory variables are dummy variables. In summary, the regression diagnostics support the
use of OLS regression.
Sensitivity analysis. Predicted R2 is 0 and is within 0.2 of R2 , showing cross-validation.
New samples are likely to yield similar results. R2 being small, we need to rule out the possibility
of a few inuential outliers affecting the model predictions. A procedure to identify the inuential
outliers, used earlier in Chapter 4, was used and this identied no inuential outliers. Robust
regression, gave a reduced F(7,2453) = 1.56 which was signicant at (P > |t| = 0.0000). Again,
church group hospitals had signicant lower mortality rates from hospitals owned by federal
government (95% CI is [0.09, 1.33]) and local government (95% CI is [0.05, 0.58]). The signs of
the coefcients were the same as before. This shows that the test results are not seriously
affected by robust regression because inuential outliers were not present. In summary, the
153
research hypothesis is supported in two groups and no inuential outliers were found by robust
regression. The robust regression results are given in Table 48
Table 48
RQ5 - Robust Regression Results for 30-day Risk Adjusted
Readmission Rate for Heart Attack by Owner Groups
SE
P > |t|
[95%
CI]
0.70
0.352
1.99
0.047
0.0095
1.391
0.156
0.115
1.36
0.175
-0.0694
0.382
Government - local
0.258
0.130
1.99
0.047
0.00341
0.513
Government - state
0.357
0.232
1.54
0.124
-0.0975
0.811
Proprietary
0.065
0.092
0.71
0.478
-0.1145
0.244
Voluntary non-pr-other
0.038
0.092
0.41
0.684
-0.144
0.2189
Voluntary non-pr-private
0.137
0.080
1.70
0.089
-0.0208
0.294
Constant term
19.79
0.067
294.34
0.000
19.658
19.922
154
Table 49
RQ5- Regression results for 30-day risk adjusted mortality rate for heart failure by
owner groups
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
Government - federal
42
0.70
0.24
2.94
0.00
0.23
1.16
0.45
0.67
Government - hospital
415
0.32
0.10
3.09
0.00
0.12
0.52
0.21
0.58
Government - local
339
0.41
0.10
3.95
0.00
0.21
0.62
0.28
0.61
Government - state
54
0.11
0.22
0.51
0.61
-0.32
0.54
0.07
0.53
599
0.02
0.10
0.22
0.83
-0.17
0.21
0.01
0.50
476
non-pr-
650
0.14
0.10
1.41
0.16
-0.05
0.32
0.08
0.53
non-pr-
1316
0.09
0.08
1.08
0.28
-0.07
0.25
0.06
0.52
11.03
0.07
155.03
0.00
10.89
11.17
area/authority
Proprietary
Voluntary
non-pr-
R2
Regression diagnostics:
0.643,
Levenes robust test for heteroscedasticity using median: W50 = 0.734, df (7, 2894), Pr > F =
Shapiro-Francia test for normality of residuals: W = 0.988, Z = 4.000, Prob > Z = 0.00003,
R2 has a very low value of 0.00767 showing that the regression only explains only 0.8% of the
variation in the response variable. This was expected because of the strong inuence of omitted
explanatory variables such as patients risk factors that play a large part on mortality rate, and
making a major contribution to R2 . The adjusted R2 takes into account the number of regression
parameters, and is lower = 0.006. The RMSE has a high value of 1.55, showing that the model t
is low. Predicted R2 is 0.004 and within 0.2 of R2 , showing cross validation. New samples are
likely to yield similar results. The overall F statistic (F(7,3883) = 4.765) is signicant at p of
155
0.0000. On the basis of two-sided tests of signicance, three regression coefcients are signicant
and positive, showing that church group hospitals have signicantly less mortality than federal
government (95% CI is [0.23, 1.16], ES = 0.45), government - hospital area/authority (95% CI is
[0.12 , 0.52], ES = 0.21), and local government (95% CI is [0.21, 0.62], ES = 0.28) in heart
failure. Cohens effect size was calculated as described in Chapter 4. In summary, the research
hypothesis is supported in three of the seven owner groups with effect sizes between low and
medium.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 49. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because of the large sample
size (N > 3,800). The normality robust Levenes test failed to reject homoscedasticity
(Pr > F = 0.643). Multicollinearity is not a problem, as shown by the low VIF of 1.68.
Multicollinearity was not expected to be a problem because the independent variables (ownership
group) are denitely not correlated. Ramsey reset test for model specication errors was not run
because all explanatory variables are dummy variables. In summary, the regression diagnostics
validated the OLS regression.
Sensitivity analysis. Predicted R2 is 0.004, and within 0.2 of R2 , showing
cross-validation for new cases. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in Chapter 4 was used and this identied only one inuential outlier, a hospital
(owned by district/authority). Robust regression, gave an overall F statistic of F(7,3883) = 4.02
which was signicant at (P > |t| = 0.0002). Again, church group hospitals had signicant lower
mortality rates than hospitals owned by federal, district or area, and local government groups. The
signs of the coefcients were the same as before. This shows that the test results are not
156
seriously affected by the presence of an inuential outlier. The robust regression results are given
in Table 50.
Table 50
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality
Rate for Heart Failure by Owner Groups
SE
P > |t|
[95%
CI]
Government - federal
0.643
0.247
2.61
0.009
0.159
1.126
0.314
0.103
3.05
0.002
0.112
0.516
Government - local
0.402
0.109
3.69
0.000
0.188
0.615
Government - state
0.103
0.220
0.47
0.639
-0.328
0.535
Proprietary
0.0195
0.094
0.21
0.836
-0.165
0.204
Voluntary non-pr-other
0.103
0.092
1.12
0.264
-0.078
0.284
Voluntary non-pr-private
0.117
0.082
1.43
0.154
-0.044
0.278
Constant term
10.963
0.070
156.07
0.000
10.825
11.10
157
Table 51
RQ5- Regression results for 30-day risk adjusted readmission rate for heart failure
by owner groups
Count
Robust SE
P>t
[95%
CI]
Cohens d
CLES (1)
Government - federal
47
0.47
0.29
1.60
0.11
-0.11
1.05
0.23
0.59
Government - hospital
428
0.60
0.13
4.44
0.00
0.33
0.86
0.30
0.62
Government - local
348
0.58
0.14
4.05
0.00
0.30
0.86
0.28
0.61
Government - state
55
0.54
0.28
1.94
0.05
-0.01
1.08
0.26
0.60
603
0.70
0.13
5.38
0.00
0.44
0.96
0.33
0.63
471
non-pr-
648
0.33
0.13
2.56
0.01
0.08
0.58
0.15
0.56
non-pr-
1304
0.31
0.11
2.75
0.01
0.09
0.52
0.15
0.56
24.14
0.10
250.97
0.00
23.95
24.33
area/authority
Proprietary
Voluntary
non-pr-
R2
RMSE = 2.048, Model sum of squares = 173.8, Residual sum of squares = 16346
Regression diagnostics:
0.143,
Levenes robust test for heteroscedasticity using median: W50 = 1.56, df (7, 3896), Pr.
F=
Shapiro-Francia test for normality of residuals: W = 0.988, Z = 3.968, Prob. > Z = 0.00004,
low value of 0.0105 showing that the regression only explains only 1.1% of the variation in the
response variable. This was expected because of the strong inuence of omitted explanatory
variables such as patients risk factors and nature of ailment that play a large part on readmission
rate, thus making a major contribution to R2 . The adjusted R2 is lower = 0.00875. RMSE has a
high value of 2.05, showing that the model t is low. Predicted R2 is 0.007 and within 0.2 of R2 .
158
The overall F statistic (F(7,3896) = 5.76) is signicant at p of 0.0000. On the basis of two-sided
tests of signicance, ve regression coefcients are signicant and positive, showing that church
group hospitals have signicantly less mortality than government - hospital area/authority (95%
CI is [0.33 , 0.86], ES = 0.30), local government (95% CI is [0.30, 0.86], ES = 0.28), proprietary
(95% CI is [0.44, 0.96], ES = 0.33), nonprot-other (95% CI is [0.08, 0.58], ES = 0.15), and
nonprot-private (95% CI is [0.09, 0.52], ES = 0.15) in heart failure. Cohens effect size was
calculated as described in Chapter 4. In summary, the research hypothesis is supported in ve of
seven owner groups with effect sizes between low and medium.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 51. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because of the large sample
size (N > 3,900). The normality robust Levenes test failed to reject homoscedasticity
(Pr > F = 0.143). Multicollinearity is not a problem, as shown by the low VIF of 1.68.
Multicollinearity was not expected to be a problem because the independent variables (ownership
group) are denitely not correlated. Ramsey reset test for model specication errors was not run
because all explanatory variables are dummy variables. In summary, the regression diagnostics
validated the OLS regression.
Sensitivity analysis. Predicted R2 is 0.007, and within 0.2 of R2 , showing
cross-validation for new cases. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in Chapter 4, was used and this identied two inuential outliers: a hospital (owned
by proprietary group) with a low readmission rate = 15.9 and a standardized residual of -4.4 and
another hospital owned by local government group with a very high rate of 34.4 and a
standardized residual of 4.7. Robust regression, gave an overall F statistic (F(7,3896) = 6.35)
which was signicant at (P > |t| = 0.0000). As with OLS regression, church group hospitals
turned out to have signicant lower mortality rates than hospitals owned by district or area, local
159
government, proprietary, nonprot-other and nonprot private groups. The signs of the
coefcients were the same as before. This shows that the test results are not seriously affected by
the presence of the two inuential outliers. In summary, the research hypothesis is supported in
ve out of seven owner groups and the test results are not affected by inuential outliers. The
robust regression results are given in Table 52.
Table 52
RQ5 - Robust Regression Results for 30-day Risk Adjusted
Readmission Rate for Heart Failure by Owner Groups
SE
P > |t|
[95%
CI]
Government - federal
0.346
0.305
1.14
0.256
-0.25
0.944
0.636
0.133
4.78
0.000
0.375
0.897
Government - local
0.508
0.141
3.61
0.000
0.232
0.785
Government - state
0.517
0.284
1.82
0.069
-0.04
1.074
Proprietary
0.698
0.123
5.70
0.000
0.458
0.938
0.284
0.121
2.35
0.019
0.047
0.521
0.308
0.107
2.88
0.004
0.098
0.518
Constant term
24.07
0.092
262.09
0.000
23.89
24.246
160
Table 53
RQ5- Regression results for 30-day risk adjusted mortality rate for pneumonia by
owner groups
DV: Mortality (heart
failure)
IV: Own code
Government - federal
Government - hospital
area/authority
Government - local
Government - state
Proprietary
Voluntary
non-prchurch (base group)
Voluntary
non-prother
Voluntary
non-prprivate
Constant term
Count
Robust SE
P > |t|
[95%
CI]
Cohens d
CLES (1)
55
458
0.28
0.68
0.24
0.13
1.15
5.41
0.25
0.00
-0.20
0.43
0.76
0.92
0.15
0.35
0.56
0.64
382
55
604
483
0.76
0.39
0.34
0.13
0.29
0.12
5.91
1.36
2.92
0.00
0.17
0.00
0.51
-0.17
0.11
1.02
0.95
0.57
0.41
0.21
0.18
0.66
0.58
0.57
682
0.30
0.11
2.65
0.01
0.08
0.52
0.16
0.56
1355
0.20
0.10
1.98
0.05
0.00
0.39
0.11
0.54
11.32
0.09
132.46
0.00
11.15
11.49
signicantly less mortality than government - hospital area/authority (95% CI is [0.43 , 0.92], ES
= 0.35), local government (95% CI is [0.51, 1.02], ES = 0.41), proprietary (95% CI is [0.11,
0.57], ES = 0.18), nonprot-other (95% CI is [0.08, 0.52], ES = 0.16), and nonprot-private (95%
CI is [0.00, 0.39], ES = 0.11) in pneumonia. Cohens effect size was calculated as described in
Chapter 4. In summary, the research hypothesis is supported in ve of the seven owner groups,
with effect sizes between low and medium.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 53. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because of the large sample
size (N > 4,000). The normality robust Levenes test failed to reject homoscedasticity
(Pr > F = 0.9). Multicollinearity is not a problem, as shown by the low VIF of 1.68.
Multicollinearity was not expected to be a problem because the independent variables (ownership
161
group) are denitely not correlated. Ramsey reset test for model specication errors was not run
because all explanatory variables are dummy variables. In summary, the regression diagnostics
validated the OLS regression.
Sensitivity analysis. Predicted R2 is 0.007 and within 0.2 of R2 , showing
cross-validation for new cases. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in Chapter 4, was used and this identied three inuential outliers: a hospital (owned
by nonprot - private group) with a high mortality rate = 20.9 and a residual of 4.9 and two
hospitals owned by proprietary group with high mortality rates of 21.6 and 20.1 with residuals of
5.23 and 4.44 respectively. Robust regression, gave an overall F statistic (F(7,4033) = 7.41)
which was signicant at (P > |t| = 0.0000). Similar to OLS regression, church group hospitals
had signicant lower mortality rates government-hospital area/authority, local government,
proprietary, nonprot-other, and nonprot-private groups in pneumonia. The signs of the
coefcients were the same as before. This shows that the test results are not seriously affected by
the presence of the three inuential outliers. In summary, the robust regression showed that the
research hypothesis is supported in ve out of seven owner groups as in OLS regression. The
162
SE
P > |t|
[95%
CI]
Government - federal
0.368
0.269
1.37
0.171
-0.159
0.895
0.654
0.123
5.31
0.000
0.413
0.896
Government - local
0.734
0.129
5.67
0.000
0.480
0.988
Government - state
0.437
0.269
1.62
0.105
-0.091
0.964
Proprietary
0.318
0.115
2.76
0.006
0.0925
0.545
Voluntary non-pr-other
0.293
0.112
2.61
0.009
0.0724
0.513
Voluntary non-pr-private
0.206
0.100
2.06
0.040
0.0094
0.402
Constant term
11.21
0.086
130.44
0.000
11.04
11.38
163
Table 55
RQ5- Regression results for 30-day risk adjusted readmission rate for pneumonia by
owner groups
Count
Robust SE.
P> |t|
[95% Conf.
Interval]
Cohens d
CLES(1)
Government - federal
56
-0.118
0.182
-0.65
0.513
-0.475
0.237
0.07
0.53
Government - hospital
459
0.04
0.106
0.41
0.683
-0.165
0.252
0.03
0.51
Government - local
382
0.055
0.117
0.47
0.635
-0.173
0.284
0.03
0.51
Government - state
55
0.538
0.241
2.24
0.025
0.066
1.010
0.32
0.63
610
0.260
0.105
2.48
0.013
0.054
0.466
0.15
0.56
476
non-pr-
670
0.120
0.103
1.16
0.244
-0.082
0.323
0.07
0.53
non-pr-
1333
0.189
0.0898
2.10
0.036
0.0128
0.365
0.11
0.54
18.06
0.077
233.84
0.000
17.908
18.211
DV:
Readmission
(pneumonia)
IV: Own code
area/authority
Proprietary
Voluntary
non-pr-
R2
Model degrees of freedom = 7, Residual degrees of freedom = 4033, F (7,4033) = 2.208, Prob. >
Regression diagnostics:
Levenes robust test for heteroscedasticity using median: W50 = 1.67, df (7, 4033), Pr > F = 0.072,
Shapiro-Francia test for normality of residuals: W = 0.980, Z = 4.27, Prob > Z = 0.00001,
R2 has a very low value of 0.004 showing that the regression only explains only 0.4% of
the variation in the response variable. This was expected because of the strong inuence of
omitted explanatory variables such as patients risk factors and nature of ailment that play a large
part on readmission rates, thus making a major contribution to R2 from the regression. The
adjusted R2 is lower = 0.002. RMSE has a high value of 1.7, showing that the model t is low.
Predicted R2 is 0, and is within 0.2 of R2 . The overall F statistic (F(7,4033) = 2.21) is signicant
at p of 0.031. On the basis of two-sided tests of signicance, three regression coefcients are
signicant and positive, showing that church group hospitals have signicantly less readmission
164
than state government (95% CI is [0.07 , 1.01], ES = 0.03), proprietary (95% CI is [0.05, 0.47],
ES = 0.03), and nonprot-private (95% CI is [0.01, 0.36], ES = 0.11) in pneumonia. Cohens
effect size was calculated as described in Chapter 4 and the values show small effect sizes. In
summary, the research hypothesis is supported in three of the seven owner groups; but the effect
sizes are below Cohens limit of 0.2 for low effect.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 55. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because of the large sample
size (N > 4,000). The normality robust Levenes test failed to reject homoscedasticity
(Pr > F = 0.072). Multicollinearity is not a problem, as shown by the low VIF of 1.68.
Multicollinearity was not expected to be a problem because the independent variables (ownership
group) are denitely not correlated. Ramsey reset test for model specication errors was not run
because all explanatory variables are dummy variables. The regression diagnostics show that OLS
regression is valid.
Sensitivity analysis. Predicted R2 is 0 and within 0.2 of R2 , showing cross-validation
for new cases. R2 being small, we need to rule out the possibility of a few inuential outliers
affecting the model predictions. A procedure to identify the inuential outliers, used earlier in
Chapter 4, was used and this identied three inuential outliers: a hospital owned by nonprot other group, a hospital owned by proprietary group, and another owned by nonprot - private
groups with high readmission rates of 25.9, 27.6, and 26.9 with residuals of 4.6,5.5, and 5.1
respectively. Robust regression, gave an F(7,4033) = 2.07 which was signicant at (P > |t| =
0.0431). Three coefcients are positive and signicant. Church group hospitals had signicant
lower mortality rates than state government, proprietary, and nonprot-private groups in
pneumonia. The signs of the coefcients were the same as before. This shows that the OLS
regression test results are not affected by the presence of the three inuential outliers. In
summary, the robust regression showed that the research hypothesis is supported in the same three
165
of seven owner groups which were identied by OLS regression. The robust regression results are
given in Table 56.
Table 56
RQ5 - Robust Regression Results for 30-day Risk Adjusted
Readmission Rates for Pneumonia by Owner Groups
SE
P > |t|
[95%
CI]
-0.044
0.233
-0.19
0.848
-0.501
0.412
0.104
0.108
0.96
0.335
-0.107
0.315
Government - local
0.047
0.113
0.42
0.677
-0.175
0.269
Government - state
0.549
0.235
2.34
0.019
0.089
1.009
Proprietary
0.236
0.101
2.34
0.019
0.038
0.434
Voluntary non-pr-other
0.119
0.0988
1.21
0.227
-0.074
0.313
Voluntary non-pr-private
0.226
0.088
2.57
0.010
0.0536
0.399
17.951
0.0755
237.61
0.000
17.802
18.1
Constant term
Government - Federal
Government - Hospital
District or Authority
Government - Local
Government - State
Proprietary
Voluntary non-prot Church (Base group)
Voluntary non-prot Other
Voluntary non-prot - Private
Outcome
variables
mortality
rates
readmission
Heart attack
ns
0.49 (0.27)
Mortality
rates
Heart Failure
0.7 (0.45)
0.32 (0.21)
Pneumonia
ns
0.68 (0.35)
Heart attack
0.71 (0.52)
ns
Readmission
rates
Heart Failure
ns
0.6 (0.30)
Pneumonia
ns
0.6 (0.03)
0.64 (0.35)
ns
0.41 (0.22)
0.41 (0.28)
ns
ns
0.76 (0.41)
ns
0.34 (0.18)
0.32 (0.23)
ns
ns
0.58 (0.28)
ns
0.7 (0.33)
0.58 (0.03)
ns
0.7 (0.15)
ns
ns
0.3 (0.16)
ns
0.33 (0.15)
ns
ns
ns
0.2 (0.11)
ns
0.31 (0.15)
0.19 (0.11)
166
167
Robust SE.
P > |t|
[95%
CI]
0.029
0.0363
0.81
0.417
-.0417
.100
Constant term
16.46
.034
479.40
0.000
16.39
16.53
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 2.75, df = 2, p = 0.2526,
residuals: W = 0.994, Z = 3.782, Prob > Z = 0.00008,
explanatory variable,
Ramsey RESET test using powers of tted values: F (3,2787) = 3.43, Prob > F = 0.0165
R2 has very low value of 0.0002, showing that only 0.02% of the variation in heart attack
mortality is explained by satisfaction. This was expected because of omitted explanatory
variables such as patients inherent risk characteristics and quality of care. The adjusted R2 is
lower at 0. The RMSE has a high value of 1.8, showing that the model t is very low. Predicted
R2 is 0, and within 0.2 of R2 . The overall F (F (1,2790) = 0.66) is not signicant (Prob > F =
0.4170), showing that mortality rate for heart attack is not signicantly associated with poor
satisfaction component. For simple regression, the F statistic gives the same value as the
two-sided test for signicance of the regression coefcient. As noted, the coefcient was not
168
signicant. (95% CI was [-0.04,0.1]). ES of 0.0002 was below Cohens limit of 0.02 for small
effect. In summary, the research hypothesis is not supported for mortality rate in heart attack.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 58. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because of large sample size
(N > 4,000). Whites test failed to reject homoscedasticity (Pr > F = 0.2526). A residuals versus
tted values plot (given in Fig. 32) showed that the pattern of data points showed some evidence
of heteroscedasticity at higher values of mortality but not to a great extent around the mean level.
Ramsey reset test for model specication errors was signicant Prob > F = 0.0165,
showing that there could be nonlinearities in the model. The Box-Tidwell test also showed
signicant nonlinearity (Nonlin. dev. = 8.205, P = 0.004). An ACPR plot was created for a visual
check and is given in Fig. 31.
Figure 31. ACPRplot for Heart Attack Mortality Rate - Satisfaction Component 1
Figure 32. RVFplot for Heart Attack Mortality Rate - Satisfaction Component 1
However, this graph showed that the heteroscedasticty was not substantial. Therefore, the
OLS regression result that the coefcient was not signicant and ES was very low, was found
169
acceptable. In summary, regression diagnostics provide support to OLS regression. The possible
nonlinearity in the model was not substantial.
Sensitivity analysis. Predicted R2 is 0 and within 0.2 of R2 , showing cross-validation
for new cases. R2 being small, we need to rule out the possibility of a few inuential outliers
affecting the model predictions. A procedure to identify the inuential outliers, used earlier in
Chapter 4, was used and this identied one inuential outlier: a hospital with high mortality rate
of 24.9, and standardized residual of 4.7. Robust regression, gave an F(1, 2790) = 0.02 which was
not signicant at (P > |t| = 0.8941). This shows that the OLS regression test results were not
affected by the presence of the inuential outlier. In summary, robust regression also failed to
support the research hypothesis for heart attack mortality rate. The robust regression results are
given in Table 59.
Table 59
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate
for Heart Attack with Patient Satisfaction Component 1
30-day risk adjusted mortality rate (Heart
Robust SE.
P > |t|
[95%
CI]
0.005
0.0357
0.13
0.894
-0.065
0.075
16.413
0.0346
474.85
0.000
16.345
16.48
attack)
Patient satisfaction - component 1
Constant term
170
Table 60
RQ5- Regression results for 30-day risk adjusted readmission rate for
heart attack by satisfaction component 1
Robust SE.
P>t
[95%
CI]
0.265
0.0288
9.20
0.000
0.208
.321
Constant term
19.86
.0267
742.53
0.000
19.81
19.91
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 15.77, degrees of freedom = 2, p = 0.0004,
Shapiro-Francia test for normality of residuals: W = 0.994, Z = 3.851, Prob > Z = 0.00006,
Multicollinearity test: Not needed with only one explanatory variable,
0.035, showing that only 3.5% of the variation in heart attack readmission is explained by
satisfaction. This was expected because of omitted explanatory variables such as patients
inherent risk characteristics and quality of care. The adjusted R2 is lower at 0.033. The RMSE has
a high value of 1.31, showing that the model t is very low. Predicted R2 is 0.034 and within 0.2
of R2 . The overall F statistic (F (1, 2398) = 84.61) is signicant (Prob > F = 0.0000), showing
that the readmission rate in heart attack is signicantly associated with patient satisfaction
component poor satisfaction. For simple regression, the F statistic gives the same value as the
two-sided test for signicance of the regression coefcient. A unit increase in Poor satisfaction
component score is associated with 0.26 increase in risk adjusted 30-day readmission rate in heart
171
attack. The coefcient was positive as expected and signicant: 95% CI was [0.21,0.32]). ES of
0.035 was above Cohens limit of 0.02 for small effect and below the limit of 0.15 for medium
effect. In summary, the research hypothesis is supported by the OLS regression results.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 60. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because of the large sample
size (N = 2,400). Whites test rejected homoscedasticity (Pr. > F = 0.00004), thus necessitating
regression with robust standard errors. A residuals versus tted values plot (given in Fig. 33)
showed that the pattern of data points showed some evidence of heteroscedasticity at higher
values of mortality but not to a great extent around the mean level. Regression was done with
robust standard errors in view of heteroscedasticity. Ramsey reset test for model specication
errors was signicant Prob > F = 0.0149, pointing to possible nonlinearity in the model. But,
Box-Tidwell test for nonlinearity, was not signicant (P = 0.872). Augmented component plus
residual plot with Locally Weighted Scatterplot Smoothing (LOWESS) line (shown in Fig. 34
also did not show much nonlinearity.
Figure 33. RVFplot for Heart Attack Readmission Rate - Satisfaction Component 1
Figure 34. ACPRplot for Heart Attack Readmission Rate - Satisfaction Component 1
172
In summary, the regression diagnostics provided support for the validity of OLS
regression.
Sensitivity analysis. Predicted R2 is 0.034 and within 0.2 of R2 , showing
cross-validation for new cases. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in Chapter 4, was used and this identied no inuential outliers. Robust regression,
gave an overall F statistic (F(1, 2398) = 71.20) which was not signicant at (P > |t| = 0.0000).
The coefcient was again positive and signicant with 95% CI of [0.18, 0.29]. The value was
only slightly different from previous OLS regression. In summary, robust regression did not nd
any inuential outliers and supported the research hypothesis for heart attack readmission rate.
The robust regression results are given in Table 61.
Table 61
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission
Rate for Heart Attack with Patient Satisfaction Component 1
Robust SE.
P > |t|
[95%
CI]
0.236
0.028
8.44
0.000
0.181
0.29
Constant term
19.84
0.027
735.31
0.000
19.78
19.89
173
Table 62
RQ5- Regression results for 30-day risk adjusted mortality rate for heart
failure by patient satisfaction component 1
30-day risk adjusted mortality rate (Heart
Robust SE.
P>t
[95%
CI]
failure)
Patient satisfaction - component 1
Constant term
-0.3
0.0256
-11.70
0.000
-0.35
-0.25
11.14
0.0265
420.26
0.000
11.08
11.19
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 10.98, df = 2, p = 0.0041,
residuals: W = 0.987, Z= 4.399, Prob. > Z = 0.00001,
explanatory variable,
Ramsey RESET test using powers of tted values: F(3,3414) = 2.80, Prob. > F = 0.0387
0.035, showing that only 3.5% of the variation in heart failure mortality is explained by
satisfaction. This was expected because of omitted explanatory variables such as patients
inherent risk characteristics and care of quality. The adjusted R2 is lower at 0.035. The RMSE has
a high value of 1.55, showing that the model t is very low. Predicted R2 is 0.034, and within 0.2
of R2 . The overall F statistic (F( 1, 3417)= 136.98) is signicant (at P > t = 0.0000), showing
that the mortality rate in heart failure is signicantly associated with patient satisfaction
component poor satisfaction. For simple regression, the F statistic gives the same value as the
two-sided test for signicance of the regression coefcient. The coefcient was negative while
the expected sign was positive and signicant: 95% CI was [-.0.35, -0.25]. ES of 0.04 was above
Cohens limit of 0.02 for small effect and below the limit of 0.15 for medium effect. A unit
increase in Poor satisfaction component score is associated with 0.03 decrease in risk adjusted
174
30-day readmission rate in heart failure. In summary, the research hypothesis is not supported for
heart failure mortality rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 62. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because of the large sample
size (N > 3,400). Whites test rejected homoscedasticity (Pr > F = 0.0041), thus necessitating
regression with robust standard errors. A residuals versus tted values plot (given in Fig. 35
showed that the pattern of data points showed some evidence of heteroscedasticity at lower tted
values of mortality but not to a great extent around the mean level. Regression was done with
robust standard errors in view of heteroscedasticity. Ramsey reset test for model specication
errors was signicant (Prob > F = 0.0387), pointing to possible nonlinearity in the model. But,
Box-Tidwell test for nonlinearity was not signicant (P = 0.749). Augmented component plus
residual plot with lowess line (shown in Fig. 36 also did not show much nonlinearity. In
summary, the regression diagnostics justied OLS regression.
Figure 35. RVFplot for Heart Failure Mortality Rate - Satisfaction Component 1
Figure 36. ACPRplot for Heart Failure Mortality Rate - Satisfaction Component 1
175
Table 63
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Failure with Patient Satisfaction Component 1
Robust SE.
P > |t|
[95%
CI]
-0.299
0.0268
-11.16
0.000
-0.351
-0.246
Constant term
11.059
0.0263
420.10
0.000
11.007
11.11
176
Robust SE.
P > |t|
[95%
CI]
0.43
0.036
11.79
0.000
0.35
0.5
24.49
0.035
690.66
0.000
24.42
24.55
failure)
Patient satisfaction - component 1
Constant term
Summary of regression results:
Model degrees of freedom = 1, Residual degrees of freedom = 3397, F (1,3397) = 140.07, Prob. > F = 0.0000
R2 = 0.039, Adjusted R2 = 0.039, Predicted R2 = 0.038
Ramsey RESET test using powers of tted values: F(3,3394) = 7.11, Prob. > F = 0.0001
R2 has low value of 0.039, showing that only 3.9% of the variation in heart failure
readmission is explained by satisfaction. This was expected because of omitted explanatory
variables such as patients inherent risk characteristics and quality of care. The adjusted R2 is
lower at 0.035. The RMSE has a high value of 1.55, showing that the model t is low. Predicted
R2 is low at 0.034 and within 0.2 of R2 . The overall F statistic (F( 1, 3417)= 136.98), is
signicant at Prob. > F = 0.0000, showing that the readmission rate in heart failure is
signicantly associated with patient satisfaction component 1 (poor satisfaction). For simple
regression, the F statistic gives the same value as the two-sided test for signicance of the
regression coefcient. The coefcient was positive as expected and signicant: 95% CI was
[0.35, 0.5]. ES of 0.04 was above Cohens limit of 0.02 for small effect and below the limit of
177
0.15 for medium effect. A unit increase in Poor satisfaction component score is associated with
0.43 increase in risk adjusted 30-day readmission rate in heart failure. In summary, the research
hypothesis is supported by the OLS regression results.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 64. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because of the large sample
size of 3,400. Whites test rejected homoscedasticity (Pr > F = 0.0084), thus necessitating
regression with robust standard errors. A residuals versus tted values plot (given in Fig. 37
showed that the pattern of data points showed some evidence of heteroscedasticity at lower tted
values of mortality but not to a great extent around the mean level. Ramsey reset test for model
specication errors was signicant Prob > F = 0.0001, pointing to possible nonlinearity in the
model. Box-Tidwell test for nonlinearity was also signicant (P = 0.001). Augmented component
plus residual plot with lowess line (shown in Fig. 38 did not show much nonlinearity. In
summary, the regression diagnostics showed support for using OLS regression.
Figure 37. RVFplot for Heart Failure Readmission Rate - Satisfaction Component 1
Figure 38. ACPRplot for Heart Failure Readmission Rate - Satisfaction Component 1
178
Table 65
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission
Rate for Heart Failure with Patient Satisfaction Component 1
Robust SE.
P > |t|
[95%
CI]
0.41
0.036
11.42
0.000
0.338
0.479
Constant term
24.42
0.035
696.65
0.000
24.35
24.490
179
Table 66
RQ5- Regression results for 30-day risk adjusted mortality rate for
pneumonia by patient satisfaction component 1
30-day risk adjusted mortality rate (pneumo-
Robust SE.
P > |t|
[95%
CI]
-0.1
0.032
-3.15
0.002
-0.162
-0.038
11.55
0.032
360.32
0.000
11.49
11.62
nia)
Patient satisfaction - component 1
Constant term
Summary of regression results:
= 9.95, Prob. > F = 0.0016,
RMSE = 1.8885 , Model sum of squares = 33.44, Residual sum of squares = 12367.78
Effect size:
f 2 = 0.0031; Cohen considers f 2 of 0.02 to be small, 0.15 medium and 0.35 as large effect.,
= 10.79, Power = 0.890, (From Table 9.3.2 of Cohen (1988) for a=0.05)
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 3.80, df = 2, p = 0.1495,
residuals: W = 0.984, Z= 4.589, Prob. > Z = 0.00001,
explanatory variable,
Ramsey RESET test using powers of tted values: F(3,3465) = 0.21, Prob. > F = 0.8872
0.0027, showing that only 0.27% of the variation in pneumonia mortality is explained by
satisfaction. This was expected because of omitted explanatory variables such as patients
inherent risk characteristics and quality of care. The adjusted R2 taking into account the number
of regression parameters and obtained by using the mean sum of squares is lower at 0.0024. The
RMSE has a high value of 1.55, showing that the model t is very low. Predicted R2 is also low at
0.034, but within 0.2 of R2 . The overall F statistic (F (1,3468) = 9.95), is signicant at Prob > F
= 0.0016, showing that the mortality rate in pneumonia is signicantly associated with patient
satisfaction component poor satisfaction. For simple regression, the F statistic gives the same
value as the two-sided test for signicance of the regression coefcient. The coefcient was
180
negative while the expected sign was positive and signicant. 95% CI was [-0.16, -0.038]. ES of
0.003 was much below Cohens limit of 0.02 for small effect. The regression coefcient was not
interpreted because of the very small effect size. In summary, the research hypothesis is not
supported in pneumonia mortality rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 66. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because no hypotheses are
tested. Whites test failed to reject homoscedasticity (Pr > F = 0.1495). However, this test is
sensitive to normality assumption and so a residuals versus tted values plot was created and is
given in Fig. 39. This showed that the pattern of data points showed some evidence of
heteroscedasticity at lower tted values of mortality but not to a great extent around the mean
level. Ramsey reset test for model specication errors was not signicant (Prob. > F = 0.8872),
pointing to no specication errors in the model. Box-Tidwell test for nonlinearity was also not
signicant (P = 0.757). Augmented component plus residual plot with lowess line (shown in Fig.
40 also did not show much nonlinearity. In summary, the regression diagnostics validated the
OLS regression.
181
Robust SE.
P > |t|
[95%
CI]
-0.10
0.033
-3.10
0.002
-0.165
-0.037
Constant term
11.45
0.032
356.50
0.000
11.386
11.512
182
Table 68
RQ5- Regression results for 30-day risk adjusted readmission rate for
pneumonia by patient satisfaction component 1
30-day risk adjusted readmission rate (pneu-
Robust SE.
P > |t|
[95%
CI]
monia)
Patient satisfaction - component 1
Constant term
0.306
0.032
9.67
0.000
0.244
0.37
18.2
0.029
631.8
0.000
18.14
18.25
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 25.31, df = 2, p = 0.000,
residuals: W = 0.986, Z= 4.475, Prob > Z = 0.00001,
explanatory variable,
Ramsey RESET test using powers of tted values: F(3,3424) = 1.86, Prob. > F = 0.1344
183
unit increase in readmission rate in pneumonia. In summary, the research hypothesis is supported
in pneumonia readmission rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 68. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because no hypotheses are
tested. Whites test rejected homoscedasticity (Pr > F = 0.000). However, this test is sensitive to
normality assumption and so a residuals versus tted values plot was created and is given in Fig.
41. This showed that the pattern of data points showed some evidence of heteroscedasticity at
lower tted values of mortality but not to a great extent around the mean level. Ramsey reset test
for model specication errors was not signicant (Prob > F = 0.1344), pointing to no
specication errors in the model. Box-Tidwell test for nonlinearity was also not signicant (P =
0.567). Augmented component plus residual plot with lowess line (shown in Fig. 42 also did not
show much nonlinearity. In summary, the regression diagnostics validated the OLS regression.
Figure 41. RVF plot for Pneumonia Readmission Rate - Satisfaction Component 1
184
Robust SE.
P > |t|
[95%
CI]
0.28
0.029
9.59
0.000
0.22
0.336
18.12
0.029
635.00
0.000
18.06
18.174
monia)
Patient satisfaction - component 1
Constant term
5.5 and 5.6 Relationships of Outcome Variables with Process of Care Quality Components
The relationship between outcomes and quality were tested in a similar way. Here, all the
ve components of quality have to be included in the regression whereas only the rst component
185
Table 70
Regression Coefcients on Outcome Variables and Poor Satisfaction
Satisfaction component
Poor satisfaction
Effect size (f 2 )
Effect size description
Power
Mortality rates
Heart Failure
-0.3
0.0031
<small
0.990
Pneumonia
-0.1
0.0346
>small and <medium
0.890
Readmission rates
Heart attack
0.265
0.0346
> small and < medium
0.990
Heart Failure
0.43
0.0389
>small and <medium
0.990
Pneumonia
0.306
0.0302
>small and <medium
0.990
of patient satisfaction was included in 5.3 and 5.4; Otherwise, there could be an omitted variable
bias in the model if relevant explanatory variables are excluded.
186
Table 71
RQ5 - Regression results for 30-day risk adjusted mortality rate for heart
attack by quality components
Robust SE.
P > |t|
[95%
CI]
Quality component 1
-0.155
0.078
-1.99
0.047
-0.308
-0.002
Quality component 2
-0.021
0.086
-0.24
0.809
-0.189
0.148
Quality component 3
-0.298
0.079
-3.78
0.000
-0.453
-0.143
Quality component 4
-0.024
0.069
-0.35
0.725
-0.161
0.112
Quality component 5
0.012
0.080
0.15
0.884
-0.146
0.170
15.881
0.075
212.40
0.000
15.734
16.028
Constant term
Summary of regression results:
Model degree of freedom = 5, Residual degree of freedom = 607, F (5,607) = 3.37, Prob. > F = 0.0052,
R2 = 0.0326, Adjusted R2 = 0.025, Predicted R2 = 0.011,
Effect size:
f 2 = 0.034 ; Cohen (1988) considers f 2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 20.66, Power = 0.92 (From Table 9.3.2 of Cohen (1988) for a=0.05)
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 31.66, df = 20, p = 0.0471,
residuals: W = 0.991, Z = 3.095, Prob. > Z = 0.00099,
Ramsey RESET test using powers of the tted values: F(3, 604) = 4.83, Prob. > F = 0.0025
of 0.0326, showing that only 3.3% of the variation in heart attack mortality is explained by quality
components. This was expected because of omitted explanatory variables such as patients
inherent risk characteristics. The adjusted R2 taking into account the number of regression
parameters and obtained by using the mean sum of squares is lower at 0.025. The RMSE has a
high value of 1.85, showing that the model t is low. Predicted R2 is 0.011 and within 0.2 of R2 .
The overall F statistic (F (5,607) = 3.37) is signicant at Prob > F = 0.0052, showing that the
mortality rate in heart attack is signicantly associated with quality components. The two
signicant coefcients were negative as expected: For component 1 (heart attack, failure
187
related), the was -0.16 with a 95% CI of [-0.31, -0.002], implying that a unit increase in this
quality component is associated with a decrease of 0.16 in the 30-day risk adjusted mortality rate
in heart attack. For component 3 (surgical care related), the was -0.3 with a 95% CI of
[-0.453, -0.143], implying that a unit increase in this quality component is associated with a
decrease of 0.3 in the 30-day risk adjusted mortality rate in heart attack. ES of 0.034 was higher
than Cohens limit of 0.02 for small effect and lower than 0.15 for medium effect. In summary,
the research hypothesis is supported for two quality components in heart attack mortality rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 71. Shapiro-Francia normality test showed
that the residuals are not normally distributed, but this is not critical because no hypotheses are
tested. Whites test rejected homoscedasticity (Pr > F = 0.0471), at 95% condence level.
However, this test is sensitive to normality assumption and so a residuals versus tted values plot
was created and is given in Fig. 43. This showed that the pattern of data points showed some
evidence of heteroscedasticity at higher tted values of mortality but not to a great extent around
the mean level. Presence of some outliers is seen in the plot. Ramsey reset test for model
specication errors was signicant (Prob > F = 0.0025), pointing to probable nonlinearities in the
model. Box-Tidwell test for nonlinearity was not signicant except for quality component 3
(Nonlinear deviation 11.901 P = 0.001). For this component, an augmented component plus
residual plot with lowess line was created and shown in Fig. 44. This plot did not show much
nonlinearity, but the presence of outliers is seen. In summary, the regression diagnostics validated
the OLS regression.
Figure 43. RVF plot for Heart Attack Mortality Rate - Quality Components
188
Figure 44. ACPRplot for Heart Attack Mortality Rate - Quality Component 3
Sensitivity analysis. Predicted R2 is 0.029 and is within 0.2 of R2 , showing
cross-validation for new cases. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in Chapter 4, was used and this failed to identify any inuential outliers. Robust
regression, gave an F( 5, 607) = 5.44 which was signicant at (P > |t| = 0.0001. The quality 1
and quality 3 coefcients were again negative and signicant with 95% CI of [-0.33, -0.04] and
95% CI of [-0.48, -0.19]. The values are only slightly changed from the previous OLS regression,
showing that the OLS regression test results were not affected by the robust regression. In
summary, the research hypothesis is supported by the robust regression for two quality
components in heart attack mortality rate. Table 72 shows the robust regression results.
Table 72
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Heart Attack with Process of Care Quality Components
Robust SE.
P > |t|
[95%
CI]
Quality component 1
-0.184
0.074
-2.48
0.013
-0.33
-0.039
Quality component 2
-0.009
0.074
-0.13
0.899
-0.155
0.136
Quality component 3
-0.334
0.074
-4.49
0.000
-0.48
-0.188
Quality component 4
0.0032
0.075
0.04
0.965
-0.143
0.15
Quality component 5
0.0656
0.074
0.88
0.379
-0.081
0.212
Constant term
15.819
0.074
212.89
0.000
15.673
15.965
189
Robust SE.
P > |t|
[95%
CI]
Quality component 1
-0.09
0.07
-1.32
0.19
-0.23
0.05
Quality component 2
-0.23
0.07
-3.38
0.00
-0.36
-0.10
Quality component 3
-0.03
0.06
-0.41
0.68
-0.15
0.10
Quality component 4
0.06
0.06
1.07
0.29
-0.05
0.18
Quality component 5
0.05
0.06
0.88
0.38
-0.07
0.18
19.74
0.06
320.13
0.00
19.62
19.86
Constant term
Summary of regression results:
Model degrees of freedom = 5, Residual degrees of freedom = 604, F(5,604) = 3.13, Prob. > F = 0.0085,
R2 = 0.03, Adjusted R2 = 0.02, Predicted R2 = 0.01,
Effect size:
f 2 = 0.0298; Cohen(1988) considers f 2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 18.33, Power = 0.99 (From Table 9.3.2 of Cohen (1988) for a=0.05)
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 35.12, df = 20, P = 0.0195,
residuals: W = 0.997, Z = 0.376, Prob. > Z = 0.353,
Ramsey RESET test using powers of the tted values: F(3, 601) = 0.99, Prob. > F = 0.3992
R2 has low value of 0.03, showing that only 3% of the variation in heart attack
readmission is explained by quality components. This was expected because of the presence of
omitted explanatory variables such as patients inherent risk characteristics. The adjusted R2
taking into account the number of regression parameters and obtained by using the mean sum of
squares is lower at 0.02. The RMSE has a high value of 1.52, showing that the model t is low.
Predicted R2 is 0.01, and within 0.2 of R2 . The overall F statistic (F (5,604) = 3.13) is signicant
190
at Prob > F = 0.0052, showing that the readmission rate in heart attack is signicantly associated
with quality components. The only signicant coefcient was negative as expected. The
coefcient for component 2 (pneumonia related) was -0.23 with a 95% CI of [-0.35, -0.095],
implying that a unit increase in this quality component is associated with a decrease of 0.23 in the
30-day risk adjusted readmission rate in heart attack. ES of 0.03 was higher than Cohens limit of
0.02 for small effect and lower than 0.15 for medium effect. In summary, the research hypothesis
is supported for pneumonia related quality component in heart attack readmission rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 73. Shapiro-Francia normality test showed
that the residuals are normally distributed. Whites test rejected homoscedasticity
(Pr > F = 0.0195), at 95% condence level. However, this test is sensitive to normality
assumption and so a residuals versus tted values plot was created and is given in Fig. 45. This
showed that the pattern of data points showed little evidence of heteroscedasticity. Presence of
some outliers is seen in the plot. Ramsey reset test for model specication errors was not
signicant (Prob > F = 0.3992). Box-Tidwell test for nonlinearity was not signicant except for
quality component 1 (Nonlinear deviation 6.818 P = 0.010). For this component, an augmented
component plus residual plot with lowess line was created and shown in Fig. 46. This plot did
not show much nonlinearity, but the presence of outliers is seen. In summary, the regression
diagnostics provided validity for OLS regression.
Figure 45. RVF Plot for Heart Attack Readmission Rate - Quality Components
191
Figure 46. ACPRplot for Heart Attack Readmission Rate - Quality Component 1
Sensitivity analysis. Predicted R2 is 0.01 and within 0.2 of R2 = 0.03, showing
cross-validation for new cases. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers,
used earlier in Chapter 4, was used and this failed to identify any inuential outliers. Robust
regression, gave an F( 5, 604) = 3.58 which was signicant at (P > |t| = 0.0034. The quality 2
coefcient was again negative and signicant with 95% CI of [-0.36, -0.11]. The value was only
slightly changed to -0.24 from previous -O.23. This shows that the OLS regression test results
were not affected by the robust regression. In summary, the robust regression conrms that the
research hypothesis is supported for pneumonia related quality component in heart attack
readmission rate. The robust regression results are given in Table 74.
Table 74
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate
for Heart Attack with Process of Care Quality Components
Robust SE.
P > |t|
[95%
CI]
Quality component 1
-0.079
0.063
-1.25
0.212
-0.202
0.045
Quality component 2
-0.238
0.063
-3.77
0.000
-0.362
-0.114
Quality component 3
-0.025
0.063
-0.40
0.693
-0.149
0.0990
Quality component 4
0.0546
0.065
0.84
0.399
-0.072
0.1817
Quality component 5
0.0683
0.064
1.07
0.284
-0.057
0.1933
Constant term
19.716
0.063
311.65
0.000
19.591
19.84
192
Table 75
RQ5 - Regression results for 30-day risk adjusted mortality rate for heart
failure by quality components
Robust SE.
P > |t|
[95%
CI]
Quality component 1
-0.09
0.06
-1.37
0.17
-0.22
0.04
Quality component 2
0.12
0.06
1.92
0.06
0.00
0.24
Quality component 3
-0.28
0.08
-3.68
0.00
-0.43
-0.13
Quality component 4
-0.09
0.06
-1.57
0.12
-0.21
0.02
Quality component 5
-0.02
0.06
-0.35
0.72
-0.14
0.10
Constant term
10.92
0.06
172.00
0.00
10.79
11.04
Effect size:
f 2 = 0.0298; Cohen considers f 2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 18.15, Power = 0.87 (From Table 9.3.2 of Cohen (1988) for a=0.05]
Regression diagnostics:
Whites test for heteroscedasticity: 2 = 43.82, df = 20, p = 0.0016,
residuals: W = 0.993, Z = 2.602, Prob. > Z = 0.00464,
Ramsey RESET test using powers of the tted values: F(3, 604) = 1.18, Prob. > F = 0.3182
R2 has low value of 0.043, showing that only 4.3% of the variation in heart failure
mortality is explained by quality components. This was expected because of omitted explanatory
variables such as patients inherent risk characteristics. The adjusted R2 taking into account the
number of regression parameters and obtained by using the mean sum of squares is lower at
193
0.035. The RMSE has a high value of 1.57, showing that the model t is low. Predicted R2 is
0.01, and within 0.2 of R2 . The overall F statistic (F (5,607) = 4.15) is signicant at Prob. > F =
0.0010, showing that the mortality rate in heart failure is signicantly associated with quality
components. The only signicant coefcient was negative as expected. The coefcient for
component 3 (surgical care related) was -0.28 with a 95% CI of [-0.43, -0.13], implying that a
unit increase in this quality component is associated with a decrease of 0.28 in the 30-day risk
adjusted mortality rate in heart failure, on average. ES of 0.03 was higher than Cohens limit of
0.02 for small effect and lower than 0.15 for medium effect. In summary, the research hypothesis
is supported for the surgical care related quality component in heart failure mortality rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 75. Shapiro-Francia normality test showed
that the residuals are not normally distributed, necessitating regression with robust standard
errors. Whites test rejected homoscedasticity (Pr > F = 0.016), at 95% condence level.
However, this test is sensitive to normality assumption and so a residuals versus tted values plot
was created and is given in Fig. 47. This showed that the pattern of data points showed little
evidence of heteroscedasticity. Presence of some outliers is seen in the plot. Ramsey reset test for
model specication errors was not signicant (Prob > F = 0.3182). Box-Tidwell test for
nonlinearity was not signicant for all quality components, thus conrming the absence of
nonlinearities in the model. In summary, the regression diagnostics validated the OLS regression.
Figure 47. RVF Plot for Heart Failure Mortality Rate - Quality Components
194
Robust SE.
P > |t|
[95%
CI]
Quality component 1
-0.082
0.063
-1.30
0.194
-0.207
0.042
Quality component 2
0.1015
0.063
1.60
0.110
-0.023
0.226
Quality component 3
-0.2762
0.064
-4.35
0.000
-0.401
-0.151
Quality component 4
-0.1091
0.064
-1.71
0.087
-0.234
0.0160
Quality component 5
-0.0231
0.064
-0.36
0.716
-0.148
0.1018
10.876
0.063
171.31
0.000
10.752
11.002
Constant term
195
Table 77
RQ5 - Regression results for 30-day risk adjusted readmission rate for
heart failure by quality components
Robust SE.
P>t
[95%
CI]
Quality component 1
-0.20
0.08
-2.40
0.02
-0.37
-0.04
Quality component 2
-0.13
0.08
-1.53
0.13
-0.30
0.04
Quality component 3
0.01
0.08
0.15
0.88
-0.15
0.18
Quality component 4
0.04
0.09
0.48
0.63
-0.13
0.21
Quality component 5
0.18
0.09
2.11
0.04
0.01
0.35
Constant term
23.97
0.08
282.07
0.00
23.80
24.14
Effect size:
f 2 = 0.0211: Cohen considers f 2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 12.92, Power = 0.77 (From Table 9.3.2/Page 420 of Cohen (1988) for a=0.05)
Regression diagnostics:
Shapiro-Francia test for normality of residuals: W = 0.997, Z = 0.476, Prob. > Z = 0.317,
Multicollinearity test: Mean VIF = 1.00,
Ramsey RESET test using powers of the tted values: F(3, 602) =
R2 has low value of 0.0207, showing that only 2.1% of the variation in heart failure
readmission is explained by quality components. This was expected because of omitted
explanatory variables such as patients inherent risk characteristics. The adjusted R2 taking into
account the number of regression parameters and obtained by using the mean sum of squares is
lower at 0.013. The RMSE has a high value of 2.1, showing that the model t is low. Predicted R2
is 0.003, and within 0.2 of R2 . The overall F statistic (F(5,605) = 2.46) is signicant at Prob. > F
= 0.0318, showing that the readmission rate in heart failure is signicantly associated with quality
components. The coefcient for component 1 (heart attack, failure related) was -0.2 with a
95% CI of [-0.37, -0.041], implying that a unit increase in this quality component is associated
196
Figure 48. RVF plot for Heart Attack Mortality Rate - Quality Components
with a decrease of 0.2 in the 30-day risk adjusted readmission rate in heart failure, on average.
Component 5 (prevention related) has a signicant coefcient = 0.18 (95% CI is [0.0004,
0.36]), but the sign is positive and the interpretation is difcult. ES of 0.021 was higher than
Cohens limit of 0.02 for small effect and the effect is taken as small. In summary, the research
hypothesis is supported for the heart attack, failure related quality component in heart failure
readmission rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 77. Shapiro-Francia normality test showed
that the residuals are normally distributed. Whites test failed to reject homoscedasticity
(Pr > F = 0.6444). However, this test is sensitive to normality assumption and so a residuals
versus tted values plot was created and is given in Fig. 48. This showed that the pattern of data
points showed little evidence of heteroscedasticity. Ramsey reset test for model specication
errors was not signicant (Prob > F = 0.33367), showing that the model may not have
nonlinearities. Box-Tidwell test for nonlinearity was not signicant except for quality
components 1 and 5. Augmented component plus residuals plots with lowess line were created
and these are shown in Fig. 49 and 50. Both do not show evidence of signicant nonlinearity. In
summary, the regression diagnostics validated the OLS regression.
Sensitivity analysis. Predicted R2 is 0.003 and within 0.2 of R2 , showing
cross-validation for new cases. R2 being small, we need to rule out the possibility of a few
inuential outliers affecting the model predictions. A procedure to identify the inuential outliers
197
Figure 49. ACPRplot for Heart Failure Readmission Rate and Quality Component 1
Figure 50. ACPRplot for Heart Failure Readmission Rate and Quality Component 5
used earlier in Chapter 4, was used and this failed to identify any inuential outliers. Robust
regression, gave an F( 5, 605) = 2.70 which was signicant at (P > |t| = 0.0202). The quality
component 1 has a regression coefcient that was again negative and signicant with 95% CI of
[-0.376, -0.035]. The value was only slightly changed to -0.205 from previous -0.2. As before,
component 5 has a positive coefcient. This shows that the OLS regression test results were not
affected by the robust regression. The robust regression results are given in Table 78. In summary,
robust regression conrms that the research hypothesis is supported for the heart attack, failure
198
Robust SE.
P > |t|
[95%
CI]
Quality component 1
-0.205
0.087
-2.37
0.018
-0.376
-0.035
Quality component 2
-0.108
0.087
-1.24
0.215
-0.278
0.0627
Quality component 3
0.044
0.087
0.51
0.612
-.126
0.2146
Quality component 4
0.049
0.089
0.56
0.578
-0.125
0.2235
Quality component 5
0.209
0.087
2.39
0.017
0.0375
0.3802
Constant term
23.94
0.087
275.39
0.000
23.777
24.119
199
Table 79
RQ5 - Regression results for 30-day risk adjusted mortality rate for
pneumonia by quality components
Robust SE.
P > |t|
[95%
CI]
Quality component 1
0.019
0.073
0.250
0.800
-0.125
0.162
Quality component 2
-0.064
0.080
-0.800
0.426
-0.221
0.094
Quality component 3
-0.278
0.081
-3.440
0.001
-0.437
-0.119
Quality component 4
-0.118
0.075
-1.560
0.119
-0.266
0.03
Quality component 5
0.078
0.072
1.090
0.278
-0.063
0.218
11.283
0.073
155.270
0.000
11.140
11.426
Constant term
Summary of regression results:
Model degrees of freedom = 5, Residual degrees of freedom = 607, F(5,607) = 3.13, Prob. > F = 0.0085,
R2 = 0.0308, Adjusted R2 = 0.0228, Predicted R2 = 0.00964,
Effect size:
f 2 = 0.032: Cohen (1988) considers f 2 of 0.02 to be small, 0.15 medium and 0.35 as large effect)
= 19.67, Power = 0.95 (From Table 9.3.2/Page 420 of Cohen (1988) for a=0.05)
Regression diagnostics:
Shapiro-Francia test
Ramsey RESET test using powers of the tted values: F(3, 604) = 1.47, Prob. > F = 0.2211
R2 has low value of 0.0308, showing that only 3.1% of the variation in pneumonia
mortality is explained by quality components. This was expected because of the omitted
explanatory variables such as patients inherent risk characteristics. The adjusted R2 taking into
account the number of regression parameters and obtained by using the mean sum of squares is
lower at 0.0228. The RMSE has a high value of 1.8 , showing that the model t is low. Predicted
R2 is 0.00964, and within 0.2 of R2 . The overall F statistic (F(5,607) = 3.13) is signicant at
Prob. > F = 0.0085, showing that the mortality rate in pneumonia is signicantly associated with
quality components. The signicant coefcient for quality component 3 (surgical care
related) was -0.28 with a 95% CI of [-0.44, -0.12], implying that a unit increase in this
200
component is associated with a decrease of 0.28 in the 30-day risk adjusted mortality rate in
pneumonia, on average. ES of 0.032 was higher than Cohens limit of 0.02 for small effect and
below the limit of 0.15 for medium effect. In summary, the research hypothesis is supported for
the surgical care related quality component in pneumonia mortality rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 79. Shapiro-Francia normality test showed
that the residuals are not normally distributed. Whites test failed to reject homoscedasticity
(Pr > F = 0.2409). However, this test is sensitive to normality assumption and so a residuals
versus tted values plot was created and is given in Fig. 51. This showed that the pattern of data
points showed little evidence of heteroscedasticity. Ramsey reset test for model specication
errors was not signicant (Prob > F = 0.2211), showing that the model may not have
nonlinearities. Box-Tidwell test for nonlinearity was not signicant for all quality components,
conrming absence of nonlinearities. In summary, the regression diagnostics establish the validity
of OLS regression.
201
value was only slightly changed to -0.283 from previous -0.278. This shows that the OLS
regression test results were not affected by the robust regression. In summary, the robust
regression conrms that the research hypothesis is supported for the surgical care related quality
component in pneumonia mortality rate. The robust regression results are given in Table 80.
Table 80
RQ5 - Robust Regression Results for 30-day Risk Adjusted Mortality Rate for
Pneumonia with Process of Care Quality Components
DV: 30-day risk adjusted readmission rate
Robust SE.
P > |t|
[95%
CI]
(pneumonia)
Independent variables:
Quality component 1
0.0218
0.074
0.29
0.769
-0.124
0.168
Quality component 2
-0.077
0.0743
-1.04
0.300
-0.223
0.069
Quality component 3
-0.283
0.0744
-3.80
0.000
-0.429
-0.137
Quality component 4
-0.104
0.0746
-1.39
0.164
-0.250
0.043
Quality component 5
0.0928
0.0745
1.25
0.213
-0.054
0.239
Constant term
11.229
0.0744
151.00
0.000
11.083
11.374
202
Table 81
RQ5 - Regression results for 30-day risk adjusted readmission rate for
pneumonia by quality components
Robust SE.
P > |t|
[95%
CI]
Quality component 1
0.089
0.066
1.35
0.176
-0.04
0.219
Quality component 2
-0.212
0.0678
-3.13
0.002
-0.345
-0.079
Quality component 3
0.089
0.0699
1.27
0.205
-0.0485
0.226
Quality component 4
-0.006
0.076
-0.08
0.937
-0.155
0.143
Quality component 5
0.0263
0.0677
0.39
0.698
-0.107
0.159
18.11
0.0697
259.89
0.000
17.975
18.249
Constant term
Summary of regression results:
Model degrees of freedom = 5, Residual degrees of freedom = 603, F(5,604) = 2.51, Prob. > F = 0.0288,
R2 = 0.0207, Adjusted R2 = 0.013, Predicted R2 = 0.001,
Effect size:
f 2 = 0.0211; Cohen considers R2 of 0.02 to be small, 0.15 medium and 0.35 as large effect,
= 13.02, Power = 0.84 (From Table 9.3.2 of Cohen (1988) for a=0.05)
Regression diagnostics:
Shapiro-Francia
Ramsey RESET test using powers of the tted values: F(3, 601) = 2.24, Prob > F = 0.0822
R2 has low value of 0.0207, showing that only 2.1% of the variation in pneumonia readmission
rate is explained by quality components. This was expected because of the omitted explanatory
variables such as patients inherent risk characteristics. The adjusted R2 taking into account the
number of regression parameters and obtained by using the mean sum of squares is lower at
0.013. The RMSEhas a high value of 1.72, showing that the model t is low. Predicted R2 is
0.001, and within 0.2 of R2 . F(5,604) = 2.51, is signicant at Prob. F = 0.0288, showing that
the readmission rate in pneumonia is signicantly associated with quality components. The
signicant coefcient for quality component 2 (pneumonia related) was -0.21 with a 95% CI
of [-0.35, -0.08] was negative as expected, implying that a unit increase in this quality component
203
is associated with a decrease of 0.21 in the 30-day risk adjusted readmission rate in pneumonia,
on average. ES of 0.0211 was slightly higher than Cohens limit of 0.02 for small effect and the
effect was assumed to be small. In summary, the research hypothesis is supported for pneumonia
related quality component for pneumonia readmission rate.
Regression diagnostic tests. The results of a battery of diagnostic tests for evaluating the
adequacy of the estimated are reported below Table 81. Shapiro-Francia normality test showed
that the residuals are normally distributed (p = 0.74). Whites test failed to reject
homoscedasticity (Pr > F = 0.8874). However, this test is sensitive to normality assumption and
so a residuals versus tted values plot was created and is given in Fig. 52. This showed that the
pattern of data points showed little evidence of heteroscedasticity. It shows a potential outlier.
Ramsey reset test for model specication errors was not signicant (Prob > F = 0.0.0822) at 5%
level, showing that the model may not have nonlinearities. In summary, the regression diagnostics
provided validation for OLS regression.
204
results were not affected by the robust regression. The robust regression results are given in
Table 82. In summary, robust regression conrmed that the research hypothesis is supported for
pneumonia related quality component for pneumonia readmission rate.
Table 82
RQ5 - Robust Regression Results for 30-day Risk Adjusted Readmission Rate
for Pneumonia with Process of Care Quality Components.
Robust SE.
P > |t|
[95%
CI]
Quality component 1
0.0765
0.0708
1.08
0.281
-0.063
0.216
Quality component 2
-0.222
0.0709
-3.13
0.002
-0.361
-0.083
Quality component 3
0.0681
0.0710
0.96
0.338
-0.071
0.2076
Quality component 4
-.00689
0.0727
-0.09
0.925
-0.150
0.136
Quality component 5
0.0320
0.0716
0.45
0.655
-0.109
0.173
Constant term
18.099
0.0711
254.53
0.000
17.960
18.239
Heart attack/failure
Pneumonia
Surgical care
Smoking cessation
Prevention
Effect size (f 2 )
Power
Outcome variables
(30-day mortality
and
readmission
rates)
Heart attack
ns
ns
0.41
-0.31
ns
0.0264
0.89
Mortality rates
Heart Failure
ns
0.12
-0.28
ns
ns
0.04
0.93
Pneumonia
ns
ns
-0.28
ns
ns
0.032
0.95
Heart attack
ns
-0.23
ns
ns
ns
0.029
0.99
Readmission rates
Heart Failure
-0.2
ns
ns
ns
0.18
0.0269
0.99
Pneumonia
ns
-0.2
ns
ns
ns
0.0256
0.89
205
Summary
In this chapter, a description was given of the archival data retrieved for analysis. The fth
research question was subdivided into three sub-questions:
For the research hypothesis that church hospitals would have lower outcomes than the other
owner groups, the evidence seems strong with an effect size between small and medium.
This evidence also brings out the importance of compassion and related qualities in
healthcare.
For the research hypothesis that greater patient satisfaction is associated with lower
outcomes, some evidence emerged but not conclusive. The hypothesis is supported for
readmission rates with a small to medium effect size but not for mortality rates.
For the research hypothesis that greater process of care quality is associated with lower
outcomes, evidence was partial and with a small to medium effect size.
The next chapter will discuss the statistical conclusions in this and the previous chapter in terms
of practical implications.
206
Chapter 6
Chapter Overview
In the preceding chapter, the data collected for the current study were discussed and
analyzed. This chapter begins with a summary of its purpose and structure. It is followed by the
main ndings related to the ve research questions that are dealt with in this study. Healthcare is
expensive in the US. Nevertheless, the patients are not terribly happy with the way they are
treated, and this has been brought out in various surveys. The government is attempting to deal
with this problem by introducing a system for performance based Medicare payments.
Government is also taking the initiative for improving process-of-care quality. Performance
improvement has now been made necessary for hospital accreditation. In these circumstances,
this study was undertaken to help hospitals identify areas for improving their patient satisfaction,
process-of-care quality and outcomes for the benet of patients. The study included ve research
questions. Chapter 6 consists of six sections:
1. Introduction
2. Review of the study ndings
3. Discussion of the ndings
4. Implications for hospitals
5. Recommendations for further research
207
6. Conclusion
The section on the discussion of ndings also includes a comparison with previous
research on this subject and threats to validity. The section on implication is presented for
providing an understanding about the factors that may inuence performance improvement in
hospitals, which is a key result area in healthcare. This study has its own limitations and these are
included in the section on further research. The conclusion will include leads to the next chapter
on relating this study to Maharishi Vedic science teachings on healthcare.
Review of ndings
The results of this study presented in Chapters 4, and 5 are summarized and discussed in
terms of the research questions and hypotheses in this section.
Research question 1.
Which dimensions of the patient perceptions of healthcare have the greatest impact on
patient satisfaction?
Patient satisfaction data obtained from HCAHPS surveys was downloaded from CMS
hospital compare database.
HCAHPS survey questions.
The HCAHPS survey uses 10 forced-response questions. To seven questions, patients
respond selecting one of three options (sometimes or never, usually and always). To the question
on Were you given information about what to do during recovery at home?, patients select
either yes or no. To the question on overall rating to the hospitals, patients can select a rating
out of 3 choices: 6 or lower (low), 7 or 8 (medium), 9 or 10 (high). To the question Would you
recommend the hospital to friends and family?, the patients should select one out of three
options: no, denitely yes, probably yes. There are thus 29 variables. A PCA was carried out on
208
satisfaction data after validating that the data conforms to PCA requirements, and using
standardized scores.
Five main dimensions were identied, accounting for 84% of the total variance.
Component 1 Poor satisfaction accounts for 59.3% of variance and is associated with
responses relating to dissatisfaction because the main loadings on this component came
from the responses sometimes or never.
Component 2 Required level of performance accounts for 11.3% of the variance and is
associated with responses on the expected level of performance (usually). This is the
minimum level of performance expected by the patients.
Component 3 Overall accounts for 6% of the variance, and is mostly associated with
responses to questions on overall rating and recommendation.
Component 4 Cleanliness accounts for 4.3% of the variance and is associated mainly with
response to cleanliness.
Component 5 Post-hospitalization care accounts for 3.5% of the variance and is associated
mainly with response to how well instructions were given to patients at discharge.
The contribution of the variables (%) to the rst ve principal components accounting for
84% of the standardized variance in patient satisfaction (as measured by HCAHPS survey) is
summarized in Table 5 in Chapter 4.
These are the main satisfaction dimensions on which hospitals should focus to improve
their satisfaction ratings. The rst component is the indicator of poor satisfaction, and accounts
for the bulk (59.3%) of the variance. Consequently, this was used as the prime indicator of patient
satisfaction, following the approach adopted by Webster (2001). This being connected with poor
satisfaction, hospitals need to reduce this component as much as possible.
209
210
The overall ratings and patient seem to be affected adversely by the poor satisfaction
component signicantly. Therefore, a hospitals efforts to improve communication and pain
management, are bound to yield improved ratings.
Expected level of performance. The main contribution to expected level of
performance comes from:
1. Staff usually explained about medicines before giving them to patients (15.1%).
2. Area around patients room is usually kept quiet at night (13.8%).
3. Pain was usually well controlled (12.2%).
4. Doctors usually communicated well (9.5%).
5. Patients usually received help when they wanted (8.6%).
Therefore, to improve the expected level of service, hospitals need to focus on following
areas:
Important role is played by properly explaining about medicines before giving them to
patients. Hospitals need to train their nurses and pharmacists to provide a complete
explanation to patients without taking their medical knowledge for granted. They should
explain clearly, avoiding any medical jargon and complicated language, keeping in mind
that any misunderstanding has the potential to cause a serious problem.
Another important principle for hospitals is the need to maintain quiet in and around patient
rooms especially at night.
This may need establishing silent hours at night (like in Maharishi University of
Management (MUM) dorms). Sometimes, sound proong patient rooms or changing the
hospital layout may be needed.
211
Important role of communication and early pain management are already covered
previously. In addition, patients expect to be provided help in a timely manner. This
requires hospitals to train all their staff to give help whenever patients request them.
Cleanliness. The main contribution to cleanliness comes from one item: Patient
rooms and bathrooms are always clean (26.4%). Therefore, to improve patient satisfaction
further, hospitals need to focus on keeping patient rooms and bathrooms clean and hygienic.
Besides improving patient satisfaction, this would also reduce cases of of hospital-acquired
infections. This is also likely to reduce safety incidents from unclean rooms and bathrooms. The
ndings showed that the overall rating and recommendation are greatly inuenced by patient
perceptions of hospital performance on these aspects.
Research Question 2.
What is the relationship between hospital ownership and patient satisfaction?
The research hypothesis that was tested for this question was as follows:
Hospitals owned by churches have higher mean patient satisfaction than the other 7
hospital ownership groups.
Hospital ownership groups for HCAHPS data were listed in Table 6 on page 82. The
ancient Vedic physician Charaka taught that four guiding principles in healthcare should be
friendship, compassion, joy in serving and equanimity. Churches run hospitals as a service and
are more likely to apply Charakas principles in practice than others.
The research hypothesis was tested statistically with an OLS regression using the rst
component score as the response variable and the ownership group binary variables as
explanatory variables. As in the case of all hypotheses tested in this study, signicances are based
on two-sided tests at 5% signicance level. The ndings from the regression were as follows:
The research hypothesis was supported in two of the seven owner groups (state government,
and proprietary). Both have signicantly higher mean poor satisfaction component than
212
the hospitals owned churches. The effect sizes (Cohens d) are 0.42 and 0.39 (low to
medium size).
The research hypothesis was not supported in ve out of seven owner groups (district /
authority, local government, voluntary nonprot - other, and voluntary nonprot - private
groups). They have signicantly lower mean poor satisfaction component than the church
group and the difference in mean poor satisfaction component between the church and
federal government groups was not statistically signicant.
Proprietary hospitals are commercial establishments, and this evidence shows that they
need to take conscious efforts and expenses to improve patient satisfaction. Among government
hospitals, only state government owned hospitals have lower patient satisfaction than church
owned hospitals. Federal government hospitals have mean patient satisfaction score not
signicantly different from church hospitals, while the other two types of government hospitals
(local and district/area) have signicantly higher patient satisfaction than church hospitals.
However, the ndings are conditional on the interpretation of component 1 as poor satisfaction.
In summary, the evidence does not give strong support for the research hypothesis which
is supported in only two out of seven owner groups.
Value based payments system for Medicare payments.
The effect sizes for statistical tests may seem small, but they need to be considered along
with the practical implications from the published CMS policy on value based purchasing (VBP)
system for Medicare payments. CMS will withhold 1% of its payments to hospitals i.e. 850
million $ in the rst year (2012). The withheld money will go into a pool that will be doled out as
bonuses that score above average on several measures. Fully 30% of the money will be utilized
for HCAHPS measures and the other 70% for process-of-care quality and other measures. Later
on, the withholding percentage will grow to 2% by 2017. Under this system, willingness to
recommend question would not be used for VBP and individual questions on cleanliness and
quietness would be averaged together. Hospitals would earn 1-10 points based on where they
213
fall in an achievement range from 50th to 95th percentile. Hospitals scoring below the 50th
percentile would receive an attainment score of zero and hospitals exceeding 95th percentile
would be awarded ten attainment points. Bonuses would be paid taking into account the
improvement and consistency in attainment.
Therefore, a hospital could lose thousands of dollars in Medicare payments alone for
effect sizes even as low as 0.1%, depending on in what percentile, its HCAHPS score lies.
Furthermore, hospitals may lose considerably in lost sales because these scores are used in the
hospital compare site maintained by CMS and patients are encouraged to use this site for
comparing different providers before selecting one for their use.
Research Question 3.
What is the relationship between hospital ownership and process-of-care quality
measures?
This research question was expressed in terms of the research hypothesis that can be
empirically tested:
Hospitals owned by church group have higher mean process-of-care quality than the other
7 groups.
Principal component analysis of quality data.
To analyze this research question, a PCA was rst done on the process-of-care quality
measures, to identify the main dimensions; otherwise, dealing with 24 quality variables would
have been difcult and time-consuming. The data concerning 24 clinical conditions covering
heart attack, heart failure, pneumonia, and surgical care were analyzed. PCA yielded ve
principal components for process-of-care quality measures.
Quality component 1 heart attack, failure related accounted for 15.8 % of the variance.
Quality component 2 pneumonia related accounted for 10.4 % of the variance.
214
Quality component 3 surgical care related accounted for 10.4 % of the variance.
Quality component 4 smoking cessation related accounted for 8.1% of the variance.
Quality component 5 prevention related accounted for 8.9% of the variance.
These were obtained after varimax orthogonal rotation and, therefore, the components are
not correlated. Each component is loaded by a set of quality indicators that are unique to a
dimension of quality.
Statistical tests on quality differences by ownership groups.
The regression results for research hypothesis 3 by OLS regression on the ve quality
components were summarized in Table 30 in Chapter 4. The ndings are as follows:
The research hypothesis is supported in the heart attack and failure related component in
one of the seven owner groups (hospital area or authority). The effect size was 0.37 (low to
medium). However, the other groups do not differ signicantly from the church group,
signifying that heart attack/failure related quality specications are being followed with no
signicant differences in the other six ownership groups.
The research hypothesis is supported in the pneumonia related component, in two of the
seven owner groups ( federal government with ES = 0.74 and state government with ES =
0.95). The effect sizes are high in both cases. However, church group hospitals scored
signicantly lower than proprietary hospitals with an ES = 0.31, a low to medium effect
size. The government hospitals need to improve their performance in respect of pneumonia
related quality specications. Proprietary hospitals score even higher than church hospitals
in this quality component.
The research hypothesis is supported in the surgical care related quality component in
one owner group ( government - hospital district or authority) with an ES of 0.60 (medium
to high). However, in contradiction to the research hypothesis, church hospitals scored
215
lower than the federal government (ES = 0.8), voluntary nonprot-other and voluntary
nonprot-private hospitals with ES of 0.41 and 0.44, respectively). Church hospitals need to
improve in surgical care related quality component.
In the smoking related quality component, no signicant difference between the
ownership groups was found.
The research hypothesis is supported in the prevention related quality component, in
three of the seven owner groups (state government with high ES of 1.1, voluntary non
prot-other with ES of 0.25, and voluntary nonprot-private with ES of 0.24). However,
hospitals owned by churches, scored lower than proprietary hospitals with ES of 0.32 (low
to medium). Prot oriented proprietary hospitals showed better quality than the other
groups in prevention.
In summary, it is, therefore, reasonable to conclude that the research hypothesis is
supported in 7 but contradicted in 5 out of 35 regression coefcients. However, the evidence is
mixed in favor of the church group.
Research Question 4.
What is the relationship between process-of-care quality measures and patient
satisfaction?
The research hypotheses under this question assume that hospitals compromise on the
process-of-care quality to gain patient satisfaction. Thus, the hypothesis to be tested for each
quality component is Patient satisfaction components have a negative association with the
process-of-care quality component scores. The quality and HCAHPS component scores were
used for this analysis. All the ve HCAHPS components were used because these are correlated;
otherwise, the regression model would have omitted variable bias. An OLS regression was carried
out for each principal component of quality as the response variable and all the patient
216
217
Research Question 5.
What is the relationship between outcome measures and patient satisfaction, ownership
type and process-of-care quality measures?
This was divided into six sub-questions for each type of outcome covering the three
medical conditions: heart attack, heart failure and pneumonia.
Research hypotheses to be tested. The research hypotheses that will be tested under the
six research questions are:
5.1 and 5.2 The mean outcomes for church group owned hospitals are signicantly less than the
other seven groups.
5.3 and 5.4 Greater patient satisfaction is associated with decreased outcomes.
5.5 and 5.6 Greater process-of-care quality is associated with decreased outcomes.
5.1 and 5.2 Relationships of outcome variables with ownership types.
Two outcome variables tested were:
1. 30-day risk adjusted mortality rate for heart attack, heart failure and pneumonia.
2. 30-day risk adjusted readmission rate for heart attack, heart failure and pneumonia.
The mortality outcome measures in CMS use 30 days from day of admission as the
window for death or survival. These have been adjusted for confounding variables that are beyond
the control of hospitals such as age, gender and sickness of patients when admitted using
sophisticated statistical modeling techniques. (Qualitynet, 2011) Similarly, the three readmission
rates measure hospital specic, risk adjusted all cause 30-day readmission rates for patients
discharged alive. The published rates give percent of patients who were admitted under a medical
condition and died or readmitted within 30 days of being treated.
218
The rst two research questions 5.1 and 5.2 were tested using an OLS regression on
outcomes as response variables and using binary variables to represent 8 ownership groups as
explanatory variables.
Research hypothesis tested was: Church owned hospitals will have signicantly lower
mean outcomes than the other seven hospital ownership groups.
Statistical comparison of mean outcomes by owner groups. The results have been
summarized in Table 57 on page 165 in Chapter 5. The ndings from testing the research
hypotheses under RQ 5.1 and 5.2 are as follows:
The research hypothesis was supported in heart attack mortality rates in three out of the
seven owner groups ( area/authority, local and proprietary). The ES ranged from 0.22 to
0.35 (small to medium). There was no signicant mean difference with the other four
owner groups.
The research hypothesis was supported in heart attack readmission rate in two of the seven
owner groups: federal (medium effect ES =0.52) and local government (small effect ES =
0.23). There was no signicant mean difference with the other ve owner groups.
The research hypothesis was supported in heart failure mortality rates in three of the seven
owner groups: federal (small to medium effect with ES = 0.45), district or authority (small
effect with ES = 0.21), and local government groups (small effect with ES = 0.28). There
was no signicant mean difference with the other four owner groups.
The research hypothesis was supported in heart failure readmission rates in ve of the seven
owner groups: area/authority (small effect with ES = 0.3), local government (small effect
with ES = 0.28), proprietary (small effect with ES = 0.33), nonproft other (small effect with
ES = 0.15) and nonprot (small effect with ES =0.15). There was no signicant mean
difference with the other two owner groups.
219
The research hypothesis was supported in pneumonia mortality rates in ve of the seven
owner groups: area/authority (small effect with ES = 0.35), local government (small effect
with ES = 0.41), proprietary (small effect with ES = 0.18), nonproft-other ( small effect
with ES = 0.16) and nonprot-private (small effect with ES =0.11). There was no
signicant mean difference with the other two owner groups.
The research hypothesis was supported in pneumonia readmission rates in four of the seven
owner groups: area/authority (small effect with ES = 0.03), local government (small effect
with ES = 0.03), proprietary (small effect with ES = 0.15) and nonprot-private (small
effect with ES =0.11). There was no signicant mean difference with the other three owner
groups.
It turned out that mortality and readmission rates in church group run hospitals are
signicantly lower in 22 out of 42 coefcients, and at least not higher than the other groups.
Therefore, it is reasonable to conclude that church group run hospitals on average, have better
outcomes. This supports Charakas, proposition of friendship, compassion, joy in serving and
equanimity is likely to yield better outcomes for patients.
It is highly likely that church group hospitals exercise these qualities higher than other
groups because of their faith and their long history of rendering service to suffering humanity.
This points to changes in government healthcare policy to encourage faith based hospitals in
health care. Also, in medical education, these qualities should be brought to the attention of
students studying for a career in healthcare. Hospitals need to start exploring if these qualities can
be nourished in their staff by modifying their training and compensation policies. The effect sizes
are small to medium. Nevertheless, considering these are mortality and readmission rates, the
associated human cost is extremely high. Human lives are precious, and even small improvements
in outcomes benet the patients.
220
221
Pneumonia mortality rate was negatively associated with poor satisfaction component.
This result did not support the research hypothesis. The effect was between small and
medium (ES (Cohens f 2 ) = 0.035).
Pneumonia readmission rate is positively associated with poor satisfaction component.
This supported the research hypothesis. The effect size is small to medium (ES (Cohens
f 2 ) = 0.03)
The results showed that three of the six regression coefcients were signicant and
positive, supporting the research hypothesis with small to medium effect sizes. that outcome rates
could be improved by improving patient satisfaction. Interestingly, all ndings with readmission
rates supported the research hypothesis. The ndings in mortality rates were not conclusive. One
(heart attack) was not signicant while another heart failure) while contradicting the research
hypothesis was below small effect size. The third one (pneumonia) was signicant but
contradicted the research hypothesis.
5.5 and 5.6 Relationships of outcome variables with process-of-care quality
components.
The research hypothesis is : Greater process-of-care quality is associated with decreased
mortality and readmission rates. This is based on the a priori assumption that if process-of-care
quality is high, mortality and readmission rates would be lower. The relationship between
outcomes and quality were tested using OLS regression. Here, all the ve components of quality
are to be included in the regression and, therefore, transformation was not found necessary.
The process-of-care quality components have been introduced specically by AHRQ and
CMS to improve outcomes for patients. Research hypotheses were tested to verify the
relationships between the quality measures and outcome variables. The two research questions
are tested using six multiple regressions on quality variables as explanatory variables using each
outcome variable as the response variable. The research hypothesis requires the regression
222
coefcients to be negative so that any increase in quality component is associated with a decrease
in outcome. The regression results are given in Table 83 on page 204 in Chapter 5.
From this table, the ndings are:
Heart attack/ failure quality component The research hypothesis was supported in heart
failure readmission rate. The effect was between small and medium (ES (Cohens f 2 ) =
0.027). All other coefcients were not signicant.
Pneumonia quality component The research hypothesis was supported in two outcomes:
heart attack readmission and pneumonia readmission rates. The effect is between low and
medium (ES (Cohens f 2 ) = 0.03). However, the research hypothesis was contradicted in
heart failure mortality rate. It is difcult to explain the increase of 0.12 in mortality rate in
heart failure by a unit increase in pneumonia related quality component. This is probably
due to confounding variables or comorbidity issues.
Surgical care quality component The research hypothesis was supported in heart failure and
pneumonia mortality rates. The effect is between small and medium (ES (Cohens f 2 ) =
0.04 and 0.03, respectively). The research hypothesis was contradicted in heart attack
mortality rate, with an ES (Cohens f 2 ) = 0.026 (small to medium). This quality component
has relatively greater impact on mortality than other components. Hospitals should take
special care about this quality component to improve their outcomes.
Smoking cessation quality component The research hypothesis was supported in heart attack
mortality rate, showing the importance of this advice, particularly to heart patients. The
effect was small to medium (ES (Cohens f 2 ) = 0.026).
Prevention quality component The research hypothesis was not supported. However, the
coefcient was positively associated with heart failure readmission rate, contradicting the
research hypothesis. The effect is small to medium (ES (Cohens f 2 ) = 0.0.027). This
nding probably means that patients getting better informed during post-care and seek
readmission more often.
223
224
does not provide any indication to hospitals on the areas to concentrate to improve their patient
satisfaction. Also, validation is required to test if this question constitutes a principal dimension
of satisfaction, Even if this is a principal dimension, the contributions of the different questions in
the HCAHPS survey to this dimension are not known.
Therefore, a PCA on HCAHPS survey data was carried out in the present study to identify
the main dimensions to provide guidelines to hospitals.
This is vindicated by the new Value Based Purchasing (VBP) system adopted by the
federal government for all Medicare payments. Under this system, bonuses will be paid to
hospitals on the basis of their HCAHPS and other quality ratings; But while assessing the
HCAHPS ratings, the question on willingness to recommend would be ignored (CMS, 2011).
This study found that the overall rating and willingness to recommend account for only 6% of the
total variance. This study was also able to pinpoint the areas on which hospitals need to focus to
improve the patient satisfaction.
Research Question 2.
What is the relationship between hospital ownership and patient satisfaction?
After Hansmann (1980) argued that any differences between ownership types should
vanish under managed care, there have been several studies on the effect of ownership on
satisfaction, quality and outcomes. Among them, only a few focused on patient satisfaction. No
denitive conclusions have emerged, and conicting results have been reported. Some of these
studies are summarized as follows:
Baker et al. (2000) conducted a meta-study on healthcare research literature and found six
studies that reviewed on patient satisfaction. Among them, ownership differences are
conrmed in one study, suggested in a second, and are inconclusive in a third.
Pink (2003) did not nd any evidence to suggest that hospital location (rural versus urban)
or religious afliation contributed to reports of patient satisfaction in any way not explained
by the other measures.
225
Casey and Davidson (2010) used a subset of HCAHPS data for small, rural hospitals with
25 or fewer beds and reported that for-prot status has a signicant negative effect on all
HCAHPS measures except one (whether the patient room was quiet at night). Hospital
inpatient volume has a signicant negative effect on all HCAHPS measures. In this study,
the scores on each question are tested individually. Lehrman et al. (2010) used HCAHPS
data and reported that private nonprot hospitals are more likely to have superior
performance on both quality and patient satisfaction than government and for prot
hospitals. They constructed summary measures for patient satisfaction, by averaging
percentile mean scores for the HCAHPS items.
Using the fraction of patients who rated the hospital in the highest category as the primary
indicator of patient satisfaction and categorizing hospitals into quartiles of HCAHPS
ratings, Jha et al. (2008) found that the performance of for-prot hospitals was worse than
that of private and public not-for-prot hospitals in all areas.
In contrast, the present study used a PCA and constructed component scores for hospitals
after weeding out outliers and unreliable data. The study found that out of ve satisfaction
components, hospitals owned by churches came out signicantly higher than other 7 owner
groups as was predicted by the research hypothesis. In poor satisfaction component, church
group hospitals had statistically signicant lower scores than other groups except two groups
(proprietary and state government).
Research Question 3.
What is the relationship between hospital ownership and process-of-care quality
measures?
Several studies have considered patient satisfaction and process-of-care quality data from
CMS and studied their relationship. Some of these are summarized in this section.
226
Jha, Li, Orav, and Epstein (2005) analyzed HQA quality data and found that characteristics
associated with small but signicant increases in performance included being an academic
hospital, being in the Northeast or Midwest, and being a not-for-prot hospital. Ten quality
measures were used in the study, and unweighted mean performance scores were reported.
Chi-square tests and analysis of variance were used to compare the characteristics of
hospitals that reported no data to the HQA with those that reported some data for every
measure, but that were based on discharge data for fewer than 25 patients and those that
reported adequate data for at least one measure. In addition, t-tests with unequal variance
were used to compare performance measures between hospitals with adequate sample sizes
and hospitals with inadequate sample sizes.
Goldman and Dudley (2008) carried out a cross-sectional analysis of hospitals participating
in Hospital Compare in 2005, evaluating percent adherence to guidelines for 10
processes-of-care for acute myocardial infarction (AMI), heart failure (HF), and
community-acquired pneumonia (CAP) using multiple linear regression analyses and
concluded that participating rural hospitals had lower performance than their urban
counterparts, but, difference across ownership groups was not studied.
(Eggleston et al., 2008) carried out a systematic review of 31 observational studies since
1990 that used multivariate analysis to examine the relationship between quality of care and
ownership and found that ownership does appear to be systematically related to differences
in quality. However, the ndings are conicting, and they concluded that the true effect of
ownership appears to depend on the institutional context, including differences across
regions, markets and time. However, all the studies concentrated on outcome variables and
not process-of-care quality measures.
(Lehrman et al., 2010) jointly examined patient satisfaction and clinical care measures from
2,583 hospitals based on CMS data. The authors used multinomial logistic regression to
identify key characteristics of hospitals that perform in the top quartile on both, either, and
227
neither dimension of quality. Top performers on both quality measures tend to be small
(100 beds), large (> 200 beds) and rural, located in the New England or West North Central
Census divisions, and nonprot. Top performers in patient satisfaction only are most often
small and rural, located in the East South Central division, and government owned. Top
performers in clinical care only are most often medium to large and urban, located in the
West North Central division, and nongovernment owned. They used summary measures for
HCAHPS and HQA scores
In contrast, the present study has identied ve quality dimensions from a PCA and
orthogonal varimax rotation and used components scores on each quality component for group
mean comparison. It was found that the process-of-care quality is likely to be generally higher in
church hospitals than many other owner groups (particularly, government owned hospitals)
conrming the research hypothesis. However, the evidence is not overwhelming in favor of the
church group.
Research Question 4.
What is the relationship between process-of-care quality measures and patient
satisfaction?
Patient satisfaction is usually associated with hospital quality and, therefore, empirical
research on this research question is limited. Some of these are summarized in this section.
(Jha et al., 2008) used HCAHPS satisfaction scores and HQA process-of-care quality
measures and found that hospitals with a high level of patient satisfaction provided clinical
care that was somewhat higher in quality for all conditions examined. They used the
fraction of patients who rated the hospital in the highest category as the primary indicator of
patient satisfaction. They categorized all hospitals into quartiles of HCAHPS ratings and
examined the mean score for clinical quality within each quartile, using a test for trend to
determine whether a higher rating on the HCAHPS survey was associated with better
clinical HQA scores.
228
(Casey & Davidson, 2010) used a subset of HCAHPS scores and process-of-care quality
data for small rural hospitals with 25 or fewer beds. They found that heart failure and
pneumonia process-of-care measures have a statistically signicant impact on the HCAHPS
overall rating and recommendation scores but not the heart attack measures. The effect
sizes for pneumonia and heart failure scores were reasonably strong. The aggregate
process-of-care quality score which combines heart attack, heart failure and pneumonia
measures has a larger and more signicant impact on HCAHPS scores, but with a smaller
effect size.
Glickman et al. (2010) used clinical data on 6467 patients from 2001 to 2006 and Press
Ganey satisfaction ratings in the same hospitals and found that Patient satisfaction was
positively correlated with 13 of 14 acute myocardial infarction performance measures.
They concluded that higher patient satisfaction is associated with improved guideline
adherence and lower inpatient mortality rates, suggesting that patients are good
discriminators of the quality of care they receive.
(Heuer, 2004) carried out a retrospective study about the relationship between the process
quality measures and patient satisfaction ratings. A total of 41 acute care, 200-plus bed,
not-for-prot hospitals in New Jersey and eastern Pennsylvania were included. He used
accreditation data for quality and Press Ganey surveys for patient satisfaction. The results
revealed no relationship. Patient satisfaction by itself is considered as one aspect of clinical
quality. In view of difculties in dening healthcare quality, the process of care quality is
considered as an important part of clinical quality that is measurable. Even though, patients
may lack medical knowledge, yet, many studies have shown that patients can instinctively
judge the process quality and hence more satised when they get appropriate treatment.
However, the increased stress on patient satisfaction by hospitals, doctors and nurses
might be coerced into giving patients whatever they want, regardless of medical appropriateness.
Several cases have come to the researchers notice where physicians prescribed antibiotics and
229
sometimes even strong pain killers for the sake of satisfying patients, ignoring the American
Medical Association (AMA) guidelines. This was the rationale behind this research hypothesis.
The study found that in two of the satisfaction components, overall and post-care, the research
hypothesis was supported. However, in other components, the evidence is mixed.
Research Question 5.
What is the relationship between outcome measures and patient satisfaction, ownership
type and process-of-care quality measures?
This was divided into six sub-questions for each type of outcome covering medical
conditions: heart attack, heart failure and pneumonia.
5.1 and 5.2 Relationships of outcome variables with ownership types.
The relationship between hospital ownership and mortality has been consistently studied,
but a denite relationship has not been established. Some of these are summarized in this section.
In one of the earliest studies on this subject, (E. F. X. Shortell Stephen M. & Hughes, 1988)
examined the inuence of hospital ownership on mortality rates among inpatients receiving
care under Medicare for 16 selected clinical conditions that were studied as a group. Data
were obtained from the records of 214,839 patients who received care in 981 hospitals in 45
states, from July 1, 1983 through June 30, 1984. They did not nd statistically a signicant
association between mortality rates among and ownership. Several later studies reported
mixed results.
(Sloan et al., 2003) used data for 129,092 Medicare patients with AMI admitted from 1994
to 1995 and found that mortality may not vary by ownership. Using the Linear Structural
Relations model to analyze 1984 Medicare data for 239 hospitals, mortality rates were
found to be higher in for-prot hospitals (Al-Haider, 1990). A later study used 1988
mortality data for 3,782 hospitals representing all ownership types to assess short-term and
230
longterm adjusted mortality Kuhn1994MC:. Osteopathic and public hospitals had the
highest unadjusted mortality rates and private teaching hospitals had the lowest rates.
Baker et al. (2000) reviewed seven studies and found that hospital ownership was related to
mortality in ve studies and inconclusive or not related in two studies.
In a retrospective cohort study, Yoshikawa, Lai Si, and McGuire (2007) used data on 16.9
million hospitalized Medicare beneciaries for a 10-year period from 1984 to 1993. 5,127
acute-care hospitals in the United States were grouped into 6 mutually exclusive hospital
types based on teaching status and nancial structure ( for-prot [FP], not-for-prot [NFP],
osteopathic [OSTEO], public [PUB], teaching not-for-prot [TNFP], and teaching public
[TPUB]). Logistic and linear regression methods were used to examine the risk-adjusted
30-day and 6-month mortality rates. Patients at TNFP hospitals had signicantly lower
risk-adjusted 30-day mortality rates than the patients at other types of hospitals. They
concluded that, in the risk-adjusted 30-day mortality, TNFP hospitals had an overall better
performance than other hospital types.
Eggleston et al. (2008) carried out a meta-study of 31 observational studies that examined
the outcomes at US hospitals from 1984 to 2001, and found that 72% of them, using
Medicare claims, found signicantly higher mortality rates in government owned hospitals
compared to private nonprot hospitals. Nevertheless, they concluded that the true effect of
ownership appeared to depend on the institutional context including differences across
regions, markets and over time.
In contrast, the current study used the 30-day mortality and readmission rates for
Medicare patients (aged 65 and above) published by CMS for all US hospitals and evaluated the
outcomes for church group run hospitals with other groups by ANOVA and found they have
signicantly lower outcomes than other groups.
231
232
readmission rates. The authors categorized hospitals into quartiles on the basis of quality
scores, compared the predicted mortality rates (95% condence interval)and found
signicant differences in mortality rates across these categories with p-values 0.005. They
also calculated the potential number of deaths avoided if patients who were treated in the
lowest quartile had been treated at the highest quartile could be 2,200.
The last two studies seem to be the only ones that used CMS data. This line of research is
important because the inverse relationship between high HQA performance and lower
risk-adjusted mortality is an important validation for the HQA national hospital quality rating
program. The same indicators have also been adopted for accreditation of hospitals.
Instead of studying the difference across quartiles, the current research studied the
signicant differences using OLS regression with outcomes as response variables and quality
components as explanatory variables. The ndings show signicant differences in mortality rates
in all the three conditions of heart attack, heart failure and pneumonia, due to differences in
surgical care quality. Other quality indicators have no signicant effect on mortality rates.
Readmission rates are signicantly affected by pneumonia quality in heart attack and pneumonia
conditions. Heart attack/failure quality does impact on heart failure and this establishes the need
for imposing this quality specication.
233
patient satisfaction than the other seven owner groups. This study found that church owned
hospitals scored signicantly better than two other ownership groups. This study brought to
focus on the desirable qualities in healthcare professionals friendship, compassion, joy in
serving, and equanimity, enunciated by Charaka. Hospitals trying to improve their patient
satisfaction scores, are advised to develop these qualities among their staff.
3. A similar research hypothesis was tested with process-of-care quality indicators. This study
found some empirical evidence to support the research hypothesis. Previously, no other
study has attempted a PCA on process-of-care quality measures.
4. Previously, some researchers studied the relationship between satisfaction and quality.
However, these have used either a subset of HCAHPS data or Press Ganey satisfaction
scores. Some of these studies concluded that patients were conscious of quality care, and
increased satisfaction is driven by clinical quality. Nevertheless, several healthcare
personnel have been expressing concern that insistence on patient satisfaction may coerce
doctors to provide the patients whatever they want, ignoring medical appropriateness. The
present study tested a research hypothesis that patient satisfaction components had negative
association with quality components, using the component scores from HCAHPS and
quality data and found some evidence.
5. Differences in mortality rates across ownership groups have been studied. Some studies
reported that private nonprot hospitals have lower mortality than government hospitals.
However, some other studies conicted with this nding. These studies did not use all HHS
data and also not all ownership groups. However, this study used all the HHS outcome data
and found convincing empirical evidence. This again brought out the importance of the
desirable qualities advocated by Charaka for healthcare professionals.
6. This study was the rst to study the relationship between outcome variables and patient
satisfaction. This was not studied previously. This study tested the hypothesis that greater
patient satisfaction is associated with decreased outcomes and found supporting empirical
234
evidence for readmission rates. However, the evidence was mixed in the case of mortality
rates.
7. Testing the relationship between process-of-care quality indicators and outcomes is
important to validate the quality indicators. Two studies using CMS data have been
reported on mortality rates. Both grouped hospitals into quartiles of quality measures and
compared mortality rates across groups. Both studies reported that signicant difference in
mortality between the quartiles. Readmission rates were not covered. In contrast, this study
used both mortality and readmission rates and found some supporting evidence.
235
Threats to validity.
Validity issues were discussed in detail, in the Introduction. The threats to validity are
re-examined in the light of empirical results. This research falls under nonexperimental design
because presumed cause and effects are identied and measured, but other structural features of
experiment and design elements are missing in the design such as random assignment, pretest and
control groups. Instead, reliance is placed on measuring alternative explanations and statistically
controlling for them. Some issues in such designs are inherent and are discussed here relative to
the inferences using the topology described in (Shadish et al., 2002).
Construct Validity The patient responses to HCAHPS questionnaires have been generalized to
the higher order construct patient satisfaction. There are several other scales and
approaches to measure this construct. The approach adopted by this study is justied
because of the nancial implications to hospitals under MMA. The governments Medicare
payment policy is linked to the HCAHPS scores and, therefore, the study denes the broad
concept of Patient satisfaction as what was measured by HCAHPS scores. After the
empirical results, it seems more appropriate to respecify patient satisfaction to something
less broad and more appropriate such as patient response to HCAHPS. This became
apparent when the study found that the rst principal component accounting for 57% of the
variance was identied as Poor satisfaction, thus representing a lack of satisfaction.
Similarly, the process-of-care quality components were only a small subset of many AHRQ
parameters (> 400), and so, it seems more appropriate to respecify process-of-care quality
as HQA measures. Inadequate explication of constructs may lead to incorrect inferences
about the relationship between the response and explanatory variables (Shadish et al., 2002,
p. 73). However, the present terminology is retained to be consistent with the already
published research.
External validity It is difcult to extend the studys inferences over variations in persons,
settings, treatment and measurement variables because of unknown interaction of the causal
236
relationships (Shadish et al., 2002, p. 87). For example if more quality elements are
included or more dimensions are added to the HCAHPS questionnaire or to the HQA
quality measures, a new set of similar analyses is required with the new data.
Statistical conclusion validity Various possible statistical errors such as low statistical power
and violated assumptions of statistical tests are listed by Shadish et al. (2002, p. 45).
Besides the statistical validity issues covered in the Chapter 1 under this, other threats were
found during statistical analysis and appropriate corrections were applied. These are:
1. Before a PCA, sampling adequacy was tested with Kaiser-Meyer-Olkin statistic for
sampling adequacy in PCA (in statistics) (KMO) statistic.
2. Before a PCA, multivariate outliers were found using Mahalanobis method and were
dropped from the sample.
3. Homogeneity of units was ensured by using standardized scores for PCA.
4. The problem with missing data was encountered in quality data but no imputation
techniques were applied. The sample size was deemed adequate. This would be a
delimitation on the study.
5. Non-normality of the data was encountered. Detailed analysis was made both before
and after regressions and the tests were justied after visual inspection of plots and
considering sample sizes using the central limit theorem.
6. Whenever, heteroscedasticity was encountered, regression with robust standard errors
was used.
7. Linearity is a Gauss-Markov assumption to make the regression estimates Best Linear
Unbiased Estimator (BLUE). Whenever, there was a doubt about linearity, further
testing was done using Box-Tidwell method and appropriate visual plots to check the
extent of nonlinearity.
8. Detailed sensitivity analysis was done after every regression to make sure that the
results are not caused by inuential outliers.
237
9. Statistical power and effect sizes were obtained for each statistical test to ensure
statistical validity.
Internal validity threats The possible threats to internal validity are given by Shadish et al.
(2002, p. 55). Among them, the following were found applicable for this study:
Ambiguity of temporal precedence While quality and satisfaction were collected at the
same period (July, 2008 to June, 2009), mortality and readmission measure data for
the period July, 2005 to June, 2008. Therefore, a threat to internal validity arises if a
causal relationship is hypothesized between satisfaction and outcomes or quality and
outcomes because outcomes occurred before quality and satisfaction data.
Consequently, the relationships found empirically in research hypotheses 5.3, 5.4, 5.5
and 5.6 should not be considered as cause and effect relationships. Therefore, the
regression coefcients were not used for predicting the outcomes.
Selection While the satisfaction and quality data concern all patients, the outcome data
is only for Medicare patients and hence covers only patients above 65. This could
cause systematic differences over conditions in respondent characteristics that could
have caused the observed effects. (Shadish et al., 2002, p. 55) While relating the
outcomes with quality and satisfaction, the study assumed that there is no difference
between Medicare patients and general patient population in the process-of-care
quality or patient satisfaction. The nding reported by Needleman (2003) for
medical patients, Medicare rates were consistently higher than all-patient rates, but the
two were highly correlated supports this assumption. This possible selection threat is
a delimitation to the study.
238
Hospitals need to improve the communication skills of doctors, nurses and all other staff
with patients to improve their patient satisfaction. This can be done through training. The
errant staff may be identied and counseled if needed. Hospitals may need to implement
this in their HR procedures such as hiring and compensation policy.
Pain management is important for patient satisfaction and hospitals may need to incorporate
into their protocol.
Explaining about medicines before giving them to patients is important, and training may
be required for their pharmacists.
Patients expect to receive help when they need it, and this needs all round effort from
hospitals and may require layout changes if necessary.
Above all, hospitals need to impart the qualities friendship, compassion, joy in serving,
and equanimity advocated by Maharishi Vedic science. This is likely to improve their
patients satisfaction, process-of-care quality and outcomes. These can be improved by
communicating them to the staff, encouraging such behavior by suitable rewards, looking
for these qualities during hiring and by setting an example by following the same toward
their own employees. The cost of implementing these qualities is low but the likely rewards
are enormous.
239
and the effect of various initiative for performance improvement launched by various
agencies like the Joint Commission and HQA.
The outcome rates used in the study are from Medicare patients only. If these outcomes are
collected for all patients, that would add strength to this study, but collecting the data could
be an expensive task.
Conclusions
The ndings of the study expanded the work of previous studies by identifying the
components of patient satisfaction in terms of the variance accounted for instead of taking the
overall rating as the sole indicator. This helped to identify steps for hospitals in performance
improvement activities. This investigation also revealed the difference among hospital ownership
types in satisfaction and interpreted the differences in terms of friendship, compassion, joy in
serving and equanimity emphasized by Maharishi Vedic Science. Though a few authors have
brought out the importance of empathy in medical practitioners in healthcare, this was not backed
with empirical evidence. This was again brought out in the differences in the quality of care, and
patient outcomes. The ndings highlight the need for incorporating these qualities in medical
education and hospital policies.
240
Chapter 7
Healthcare in light of Maharishi Vedic Science and Maharishis Vedic approach to Total
Health.
Chapter Overview
The purpose of this chapter is to present a perspective from Maharishi Vedic Science
regarding total health which is most nourishing to life and the ideal quality of healthcare services.
This chapter is divided into six sections.
1. In the rst section, an introduction is given to Maharishi Vedic Science and how Maharishis
teachings successfully integrate subjective elements into the objective science of healthcare.
The Science of Creative Intelligence, created by Maharishi Mahesh Yogi, the leading
scientist in the eld of consciousness, highly regarded as a great educator of development of
consciousness. Maharishi did not ignore the positive contributions of objective science and
laid emphasis on combining the subjective and objective elements of knowledge to solve
modern problems. Objective science often ignores the spiritual element and the consequent
relentless pursuit of material wealth, totally ignoring the spiritual, environmental and
ethical aspects has caused many problems for health and wellness of human race.
2. Maharishis approach to health is discussed as an effective system of natural medicine,
aiming to inuence mind, physiology, behavior, and environment to maintain perfect health
and promote longevity. The potential harmful effects and ineffectiveness of modern
medicines in curing or preventing are leading researchers to focus on natural or holistic
approaches to health and wellness. Empirical evidence showing the benecial effects on
241
242
Maharishi Vedic Science - Maharishi has organized the thousands-of-years-old scattered Vedic
Literature into a perfect science and has revealed that Veda and its literature is not a
collection of ancient books or man-made philosophies, but is innite knowledge with
innite organizing power. Its source is the most fundamental level of Natures intelligence,
identied by quantum physicists as the Unied Field of Natural Law, which Maharishi has
brought to light as the Unied Field of Consciousness. This unied approach of enhancing
and enriching objective science with the subjective elements of life is called as Maharishi
Vedic Science.
Science of Creative Intelligence (SCI) Maharishi envisioned SCI as a supplementary course
that could be taught at schools and universities around the world. It would make all elds of
knowledge meaningful, revealing the laws of nature at the basis of math, art, biology, and
every other area of study. By integrating this study into the curriculum, Maharishi wanted
to provide students with the feeling of connectedness and comfort with everything he or she
is learning. The process of learning would become effortless and fullling and the student
gets the opportunity of integrating the eld of study with consciousness.
243
244
(the known), the subject (the knower), and the link between the known and the knower (the
process of knowing). These three elements together create a state of pure knowledge, where
knower, known, and the process of knowing are in the state of self-referral (Maharishi Mahesh
Yogi, 1986).
Objective science emphasizes objective means of gaining knowledge by observing
phenomena, measuring, inferring theories and deducing expected results in similar situations.
This however leads to fragmented knowledge because each eld of study leads to unconnected
bits of knowledge. Subjective elements such as how the eld of study is related other elds of
study and to societal, moral and ethical issues are left out. Maharishi stressed that the knowledge
gained by objective means is not complete until it is enriched with knowledge gained from
subjective means. Maharishi formulated Vedic technologies such as Transcendental Meditation
that facilitate a person to directly experience deeper, more fundamental and universal principles of
intelligence in consciousness (Maharishi Mahesh Yogi, 2001).
Unied Field and Consciousness. The popular notion is Maharishis concept of Unied
Field is derived from the objective science of quantum mechanics. However, Schrodingers
development of quantum mechanics was inspired by Vedic teachings. In his book, Schrodinger
(1992) explains how physical laws are only approximate and scientic measurements have
limited accuracy. He raises the question Is life based on the laws of physics?, and points to an
I (or a Consciousness) that has full control over Nature. He refers to Upanishads teachings on
Brahman, Atman and Maya. Moore (1992) has given the following quotation from Schrodinger,
which he wrote in 1925 before creating his quantum theory.
This life of yours which you are living is not merely a piece of this entire existence,
but is in a certain sense the whole; only this whole is not so constituted that it can
be surveyed in one single glance. This, as we know is what the Brahmins express in
their sacred, mystic formula which is yet really so simple and so clear: tat tvam asi
245
( a ) , this is you. Or, again, in such words as I am in the east and in the
west. I am above and below, I am this entire world
Many objective scientists identied the most fundamental level of Natures functioning as a
unied eld of all the laws of nature. Hagelin (1989) explained how scientists from different
disciplines deduced the existence of a unied eld of all laws of nature. Maharishi revived the
ancient Vedic teaching about a Unied Field (Atma in Sanskrit) of Pure Intelligence and Pure
Consciousness which is the home of all the Laws of Nature (Dharma in Sanskrit. Maharishi
described the Unied Field of Pure Consciousness as Being (Maharishi Mahesh Yogi, 2001).
By experiencing this eld of Pure Consciousness, knowledge of all Laws of Nature can be
obtained and this knowledge is pure and completely reliable, free from doubts, because
knowledge is gained not from any measurements that are subject to doubts and uncertainty, but
from directly experiencing this Field.
Comparison of modern science with Maharishi Vedic Science. Table 84 lists the
differences between modern science and Maharishi Vedic science. This table was adapted from
page 186 of (Maharishi Mahesh Yogi, 1994). Some text has been slightly modied from the
original to suit the general context because Maharishi was writing about University education in
his book.
Table 84
Comparison of modern science with Maharishi Vedic Science.
Modern Science
reality.
. . . Continued on Next Page
246
Modern Science
reality.
practice.
is a dangerous thing.
knowledge.
247
Modern Science
intention.
of fulllment.
Isolates the individual from his environment.
Notes:
1. Source: Page 186 of the book Vedic Knowledge for Everyone by Maharishi Mahesh Yogi.
2. The text has been slightly modied to t the general context; Original text had been written in the
context of University education.
(2005)
Public Policy.
D. W. Orme-Johnson
Applications of Mahar-
The article focuses on the applications of Maharishi Vedic Science to public policy.
problems.
of schools and universities and shows that this approach solves current educational
beck (1987)
in Education:Principles,
This paper adds Maharishi Technology of the Unied Field to the existing curricula
Sidhi programs shows the growth of cosmic psyche on all levels of life.
(1988)
Creation
This article conceptualizes the cosmic psyche as the simplest form of human
D. W. Orme-Johnson
rists perspective.
consistent with all known physical principles, but requires an expanded framework
Showed that the proposed identity between consciousness and the unied eld is
The new paradigm offered a practical program to bring to fruition all the various
Summary
Hagelin (1987)
Cavanaugh (1992)
Economic development
Reference
Title
Table 85
248
Its
Foundation
Mathematics
tions of Effectiveness of
Business Management
Applications of Mahar-
Vedic Science
in Light of Maharishi
From
Gorini (1988)
Schmidt-Wilk (2005)
This article uses Maharishi Vedic Science to explain the subjective relationships
eld through the subjective approach of Vedic Science are brought out.
A new understanding and language of physics based directly on the unied eld is
Hagelin (1989)
Restructuring
Physics
Summary
Reference
Title
249
250
All these published researches including others not listed in this table, have demonstrated that
Maharishi Vedic Science principles, when applied to their respective elds of modern science will
enrich the objective ndings and theories in all levels of life including: psychological,
sociological, ecological, physiological, cosmological and mathematical aspects. Accordingly, in
this dissertation, Maharishi Vedic Science principles were applied to enrich the study ndings.
Several research articles have also been published, applying these principles in their research and
the authors might not be familiar or even aware of Vedic principles.
251
Maharishis approach is based on Natural Law (to differentiate this from the
western concept of natural law, it is spelt with capitals). Since ancient times, several eastern and
western moral views have claimed that there are universal moral principles and norms and
somehow, these are available to all humans, imprinted in their minds from birth. Greek and
Roman philosophers and law-givers considered that while human made laws ought to reect the
principles of natural law and should not try to violate them. The thirteenth century Dominican
Monk St. Thomas Aquinas was the most inuential natural law theorist of all times and laid down
the connections between natural law and human laws (Aquinas, 1990). The US constitution
makers tried to use the principles laid down by Aquinas while drafting the constitution (E.g.)
They adopted Aquinas assertion These things are self-evident: God exists, the manner of His
existence, His knowledge, will and power and tried to imbibe them in their draft. In modern
times, Prof. Budziszewskis denition of Natural Law as the foundational principles of right and
wrong which are both right for all and at some level known to all is taken as the functional
description and is used as a principle to limit enforcement of natural law through human laws.
(Hensler III, 2009)
Maharishi used the term Natural Law in the place of the Vedic term Dharma (! ). This is a
broader concept than the western view of Natural Law. The Sanskrit word Dharma is derived
from the root Dhr-which denotes : upholding, supporting, nourishing and sustaining. The
252
role of Dharma is to support and uphold all aspects of human life and society. Dharma has often
been identied with Truth
Verily, that which is Dharma is truth. Therefore they say of a man who speaks
truth, He speaks the Dharma, or of a man who speaks the Dharma, He speaks the
Truth. Verily, both these things are the same. (From Brihadaranyaka Upanishad
91.4.14)
The Rig Veda (the earliest of the four Vedas) declares that Law and Truth are eternal. The
Vaiseshika sutra of Kanada (one of the six systems of Philosophy and extolled by Maharishi as
enlivening the specifying quality of intelligence and traced to functioning of cerebellum in
human physiology (Nader, 2000)) denes dharma as that from which results true happiness.
This denition is very much valid in healthcare. The concept of dharma includes the responsibility
of upholding it. The popular quotation Dharma protects those who defend it ! " "
Dharmo rakshati rakshitaha from Manu Smrithi (the Vedic book on laws) exemplies this.
In short, Maharishis concept of natural law is associated with truth and happiness and is life
supporting. Maharishi uses this concept to cover all aspects government including education,
health, economics, defense and governance (Maharishi Mahesh Yogi, 1995a).
Ayur-Veda. Maharishis approach is based on Ayur-Veda (a
), an ancient Vedic system
of medicine. The Sanskrit name is translated as the science of life by Maharishi (Maharishi
Mahesh Yogi, 1986). Ayur-Veda, itself was inspired by the teachings of Atharva Veda (a#
), the last of the four Vedas. Many of the 6,000 hymns or verses of Atharva Veda were concerned
with philosophical and spiritual ideas; but at least a thousand verses dealt with human body, its
disorders and their cures.
Adhi Raja Nada Ram has explained that
Atharva Veda is the sum total of all that is pertaining to Samhita with a
predominance of Chhandas observed, or object of observation value. Atharva Veda
represents the totality of the musculo-skeletal system the organs of action. This
253
aspect of the physiology makes the totality of Veda move. This is the value of
reverberating wholeness. (Nader, 2000)
Ayur-Veda has adopted the teachings of Atharva Veda that for a persons healthy living, both
mind and body play a composite role and cannot be isolated from each other. Charaka, one of the
founding fathers of Ayur-Veda, in his monumental work Charaka Samhita conceptualized
healthcare, as a quartet comprising four elements, ideally working together in unison: physician,
medication, helper and patients. Valiathan (2007) has discussed Charakas requirements from
each of these elements and the need for them to act in harmony. Nader (2000) identies Charaka
Samhita as the balancing, holding together and supporting quality of intelligence and associates
it with the physiology of the cell nucleus. This association is apt because the cell nucleus
carefully preserves the balance in the cell, protecting it from unwanted chemicals.
Maharishi Ayur-Veda and Maharishis Vedic approach to Health. Maharishi explains that the
most fundamental eld of life is consciousness and is the basis of all our thinking, action, and
creative efforts. Consciousness, synonymous with awareness, is also described by Maharishi
Vedic Science to be the basis of all the functions of physiology (Maharishi Mahesh Yogi, 1995a,
1995b). Maharishi Ayur-Veda is the revival of the traditional Vedic healthcare system by
Maharishi Mahesh Yogi. The ancient system had been in continuous practice for over ve
thousand years; But consequent to the passage of time and several foreign conquests in India, new
research and development was lacking and much of the original knowledge had been lost, along
with the understanding of how consciousness and physiology were connected. Maharishi brought
the effortless practice of Transcendental Meditation to the world over 50 years ago. Out of
concern for public well-being , Maharishi began to work with eminent Ayur-Vedic physicians and
western physicians in the 1980s to introduce a holistic alternative system of medicine and
healthcare that caters for both body and mind. He was convinced that a fully developed
consciousness is the basis for a perfectly functioning physiology. Maharishi mapped the concept
254
of balance that prevailed in Ayur-Veda to the Unied Field, the home of Natural Law. Maharishi
considers that balance is the natural state of life.
Imbalance arises from loss of the self-referral state and leads to pain and suffering (Maharishi
Mahesh Yogi, 1986). Maharishi recommends restoring the balance in the body through Ayur-Veda
and balance in the mind through the practice of Transcendental Meditation. He emphasizes a
prevention based approach to healthcare. Maharishi feels that A holistic approach to health
through alliance with natural law so that prevention, promotion, and cure are most effective for
health, supported by the evolutionary inuence of the natural law. Alliance with the natural law
for perfect health and reversal of ageing for life to grow in the direction of immortality
(Maharishi Mahesh Yogi, 1995a). By adopting healthy living and healthful practices which are
likely to prevent diseases, Ayur-Veda recommends herb based drugs and formulations and many
of them have been adapted in Maharishis approach to health after due testing and validation.
Healthful eating and dietary foods have also been prescribed. Maharishi always recommends
fresh, organic foods. Maharishi revived the ancient system of Ayur-Veda and improved it
considerably by combining Maharishi Vedic technology of Transcendental Meditation which
results in the experience of pure consciousness, which is accompanied by deep silence , peace and
restfulness. The resulting approach to total health is called as Maharishi Ayur-Veda (MAV) in
honor of Maharishi. For mental balance, Maharishi recommends daily practice of Asanas
(prescribed by Yoga) and Maharishi Transcendental Meditation that will surely lead one to higher
levels of consciousness. Maharishis opinion is that the self-referral consciousness is the natural
state of beings! Even striving for this state leads one to a long, healthy, stress-free and useful life
and this itself is a good enough reward.
Maharishi Ayur-Veda (MAV) Maharishis approach is applicable to all the four constituents of the
medical quartet:
1. Medications and therapies are to be based on Maharishi Ayur-Veda.
2. The other three elements: physicians, patients and attendants by practicing Transcendental
Meditation rise to higher levels of consciousness.
255
3. Physicians and attendants show more friendship, compassion, satisfaction and equanimity
and this improves their attitude to patients and their work.
4. Patients by practicing meditation and preventive practices reduce their need for hospital
visits and this reduces the healthcare system cost to the society.
Empirical evidence for the efcacy of Maharishis Vedic approach to Health. Maharishis
approach to Total Health has already produced measurable and veriable results. Some selected
published articles giving empirical evidence about Maharishis approach to total health are listed
in table 86. This table shows that Maharishis approach to total health has been found to lead to
lower health costs, reduced substance abuse, lower mortality, benecial changes in brain activity,
cardio-vascular function, anxiety level, and blood pressure, improved intelligence, and to better
pain management.
in US.
treatments.
Rainforth (1994)
gram
D. Orme-Johnson (1987)
D. W. Orme-Johnson (1973)
scendental Meditation
(1989)
covered.
Newman,
C.
Alexander,
healthcare.
N.
Gelderloos (1991)
Summary
Reference
Title
Table 86
256
tice:
blood pressure
Pressure
power. Two neural networks that may mediate these effects are
Model of TM Practice
disease.
artery disease
meditation program in the treatin patients with coronary artery disease and may be considered
Summary
Reference
Title
257
Taken
and Prevention
The
Ef-
patient dissatisfaction.
anxiety level.
matic Medicine
level.
M. C. Dillbeck (1977)
(2008)
analysis
Anderson,
Related to mortality.
mortality.
longitudinal study
The regular practice of Transcendental Meditation may have the
intelligence-related measures: A
Transcendental
meditation
providing not only immediate relief from distress but also long-
misuse: a review.
range improvements in well-being.
Orme-
Gelderloos,
Walton,
Summary
Reference
Title
258
Quebec: an update.
Summary
Reference
Title
Cost of healthcare is a serious problem in US.
259
260
261
Figure 54. Unied Field Chart - Blow-up of healthcare portion in the upper left section
262
At the lower level of concretion, a person becomes a patient due to intrinsic or external
causes. Intrinsic causes have been explained by Charaka as over-use, under-use, non-use or
misuse of senses, action or time (Valiathan, 2007). For chronic ailments, this view has been
accepted by modern science as well E.g. Under-use of the limbs could result in partial loss
of exibility. Modern medicine used to ignore the mind-body connection until recently;
But, Charaka stressed the importance of proper use of mind and time for body health.
Wahbeh, Elsas, and Oken (2008) searched for articles on mind-body therapies and
neurologic disease search terms from Medline and Psychinfo databases for clinical trials
and graded reviewed published evidence. They found that there were several conditions
where evidence for mindbody therapies was strong such as migraine headache and back and
neck pain. The therapies included meditation, relaxation, hypnosis, biofeedback, breathing
techniques, yoga, tai chi, and qigong. The desire of patients to engage in mindbody
techniques as an additional therapy to more conventional treatments needs to be recognized,
because they produce physiologic and physiological benets. The scheme would lead to a
better understanding of the core processes taking place in healthcare than modern medicine
which mostly ignores the role of mind-body connection in healthcare.
At this level of concretion, the process of treatment is entrusted to physicians, attendants
and medications, following Charakas medical quartet. Charaka has listed the most
desirable qualities that physicians should possess: knowledge, wisdom, friendship,
compassion, joy in serving, equanimity, and fearlessness. These are shown as chhandas
values at the base level. At this level of abstraction itself, the major shortcomings of
modern healthcare emerge: Unfortunately, these qualities are often ignored by modern
medicine. These qualities are not part of the curriculum of our medical schools, and they do
not even test for these qualities while admitting students. Hospitals do not assess their
employees for any of these qualities and do not arrange any training or include in their
policies. This study has found empirical evidence that hospitals owned by church groups
263
having better satisfaction levels, quality and outcomes and the reasons are attributable to
cherishing these qualities.
Practice of Maharishis Vedic technologies would improve these qualities among the
hospital staff. This is supported by a study by Gelderloos et al. (1991). They carried out a
statistical meta analysis of 24 empirical studies on the benets of Transcendental
Meditation and found that the effect size of TM on overall score of self-actualization was
approximately three times as large as that of other forms of meditation and relaxation.
Maslow, after formulating hierarchy of need, amended his model, placing
self-transcendence as a motivational step beyond self-actualization. Consequences are: (a)
a more comprehensive understanding of world-views regarding the meaning of life; (b)
broader understanding of the motivational roots of altruism, social progress, and wisdom;
(c) a deeper understanding of religious violence; (d) integration of the psychology of
religion and spirituality into the mainstream of psychology; and (e) a more multiculturally
integrated approach to psychological theory (Koltko-Rivera, 2006). Thus, it is reasonable to
conclude that any increment in self-actualization (or the subsequent self-transcendence)
will be accompanied by positive improvements in altruism and wisdom leading to the
development of the four qualities of friendship, compassion, equanimity, and joy in
serving that were stressed as essential by Charaka for hospital staff. Development
psychologist V. K. Alexander (2005) focused on the applications of Maharishi Vedic
Science to developmental psychology and summarized the profound impact of
transcendental experience on the life and behavior of the experiencer, including
improvement in wisdom and outlook.
Second level of concretion - The ancient Ayur-Veda physician has also given a detailed
description of how the problems listed in level 1 cause problems for patients Valiathan
(2007). Diseases could be in-born, exogenous or psychological. He considered patient as an
integral whole of body, mind and self and any problems with one could impact the other.
He also followed the three-in-one structure in his incisive analysis. For healthy life, diet,
264
sleep and sense control are all important and need to be properly done. For treatment,
Charaka included various options: Sacred means, rational approach and at times
psychological on the basis of establishing trusty relationships with patients. Maharishis
approach to total health also included using Vedic sounds and gems in some therapies.
At this level, the physicians need to display comprehension of the patient, pay careful
attention to symptoms, and show patience, humility and carefulness. In contrast, modern
science of healthcare does not show full attention to these details. It does not usually take
such a holistic view but relies on short-term drug based therapies. This study has found
empirical evidence that proper communication with the physicians and nurses is an
important reason for patient dissatisfaction. Studies such as by Wahbeh et al. (2008) are
now beginning to appear on alternative treatment methods and their benets.
Top level of concretion - At the most concrete level, the constituents of an ideal healthcare
system are listed. Patients develop condence in the system. Doctors and healthcare
professional develop balance, righteousness and commitment towards patients. The quality
of the process of treatment improves. Ideal outcomes of a total healthcare if Maharishis
Vedic Technology is adopted in healthcare are as follows: good patient satisfaction, quality
healthcare at affordable cost, and improved mortality. This study has found some empirical
evidence for improved mortality in hospitals owned by church groups. Many studies such
as (C. N. Alexander et al., 1989; D. W. Orme-Johnson, 2000; Schneider et al., 1997;
Zamarra et al., 1996) have provided empirical evidence for improvement in longevity by
practicing of Maharishis Vedic Technologies. Furthermore, several studies have provided
empirical evidence for improvements in health by practicing Transcendental Meditation.
Herron and Hillis (2000) have shown a reduction in government payments to physicians for
insurance enrollees after starting with Transcendental Meditation.
These studies are listed in table 86. On the basis of this evidence, Herron (2008) provided
clear insights into the sources of dysfunction in US healthcare system and practical policy
guidance about how the system can be made to work much more efciently and effectively.
265
Herrons evidence based advice is to increase the relative amount of health care resources
devoted to prevention is especially important. He offers innovative approaches to preventive
care, drawn from what has become a signicant body of research and experience with
Maharishis approach to Total Health.
Connecting to Unied Field using Maharishi Vedic Technologies. Figure 55 is the upper right
section of the UFC and represents the process of transcending the thinking level in waking state
of consciousness to experience the Unied Field of Natural Law through Maharishis Vedic
technologies: Transcendental Meditation program and TM-Sidhi programs. The top level in this
diagram represents the most concrete form of life experienced in the waking state of
consciousness. With practice of Maharishi Vedic technology programs, the consciousness ascends
to higher levels resulting in gradually contacting the Transcendental consciousness deep within
and later experiencing the Unied Field thereafter after the mental blocks and stresses vanish.
This shows that both patients and healthcare professionals by practicing TM and TM-Sidhi
programs, can experience the Pure Intelligence at the Unied Field in Transcendental
Consciousness (TC). By continual contacting this, the mind gets freed from mental blocks and
stresses. The person will gradually begin absorbing the forty qualities of the Unied Field in his
waking state of consciousness (Maharishi Mahesh Yogi, 2001). These qualities include one of
balancing, holding together and supporting quality of intelligence represented by Charaka
Samhita and this is a basic requirement for healthcare. Maharishi has given an analogy of how a
cloth frequently dipped in a vat of dye starts absorbing the color of the dye into itself. Similarly,
people tend to absorb these qualities and start showing these qualities in their waking state of
consciousness. People tend to lead longer, healthier, stress-free and useful loves while healthcare
professionals tend to provide dedicated service to their patients who get diseases from extrinsic
causes.
Connecting to Unied Field using Maharishi Vedic Technologies. Figure 56, the lower left
section of the UFC shows that the Unied Field of Natural Law is the source of all natural laws.
266
267
268
269
R
. icho Ak-kshare Chart (RAC)
Signicance of R
. icho Ak-kshare Chart (RAC). The previous section of this chapter provided a
broad structure of the eld of healthcare as a whole through the illustration and interpretation of
the Unied Field Chart. This present section further examines how the eld of healthcare unfolds
from within the innite Unied Field of Natural Law by demonstrating the knowledge contained
in the R.icho Ak-kshare verse (Mandala 1, hymn 164, verse 39) of R.k-Veda ($%
) . The
R.icho Ak-kshare verse locates the source of all the laws of nature and the fundamental level of all
the creation in the transcendental eld of consciousness, the Unied Field, and explains how the
knowledge and experience of the Unied Field completes human life. In keeping with Vedic
traditions, the verse is given in Sanskrit with the accent marks, followed by its transliteration in
English.
$& a"
'
()
a! v
+
.&
%/ ittdv) i
)nn
%
Richo ak-kshare parame vyoman yasmin Deva adhi vishwe nisheduh
Yastanna veda Kim richa Karishyati ya it tad vidus ta ime samasate
Jones (1989) gives Maharishis translation from Page 101 of a publication of Maharishi Vedic
University in 1985 as:
The verses of the Ved exist in the collapse of fullness (the Kshara of A) in the
transcendental eld, In which reside all the devas, the impulses of creative
intelligence, the laws of nature responsible for the whole manifest universe.
He whose awareness is not open to this eld, what can the verses accomplish for
him? Those who know this level of reality are established in evenness, wholeness of
life.
(Maharishi Mahesh Yogi, 1994)s commentary on this verse is important because from his
incisive analysis, he has made several remarkable discoveries :
270
The R.ichas (verses) of the R.k Veda Laws of Nature, structures of intelligence, frequencies
of consciousness are sustained in the Ak-kshar the dynamics of a (A), the dynamics of
totality, the kshar, the collapse of a% (Ak) the collapse of innity, represented by a (A),
onto its own point, represented by % (K)a% (Ak), the collapse of the unbounded eld of
intelligence onto its own point.
Maharishis discoveries from this verse are as follows:
The R.icho Ak-kshare verse shows the source of all the R.ichas, the expressions of the Veda,
in Transcendental Consciousness. Therefore he concluded that Knowledge is structured in
Consciousness.
All the theories of all disciplines of modern science are contained in the structure of the
Veda; Therefore, all theories of objective science are contained in this one verse of
R.k-Veda. By applying this distinctly profound knowledge, a R.icho Ak-kshare Chart can be
utilized to illustrate the fundamental principles, development, and applications of any eld
of study in Objective Science.
The second half of the verse begins with the interrogative, yastanna veda kim richa
karishyati He whose awareness is not open to this eld, what can the verses accomplish
for him? Although one may know about the existence of the Unied Field of Natural Law,
one cannot gain from the knowledge if one does not have the experience of pure
consciousness . For this, Maharishi designed Vedic technologies such as TM and
TM-Sidhi programs; a natural and easy way to personally experience the Unied Field of
Natural Law.
The last part of the verse ya it tad vidus ta ime samasate Those who experience this
level of reality are established in evenness and wholeness of life. When awareness is
permanently established in the eld of pure consciousness, ones thoughts and actions
spontaneously align with the totality of Natural Law; thereby, gaining the full support of
271
Natural Law and experience of supreme stability and exibility . This exposes the
weakness of objective science based education because it does not give an opportunity to
the students and faculty to experience the Unied Field and therefore are likely to miss the
balance in life. This encouraged Maharishi to establish education based on Vedic Science
(Maharishi Mahesh Yogi, 1994).
There is independent corroborative evidence of this from publications by researchers who are
outside Maharishis TM movement. For example Dienes (2008) quotes the R.icho Ak-kshare verse
citing from (Maharishi Mahesh Yogi, 1994), showing that nothing new has been discovered, since
there is nothing new under the Sun! It is instructive to note that objective science reaches a
dead-end after using all their research tools.
In accordance with Maharishis great discovery that Knowledge is structured in Consciousness,
an attempt has been made to present the structure of the eld of healthcare schematically and is
seen to follow the structure of the
R.icho Ak-kshare verse, in the form of a R.icho Ak-kshare Chart (RAC).
R
. icho Ak-kshare Chart (RAC) for healthcare eld. The aim is to show eld of healthcare
follows the structure of the Vedic verse. Figure 58 shows the complete RAC. A brief description
of the individual boxes in the RAC follow:
The rst upper box (Richo Akshare) explains that in the foundational principles of
performance improvement in hospitals (Richas) emerge from the dynamics of interaction of
fullness and point collapse of (kshara of A) public desire for long and healthy life. This
process starts with the many hospitals and service options (A) are available to public.
Through hospital performance in satisfaction, quality, and outcomes, public make a choice
(K).
The second upper box (Parame Vyoman) brings out the transcendental nature of the
fullness. Hospitals need to understand properly what the patients ultimately want ( a long
272
PARAME VYOMAN
...in the transcendental
field, self-referral
consciousness,
the Self
US hospitals need to
offer quality service
to meet the public
desire for long and
healthy life
(collapse of A) and
this requires
Principles of
performance
improvement
(richas)
Which need to be
properly understood
and implemented in
their organizations
and also making
sure that patients are
guided in making
healthy choices in
their lifestyles.
Knowledge about
what factors affect
patient satisfaction,
process of care
quality, mortality
and improving the
quality of the
healthcare staff
YASTANNA VEDA
He whose awareness
is not open to this
field...
KIMRICHA
KARISHYATI
...what can the
verses accomplish for
him?
YA ITTADVIDUS
Those who
know this level
of reality...
Hospitals lacking
this knowledge will
not able to improve
their performance
and give satisfactory
and quality
healthcare service to
the patients
YASMIN DEVA
In which reside all the
devas, the impulses of
creative intelligence, the
Laws of Nature
ADHIVISHVE
NISHEDUH
...responsible for the whole
manifest universe
this knowledge
enables hospitals to
improve their hiring
and training
policies, fine-tune
their protocols and
to strengthen their
review and survey
procedures and to
improve patient
counseling
procedures.
TA IME SAMASATE
...are established in
evenness, wholeness
of life.
273
and health life) and orient their organizations accordingly by proper use of performance
improvement principles.
The third upper box (Yasmin Deva) brings out the organizing power of the foundational
principles, showing that understanding the factors inuencing patient satisfaction, quality
and outcomes stimulates the intentions or impulses (Devas) in hospitals for improving their
performance. The rst research question attempted to identify the principal components of
patient satisfaction by carrying out a Principal Component Analysis PCA and found ve
components accounting for most of the variance. The second research question did a
similar analysis on process of care quality data and identied ve principal components.
The other three research questions probed into the relationships between satisfaction,
quality, outcomes and hospital ownership type and found signicant relationships. (Details
are given in Chapter 4) The results are presented in the fourth upper box (Adhivishve
Nisheduh). Details are given in Chapters 4 and 5. To improve their patient satisfaction and
quality ratings, hospitals need to identify steps for improving their performance on the
identied principal components. There is a signicant relationship between hospital
ownership type and satisfaction, quality and ownerships. Valiathan (2007) has shown that
the ancient Vedic Physician Charaka advocated that the healthcare staff should have
friendship, compassion, joy of serving and equanimity while dealing with patients and this
was the difference between hospitals owned by church groups and other ownership types.
Other hospitals should try to reect this in their hiring and training policies and in their
mission objectives to promote these desirable qualities in their healthcare staff. Hospitals
should realize that the public desire for long and healthy life is the aspect of fullness (A)
and its collapse (kshara) is the healthcare service they offer to patients and keeping this as
their base objective. To meet this, hospitals need to offer counseling to patients on adopting
healthy life styles.
The rst lower box (Yastanna Veda) points to ignorance and describes the case of
hospitals not having sufcient knowledge that primarily, patients seek a long and health life
274
(A) and its collapse is only the healthcare service being offered to them. They are not
aware of the factors behind patient satisfaction and quality. They are also not aware of the
relationships between satisfaction, quality, outcomes and their service.
The second lower box (Kimricha Karishyati) explains the practical output from their
ignorance. Performance improvement is not pursued properly and these hospitals lose their
market share and government payments. Patients do not fulll their wishes for a long,
healthy life.
The third lower box (Ya Ittadvidus) points to knowledge and states hospitals which
become aware of the pressing need of performance improvement and the basic expectations
of their patients .
The last lower box (Ta Ime Samasate) presents the practical consequences of this
knowledge. Such hospitals pursue performance improvement with vigor. They achieve
improved patient satisfaction, better quality and improved outcomes. They attract more
patients, more government payments and the improved revenues satisfy their stake holders.
Patients also get longer and healthier lives.
By applying the R.icho Ak-kshare verse to the subject of Improving Healthcare services in
hospitals, this study has shown that performance improvement in hospitals involves a deeper level
of knowledge about patient satisfaction, quality and outcomes than a supercial level of blindly
applying some techniques. The results of analysis presented in the Chapter IV show that
knowledge about the factors contributing patient satisfaction and quality helps hospitals to
identify the areas to focus on. The signicant relationships between satisfaction, quality,
outcomes and ownership types bring out the importance of developing qualities such as
friendship, compassion, joy of serving and equanimity among the healthcare staff. The fact that
healthcare service should be motivated by the patients desire for long and healthy life points to
the need for hospitals to use this as their mission goal. Thus all the principles of healthcare
services are contained in one Vedic verse.
275
Conclusion
Problems in US healthcare. US healthcare services are facing a crisis of condence. The cost
is high. As shown in Chapter 1, total healthcare spending is about 16.2% of GDP in 2007 and the
per capita spending on healthcare is the highest in the world. At the same time, patients are not
satised with their treatment; As described in Chapter-4, only 65% gave the top rating to their
hospital. The process of care quality rating showed the need for improvement as discussed in
Chapter-4. Mortality gures do not show US among the top ranking developed countries in the
order of life expectancy. Schneider et al. (1997) has listed other problems in US healthcare:
About 40% of the US population suffers from chronic diseases. Western healthcare has
been remarkably ineffective against such diseases.
Very little proportion of the total budget (about 1%)is being spent in 1992 on preventive
measures while 99% is spent for acute and chronic care of illnesses after they occur). A
more recent article (Woolf, 2007) conrmed this and gave an example that regularly
offering smoking cessation counseling would save society an estimated 1.3
quality-adjusted-life-years. Woolf (2007) also reports that most major chronic diseases are
amenable to prevention and an estimated 38% of US deaths are attributable to 4 behaviors :
smoking, poor diet, physical inactivity and alcohol use. Woolf (2007) estimated that
prevention accounts for only 2% to 3% of the total health spending, though strong case has
been shown in published research to invest decisively in prevention, perhaps more than in
treatment.
Citing many articles, Schneider et al. (1997) showed that adverse side effects of modern
drugs could be adding to healthcare cost and problems.
A case for a new paradigm in Healthcare was advocated. The new approach will stress on
preventive care and can make use of Maharishi Ayur-Veda to manage chronic ailments and
to improve mental conditioning. Besides therapeutic benets, Maharishis Vedic
276
technologies will also improve the attitudes healthcare staff towards patients, patients
lifestyles and longevity. The high cost of healthcare in US will also be better managed.
Present Research. These and other evidence of problems in current healthcare practices in US
have been presented in Chapter 1. This research attempted to analyze the publicly available data
on nationwide hospital wise patient satisfaction survey results, process-of-care quality
performance indicators and outcomes. Chapter-2 of this dissertation gives a brief review of
literature Chapter-3 presented the details on methodology that was adopted for the analysis.
Chapter-4 presented results and a discussion of results was given in Chapter-5. Chapter-6
discussed the results in the light Maharishis Vedic Science principles. A comparison of modern
science with Maharishi Vedic Science was given showing that this approach enhanced the
objective science approach. Application of Maharishi Vedic Science has been done by respected
scientists in areas of modern science and some of them are briey described showing that this
approach will enrich the objective ndings and theories in all levels of life. Accordingly, in this
dissertation, Maharishi Vedic Science principles were applied to enrich the study ndings.
Maharishis Vedic approach to Total Health was also briey presented showing that it could
provide a new paradigm in healthcare, particularly in addressing some of its shortcomings such
as: more emphasis on prevention by lifestyle changes, reduced risk of side effects by using tested
herbal remedies, reduced medical costs because of preventive care, and improved outcomes.
Empirical evidence was also cited showing the efcacy of Maharishis approach to Total Health.
The connections of healthcare to the Unied Field and use of Maharishi Vedic Technologies were
described using a Unied Field chart. A R.icho Ak-kshare chart was then presented followed by a
brief discussion on how the study ndings and the eld of healthcare science are contained in the
structure of one verse from R.k-Veda.
Benets from the study. This Research has brought out the principal components underlying
patient satisfaction and process of care quality and this would be very helpful to hospitals which
are trying to implement performance improvement measures. They need to address and x these
277
factors. The study also analyzed the relationships between patient satisfaction, process of care
quality and outcomes and their association with ownership types. Hospitals run by church groups
come out to be signicantly better. If this is taken together with Charakas assertion that
healthcare staff should have friendship, compassion, joy of serving and equanimity led to the
deduction that hospitals need to seek ways such as training, counseling and other measures to
inculcate these qualities in their staff. Present medical education and training ignores these
qualities. Medical admission tests like MCAT do not use this and medical colleges do not use
such a criterion. In contrast, ancient Ayur-Veda physicians (even the present ones) used these
qualities as selection criteria. Medical college curricula need to be modied to include these.
278
Bibliography
Addinsoft. (2011). XLSTAT Version 2011. 28 West 27th Street, Suite 503 New York, NY 10001
USA.
AHRQ. (2011). Inpatient Quality Indicators (IQIs) @ONLINE. Retrieved October 10, 2011,
from http://www.qualityindicators.ahrq.gov.
Alazri, M. H., & Neal, R. D. (2003, June). The Association between Satisfaction with Services
Provided in Primary Care and Outcomes in Type 2 Diabetes Mellitus. Diabetic Medicine,
20(6), 486490.
Alexander, C. N., Newman, R., Langer, E. J., Chandler, H. M., & Davies, J. L. (1989).
Transcendental Meditation, Mindfulness, and Longevity: An Experimental Study with the
Elderly. Journal of personality and social psychology, 57(6), 950-964.
Alexander, C. N., Rainforth, M. V., & Gelderloos, P. (1991). Transcendental Meditation,
Self-actualization, and Psychological ealth: A Conceptual Overview and Statistical
Meta-analysis. Journal of Social Behavior & Personality, 6(5), 189-248.
Alexander, C. N., Robinson, P., & Rainforth, M. (1994). Treating and Preventing Alcohol,
Nicotine, and Drug Abuse through Transcendental Meditation: A Review and Statistical
Meta-analysis. Alcoholism treatment quarterly, 11(1-2), 13-.
Alexander, V. K. (2005, March). Applications of Maharishi Vedic Science to Developmental
Psychology. Journal of Social Behavior & Personality, 17(1), 920.
Al-Haider, A. S. (1990). Modeling the Determinants of Hospital Mortality. Health Serv Res.
Amza, G., Paris, A., Tarcolea, C., Carp, V., & Nimigean, H. (2009). Risk and Factor Analysis. In
Proceedings of the 3rd wseas int. conference on nite differences - nite elements - nite
volumes - boundary elements. 313, Spl. Independentei. Romania.
279
280
Bruin, J. (2011, February). Stata Annotated Output Robust Regression. Retrieved October 10,
2011, from ucla stat website.
Casey, M. M., & Davidson, G. (2010). Patient Assessments and Quality of Care in Rural
Hospitals. Rural Health Research and Policy Center.
Cavanaugh, K. (1992). Maharishis Vedic Science and TechnologyThe Basis for Economic
Development and World Peace. In Modern science and vedic science (Vol. 5, p. 172-181).
CMS. (2008). National Health Expenditures 2008 Highlights. Retrieved October 10, 2011, from
cms.gov website.
CMS. (2010, March). HCAHPS Quality Assurance Guidelines v 5.0.
CMS. (2011, January 13,2011). Medicare Program: Hospital Inpatient Value-Based Purchasing
Program - Proposed Rule. Federal Register, 76(9), 2454 - 2491.
CNN. (2012, March). Family Healthcare Costs to Exceed $ 20,000 This Year. Retrieved April 28,
2012, from money.cnn.com website. Atlanta, GA.
Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences. (Second ed.). Hillsdale,
New Jersey 07642: Lawrence Erlbaum Associates.
Cohen, J. (1992). A Power Primer. Psychological bulletin, 112(1), 155 - 159.
Cohen, J., Cohen, P., West, S., & Aiken, L. (2003). Applied Multiple Regression/Correlation
Analysis for Behavioral Sciences (D. Riegert, Ed.). Mahwah, NJ 07430: Lawrence Erlbaum
Asociates, Inc.
Cranson, R. W., Orme-Johnson, D. W., Gackenbach, J., Dillbeck, M. C., Jones, C. H., &
Alexander, C. N. (1991). Transcendental Meditation and Improved Performance on
Intelligence-related Measures: A Longitudinal Study. Personality and Individual
Differences, 12(10), 11051116. Retrieved October 10, 2011, from sciencedirect.com.
Daniels, A., England, M. J., Page, A. K., & Corrigan, J. (2005). Crossing the Quality Chasm.
International Journal of Mental Health, 34(1), 5-9.
Davies, E., Shaller, D., Edgman-Levitan, S., Safran, D. G., Oftedahl, G., Sakowski, J., et al.
(2008). Evaluating the Use of a Modied CAHPS Survey to Support Improvements in
281
282
283
284
Hargraves, J. L., Hays, R. D., & Cleary, P. D. (2003). Psychometric Properties of the Consumer
Assessment of Health Plans Study (CAHPS) 2.0 Adult Core Survey. Health Services
Research, 38(6p1), 1509-1527.
Harrington, C., Woolhandler, S., Mullan, J., Carrillo, H., & Himmelstein, D. U. (2001,
September). Does Investor Ownership of Nursing Homes Compromise the Quality of
Care? Am J Public Health, 91(9), 14521455. Retrieved August 19, 2010, from
ajph.aphapublications.org website.
Havranek, E. P., Wolfe, P., Masoudi, F. A., Rathore, S. S., Krumholz, H. M., & Ordin, D. L.
(2004). Provider and Hospital Characteristics Associated With Geographic Variation in the
Evaluation and Management of Elderly Patients With Heart Failure. Archives of Internal
Medicine, 164(11), 1186-1191. Retrieved October 10, 2011, from ama-assn.org website.
Hayward, R. A., & Hofer, T. P. (2001, July). Estimating Hospital Deaths Due to Medical Errors:
Preventability Is in the Eye of the Reviewer. JAMA, 286(4), 415420. Retrieved August 10,
2010, from jama.ama-assn.org website.
Health, D. of, & Services, H. (2011, December). Calculation of 30-day Risk-Standardized
Mortality Rates and Rates of Readmission @ONLINE. Retrieved December 10, 2011, from
hospitalcompare.hhs.gov website.
Heaton, D. P., Schmidt-Wilk, J., & Travis, F. (2004). Constructs, methods, and measures for
researching spirituality in organizations. Journal of Organizational Change Management,
17(1), 62-82.
Hensler III, L. W. (2009, December). A Modest Reading of St. Thomas Aquinas on the
Connection between Natural Law and Human Law. Creighton Law Review, 43(1),
153174.
Herron, R. E. (2008). New Knowledge for New Results. 1100 North 4th street, Faireld, Iowa
52556: 1st world publishing.
Herron, R. E., & Hillis, S. L. (2000). The Impact of the Transcendental Meditation Program on
Government Payments to Physicians in Quebec: an Update. American journal of health
285
286
287
Journal of Medicine, 324(6), 377384. Retrieved August 22, 2010, from doi
(10.1056/NEJM199102073240605.
Lehrman, W. G., Elliott, M. N., Goldstein, E., Beckett, M. K., Klein, D. J., & Giordano, L. A.
(2010). Characteristics of Hospitals Demonstrating Superior Performance in Patient
Experience and Clinical Process Measures of Care. Medical Care Research and Review,
67(1), 38-55.
Lumley, T., Diehr, P., Emerson, S., & Chen, L. (2002). The Importance of the Nnormality
Assumption in Large Public Health Data Sets. Annual review of public health, 23(1),
151-169.
Maharishi Mahesh Yogi. (1986). Life Supported by Natural Law : Discovery of the Unied Field
of all the laws of nature and the Maharishi Technology of the Unied Field Integration of
Modern Science and Ancient Vedic Science to Perpetuate the Rising Age of Enlightenment
and bring Fulllment to every area of society : lectures. Age of Enlightenment Press,
Washington DC.
Maharishi Mahesh Yogi. (1994). Vedic Knowledge for Everyone. Maharishi Vedic University
Press.
Maharishi Mahesh Yogi. (1995a). Maharishi Forum of Natural Law and National Law for
Doctors. Maharishi Vedic university.
Maharishi Mahesh Yogi. (1995b). Maharishis Absolute Theory of Government: Automation in
Administration.
Maharishi Mahesh Yogi. (2001). Science of Being and Art of Living (2001st ed.).
Harmondsworth, Middlesex, England: Penguin Books Ltd.,Middlesex, England.
Marshall, G. N., Elliot, M., Morales, L. S., Spritzer, K., & Hays, R. D. (2001). Conrmatory
Factor Analysis of the Consumer Assessment of Health Plans Study (CAHPS) 1.0 Core
Survey. Psychological Assessment, 13(2), 216.
McClellan, D. O., Mark B. Staiger. (2000). The Changing Hospital Industry: Comparing
For-Prot and Not-for-Prot Institutions [out-of-print volume]. In D. M. Cutler (Ed.),
288
289
2162-2181.
OMalley, A. J., Zaslavsky, A. M., Hays, R. D., Hepner, K. A., Keller, S., & Cleary, P. D. (2005).
Exploratory Factor Analyses of the CAHPS Hospital Pilot Survey Responses across and
within Medical, Surgical, and Obstetric Services. Health Services Research, 40(6P2),
2078-2095.
ONeill, C., Harrington, C., Kitchener, M., & Saliba, D. (2003, December). Quality of Care in
Nursing Homes: An Analysis of Relationships among Prot, Quality, and Ownership.
Medical Care, 41(12), 13181330. Retrieved August 19, 2010, from jstor.org website.
Orme-Johnson, D. (1987, September). Medical Care Utilization and the Transcendental
Meditation Program. Psychosomatic Medicine, 49(5), 493-507.
Orme-Johnson, D. W. (1973). Autonomic Stability and Transcendental Meditation.
Psychosomatic Medicine, 35, 341-349.
Orme-Johnson, D. W. (1988). The Cosmic Psyche as the Unied Source of Creation. In Modern
science and vedic science (Vol. 2, chap. 1). Maharishi University of Management.
Orme-Johnson, D. W. (2000, October). An Overview of Charles Alexanders Contribution to
Psychology: Developing Higher States of Consciousness in the Individual and the Society
(Vol. 7) (No. 4). Springer Science & Business Media B.V.
Orme-Johnson, D. W. (2005, March). Applications of Maharishi Vedic Science to Public Policy.
Journal of Social Behavior & Personality, 17(1), 377382.
Palosky, C., & Singh, R. (2011, September). News Release. on-line.
Park, H. M. (2008). Univariate Analysis and Normality TestUsing SAS, Stata, and SPSS
(Working Paper). The University Information Technology Services (UITS) Center for
Statistical and Mathematical Computing, Indiana University.
PASW. (2007). PASW Statistics 18 Core System Users Guide. 233 South Wacker Drive, 11th
Floor Chicago, IL 60606-6412: SPSS Inc.
Pett, M. A., Lackey, N. R., & Sullivan, J. J. (2003). Making Sense of Factor Analysis (1st ed.).
2455 Teller Road, Thousand Oaks, CA 91320: Sage Publications.
290
Pham, H. H., Coughlan, J., & OMalley, A. S. (2006). The Impact of Quality-Reporting
Programs on Hospital Operations. Health Affairs, 25(5), 1412-1422.
Phillips, R. L., Bartholomew, L. A., Dovey, S. M., Fryer, G. E., Miyoshi, T. J., & Green, L. A.
(2004, April). Learning from Malpractice Claims about Negligent, Adverse Events in
Primary Care in the United States. Quality and Safety in Health Care, 13(2), 121126.
Retrieved August 22, 2010, from bmj.com website.
Pink, G. H. (2003). Hospital Efciency and Patient Satisfaction. Health Services Management
Research, 16(1), 24-38.
Qualitynet. (2011, September). Mortality Measures Overview - Publicly Reporting
Risk-standardized, 30-day mortality measures for ami, hf, and pn.
Reuters, T. (2010, June). 100 Top Hospitals: Health System Benchmark Study. 777 E.
Eisenhower Parkway Ann Arbor, MI 48108.
Romano, P. S., Geppert, J. J., Davies, S., Miller, M. R., Elixhauser, A., & McDonald, K. M.
(2003, March). A National Prole of Patient Safety in U.S. Hospitals. Health Aff , 22(2),
154166. Retrieved August 22, 2010, from healthaffairs.org website.
Roth, P. L. (1994). Missing Data: A Conceptual Review for Applied Psychologists. Personal
Psychology, 47(3), 537-.
Rothman, A. A., Park, H., Hays, R. D., Edwards, C., & Dudley, R. A. (2008). Can Additional
Patient Experience Items Improve the Reliability of and Add New Domains to the CAHPS
Hospital Survey? Health Services Research, 43(6), 2201-2222.
Sanda, M. G., Dunn, R. L., Michalski, J., Sandler, H. M., Northouse, L., Hembroff, L., et al.
(2008, March). Quality of Life and Satisfaction with Outcome among Prostate-Cancer
Survivors. New England Journal of Medicine, 358(12), 12501261. Retrieved August 22,
2010, from doi (10.1056/NEJMoa074311).
Schmidt-Wilk, J. (2005). Applications of Maharishi Vedic Science to Business Management.
Journal of Social Behavior and Personality, 17(1), 229-234.
291
Schneider, R. H., Charles, B. M., Sands, D., Gerace, D. D., Averbach, R. E., & Rothenberg, S.
(1997). The Signicance of the Maharishi Vedic Approach to Health for Modern Health
Care and Medical Education. In Modern science and vedic science. Faireld, Iowa:
Maharishi University of Management. Retreived on October,23 2011 from
mum.edu/pdf msvs website.
Schneider, R. H., & Fields, J. Z. (2006). How to Prevent and Reverse Heart Disease with the
Maharishi Vedic Approach to Health. Basic Health Publications Inc.
Schrdinger, E. (1992). What is Life?: the Physical Aspect of the Living Cell; with, Mind and
Matter; & Autobiographical Sketches. Cambridge University Press.
Schuster, M. A., McGlynn, E. A., & Brook, R. H. (2005). How Good is the Quality of Health
Care in the United States? Milbank Quarterly, 83(4), 843-895.
Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and Quasi-experimental
Designs for Generalized Causal Inference. Boston, MA, US: Houghton, Mifin and
Company.
Shaughnessy, D. F., Peter W. Hittle. (2002, September). Overview of Risk Adjustment and
Outcome Measures for Home Health Agency OBQI Reports:Highlights of Current
Approaches and Outline of Planned Enhancements. UCHSC, Denver, CO.
Shen, Y.-C. (2002). The Effect of Hospital Ownership Choice on Patient Outcomes after
Treatment for Acute Myocardial Infarction. Journal of health economics, 21, 901-922.
Shortell, E. F. X., Stephen M. & Hughes. (1988). The Eeffects of Regulation, Competition, and
Ownership on Mortality Rates among Hospital Inpatients. New, 318 (17), 1100-1107.
Shortell, S. M., Jones, R. H., Rademaker, A. W., Gillies, R. R., Dranove, D. S., Hughes, E. F. X.,
et al. (2000, Feb.). Assessing the Impact of Total Quality Management and Organizational
Culture on Multiple Outcomes of Care for Coronary Artery Bypass Graft Surgery Patients.
Medical Care, 38(2), 207217. Retrieved August 22, 2010, from jstor.org website.
Sloan, F. A., Picone, G. A., Taylor, D. H., & Chou, S.-Y. (2001, January). Hospital Ownership
and Cost and Quality of Care: is there a dimes worth of difference? Journal of Health
292
Economics, 20(1), 121. Retrieved August 19, 2011, from sciencedirect.com website.
Sloan, F. A., Trogdon, J. G., Curtis, L. H., & Schulman, K. A. (2003, October). Does the
Ownership of the Admitting Hospital Make a Difference? Outcomes and Process of Care of
Medicare Beneciaries Admitted with Acute Myocardial Infarction. Medical Care, 41(10),
11931205. Retrieved January 1, 2011, from jstor.org website.
Stevens, J. (1999). Intermediate Statistics: A Modern Approach (2nd ed.). Lawrence Erlbaum
Associates (Hillsdale, N.J.).
Stevens, J. (2002). Applied Multivariate Statistics for the Social Ssciences. Lawrence Erlbaum
Associates.
Teleki, S. S., Kanouse, D. E., Elliott, M. N., Hiatt, L., De Vries, H., & Quigley, D. D. (2007,
Spring2007). Understanding the Reporting Practices of CAHPS Sponsors. Health Care
Financing Review, 28(3), 17-30.
Thomas, E. J., Orav, E., & Brennan, T. A. (2000). Hospital Onwership and Preventable Adverse
Events. Journal of General Intern Medicine, 15(4), 211-219.
Thornlow, D. K., & Stukenborg, G. J. (2006). The Association Between Hospital Characteristics
and Rates of Preventable Complications and Adverse Events. Medical Care, 44(3)(3),
265-269.
Today, U. (2012, February). U.S. Median Household Income up 4% at end of 2011. Delaware,
USA.
Travis, F., & Wallace, R. (1999, September). Autonomic and eeg patterns during eyes-closed rest
and transcendental meditation (tm) practice: The basis for a neural model of tm practice.
Consciousness and Cognition, 8(3), 302318. Retrieved October 19, 2011, from
sciencedirect.com website.
Valiathan, M. S. (2007). The legacy of caraka (1st ed.). Chennai: Orient Longman Private
Limited.
Vincent, C., Neale, G., & Woloshynowych, M. (2001, March). Adverse events in british
hospitals: preliminary retrospective record review. BMJ, 322(7285), 517519. Retrieved
293
294
White, H. (1980, February). Using least squares to approximate unknown regression functions.
International economic review, 21(1), 149 - 170.
White, K., & Ozcan, Y. (1996). Church ownership and hospital efciency. Hosp Health Serv
Adm., 41 (3), 297-310.
Wooldridge. (2006). Introductory econometrics - a modern approach (3rd ed.). Mason, OH
45040: Thomson South-Western.
Woolf, H. S. (2007). Potential health and economic consequences of misplaced priorities. JAMA,
297(5), 523-526.
Yoshikawa, T., Lai Si, T.-A., & McGuire, J. (2007). Corporate governance reform as institutional
innovation: The case of japan. Organization Science, 18(6), 973-988.
Yuan, Z., Cooper, G., Einstadter, D., Cebul, R. D., & Rimm, A. A. (2000). The association
between hospital type and mortality and length of stay. Med, 38(2), 231-245.
Zamarra, J. W., Schneider, R. H., Besseghini, I., Robinson, D. K., & Salerno, J. W. (1996, April).
Usefulness of the Transcendental Meditation Program in the Treatment of Patients with
Coronary Artery Disease. The American Journal of Cardiology, 77(10), 867870.
Retrieved October 10, 2011, from sciencedirect.com website.
295
Appendix A
296
Table 87
h comp 1 u p
h comp 1 a p
h comp 2 sn p
h comp 2 u p
h comp 2 a p
h comp 3 sn p
h comp 3 u p
h comp 3 a p
h comp 4 sn p
h comp 4 u p
h comp 4 a p
h comp 5 sn p
h comp 5 u p
h comp 5 a p
10001
10005
10006
10007
10008
10009
10010
10011
10012
10015
h comp 1 sn p
Provider
7
5
7
7
4
5
1
5
9
5
24
12
16
20
15
16
12
18
19
14
69
83
77
73
81
79
87
77
72
81
5
2
2
6
0
1
1
3
5
3
14
10
11
12
7
10
6
15
12
10
81
88
87
82
93
89
93
82
83
87
21
6
11
14
8
6
9
13
12
9
28
20
26
27
21
19
19
24
27
23
51
74
63
59
71
75
72
63
61
68
10
4
6
11
3
11
1
6
12
4
27
16
20
25
26
21
16
25
18
16
63
80
74
64
71
68
83
69
70
80
28
24
25
24
18
18
16
24
25
11
17
12
20
18
17
19
11
16
17
13
55
64
55
58
65
63
73
60
58
76
297
Table 88
Provider
h clean hsp sn p
h clean hsp u p
h clean hsp a p
h quiet hsp sn p
h quiet hsp u p
h quiet hsp a p
h comp 6 y p
h comp 6 n p
h hsp rating 0 6
h hsp rating 7 8
h hsp rating 9 10
h recmnd dn
h recmnd py
h recmnd dy
10001
10005
10006
10007
10008
10009
10010
10011
10012
10015
18
8
17
12
7
7
6
17
13
13
24
15
21
24
21
13
17
29
21
19
58
77
62
64
72
80
77
54
66
68
9
8
7
8
4
9
3
10
9
8
29
18
22
28
28
20
21
29
28
22
62
74
71
64
68
71
76
61
63
70
76
83
82
69
87
75
86
75
78
82
24
17
18
31
13
25
14
25
22
18
13
6
10
16
7
8
3
8
10
7
25
19
26
26
31
20
21
27
30
30
62
75
64
58
62
72
76
65
60
63
6
4
5
8
6
3
1
4
8
5
25
25
29
34
27
26
18
24
31
23
69
71
66
58
67
71
81
72
61
72
17
298
Appendix B
299
Table 89
Outcomes - Sample data
Providera
10001
10005
10006
10007
10008
10009
10010
10011
10012
10015
a
HAMORT1b
16.10
19.80
15.80
16.20
NA
NA
15.50
15.10
19.30
NA
HFMORT1c
8.80
13.20
9.70
12.50
12.50
11.70
12.30
13.60
11.40
9.40
PNMORT1d
7.90
13.60
10.80
12.50
13.50
11.60
11.10
13.90
11.50
9.60
HAMORT2e
20.20
20.90
16.70
.00
.00
.00
.00
20.10
21.20
.00
HFMORT2f
24.00
24.80
24.70
25.50
23.70
24.00
24.80
22.90
23.40
24.50
PNMORT2g
17.90
18.60
17.70
21.20
18.10
15.70
19.20
19.10
18.90
18.00
NA
300
Appendix C
99
62
96
50
100
100
100
95
100
NA
AMI2
99
100
91
0
0
0
75
88
91
NA
AMI3
99
89
95
100
NA
100
86
98
94
NA
AMI5
100
50
100
NA
NA
100
100
99
100
NA
AMI4
NA
NA
NA
NA
NA
NA
NA
NA
33
NA
AMI7A
76
NA
93
NA
NA
NA
NA
85
67
NA
AMI8A
99
98
96
88
70
98
83
98
100
87
HF2
91
87
79
56
62
100
81
98
89
85
HF3
74
77
79
98
6
97
93
94
94
97
HF1
100
95
100
100
NA
100
100
100
100
88
HF4
92
97
88
94
46
98
88
82
100
96
PN2
89
97
96
94
87
98
96
96
99
96
PN5C
90
97
95
85
86
100
98
95
93
85
PN3B
98
94
99
100
NA
100
95
100
100
100
PN4
AMI2
98
78
96
60
NA
100
86
98
95
NA
10001
10005
10006
10007
10008
10009
10010
10011
10012
10015
Provider
AMI1
Provider
Table 90
301
77
92
90
91
59
100
100
89
87
96
10001
10005
10006
10007
10008
10009
10010
10011
10012
10015
84
81
88
90
47
97
97
85
98
92
PN7
92
96
98
86
100
83
83
95
97
86
SCIPINF1
88
94
95
88
100
83
83
78
93
86
SCIPINF3
98
95
97
89
100
71
71
98
97
93
SCIPINF2
81
77
93
82
100
100
100
82
98
90
SCIPVTE2
85
80
97
82
100
100
100
89
98
90
SCIPVTE1
96
NA
84
NA
NA
NA
NA
60
NA
NA
SCIPINF4
93
100
98
100
100
100
100
100
100
100
SCIPINF6
58
50
95
100
100
100
100
83
100
NA
SCIPCARD2
PN6
Provider
Table 91
302
303
Appendix D
304
Appendix E
.372**
-.759**
.862**
.326**
-.709**
.787**
h comp 3 u p
h comp 3 a p
h comp 4 sn p
h comp 4 u p
h comp 4 a p
h comp 5 sn p
-.565**
.567**
.837**
.508**
-.740**
.854**
.494**
-.684**
h comp 6 y p
h comp 6 n p
h hsp rating 0 6
h hsp rating 7 8
h hsp rating 9 10
h recmnd dn
h recmnd py
h recmnd dy
* Correlation is signicant at
0.05 level
** Correlation is signicant at
0.01 level
Notes:
.090**
-.392**
h quiet hsp sn p
h quiet hsp a p
.595**
h clean hsp a p
h quiet hsp u p
.428**
-.638**
h clean hsp u p
.707**
.883**
h comp 3 sn p
h clean hsp sn p
-.627**
h comp 2 a p
-.032*
.449**
h comp 2 u p
-.653**
.737**
h comp 2 sn p
h comp 5 a p
-.842**
h comp 1 a p
h comp 5 u p
1
.544**
h comp 1 sn p
h comp 1 u p
h comp 1 sn p
-.541**
.471**
.509**
-.666**
.588**
.603**
.355**
-.341**
-.685**
.575**
.617**
-.620**
.650**
.444**
-.713**
.324**
.611**
-.792**
.748**
.489**
-.786**
.768**
.581**
-.772**
.760**
.598**
-.911**
h comp 1 u p
.685**
-.546**
-.747**
.792**
-.628**
-.800**
-.507**
.497**
.634**
-.414**
-.690**
.713**
-.628**
-.633**
.780**
-.193**
-.780**
.858**
-.642**
-.739**
.879**
-.677**
-.808**
.805**
-.710**
-.747**
h comp 1 a p
Table 92
-.571**
.410**
.718**
-.641**
.434**
.732**
.497**
-.497**
-.472**
.244**
.578**
-.522**
.416**
.508**
-.628**
.055**
.669**
-.668**
.405**
.692**
-.669**
.421**
.698**
-.847**
.603**
h comp 2 sn p
-.428**
.375**
.397**
-.527**
.459**
.486**
.254**
-.247**
-.664**
.559**
.596**
-.537**
.573**
.375**
-.610**
.338**
.524**
-.648**
.615**
.398**
-.665**
.661**
.481**
-.935**
h comp 2 u p
.539**
-.432**
-.584**
.637**
-.499**
-.649**
-.390**
.386**
.652**
-.481**
-.654**
.590**
-.567**
-.476**
.686**
-.250**
-.647**
.729**
-.590**
-.573**
.741**
-.628**
-.632**
h comp 2 a p
-.609**
.438**
.762**
-.708**
.496**
.788**
.552**
-.524**
-.453**
.145**
.645**
-.696**
.518**
.717**
-.645**
.005
.772**
-.702**
.348**
.822**
-.870**
.442**
h comp 3 sn p
-.392**
.363**
.326**
-.517**
.495**
.425**
.251**
-.232**
-.666**
.618**
.540**
-.557**
.627**
.356**
-.592**
.339**
.485**
-.655**
.683**
.327**
-.826**
h comp 3 u p
.597**
-.474**
-.657**
.728**
-.583**
-.728**
-.485**
.457**
.650**
-.430**
-.702**
.743**
-.669**
-.645**
.730**
-.189**
-.751**
.800**
-.594**
-.695**
h comp 3 a p
-.642**
.472**
.784**
-.691**
.478**
.776**
.554**
-.529**
-.378**
.103**
.558**
-.572**
.386**
.631**
-.600**
-.010
.727**
-.747**
.269**
h comp 4 sn p
-.401**
.377**
.320**
-.484**
.456**
.405**
.244**
-.204**
-.583**
.569**
.443**
-.414**
.496**
.233**
-.555**
.372**
.411**
-.841**
h comp 4 u p
.637**
-.525**
-.660**
.722**
-.583**
-.716**
-.479**
.438**
.615**
-.451**
-.620**
.607**
-.559**
-.515**
.720**
-.252**
-.692**
h comp 4 a p
305
h comp 1 sn p
.735**
h comp 6 n p
h hsp rating 0 6
* Correlation is signicant at
0.05 level
** Correlation is signicant at
0.01 level
Notes:
-.612**
.553**
h comp 6 y p
h recmnd dy
-.523**
h quiet hsp a p
.472**
-.500**
h quiet hsp u p
.701**
.246**
h quiet hsp sn p
h recmnd dn
.627**
h clean hsp a p
h recmnd py
-.669**
h clean hsp u p
.531**
.535**
h clean hsp sn p
-.701**
.650**
h comp 5 a p
h hsp rating 9 10
-.823**
h comp 5 u p
h hsp rating 7 8
1
.010
h comp 5 sn p
h comp 4 a p
h comp 4 u p
h comp 4 sn p
h comp 3 a p
h comp 3 u p
h comp 3 sn p
h comp 2 a p
h comp 2 u p
h comp 2 sn p
h comp 1 a p
h comp 1 u p
h comp 5 sn p
-.101**
.160**
-.052**
-.114**
.169**
.024
-.007
.036*
-.309**
.397**
.138**
-.115**
.246**
-.046**
-.426**
h comp 5 u p
.570**
-.450**
-.632**
.645**
-.519**
-.643**
-.470**
.496**
.566**
-.403**
-.584**
.616**
-.566**
-.524**
h comp 5 a p
-.519**
.388**
.620**
-.612**
.473**
.632**
.399**
-.382**
-.358**
.106**
.520**
-.881**
.567**
h clean hsp sn p
Table 93
-.417**
.382**
.354**
-.554**
.519**
.467**
.249**
-.239**
-.621**
.491**
.589**
-.889**
h clean hsp u p
.528**
-.435**
-.548**
.658**
-.561**
-.620**
-.365**
.350**
.555**
-.341**
-.627**
h clean hsp a p
-.526**
.447**
.519**
-.645**
.556**
.601**
.375**
-.357**
-.867**
.511**
h quiet hsp sn p
-.284**
.331**
.096**
-.365**
.420**
.220**
.089**
-.074**
-.871**
h quiet hsp u p
.465**
-.447**
-.352**
.580**
-.561**
-.470**
-.265**
.247**
h quiet hsp a p
.532**
-.423**
-.582**
.531**
-.380**
-.583**
-.915**
h comp 6 y p
-.541**
.446**
.562**
-.553**
.404**
.598**
h comp 6 n p
306
h comp 1 sn p
-.878**
.885**
.669**
-.828**
h hsp rating 7 8
h hsp rating 9 10
h recmnd dn
h recmnd py
h recmnd dy
* Correlation is signicant at
0.05 level
** Correlation is signicant at
0.01 level
Notes:
1
.597**
h hsp rating 0 6
h comp 6 n p
h comp 6 y p
h quiet hsp a p
h quiet hsp u p
h quiet hsp sn p
h clean hsp a p
h clean hsp u p
h clean hsp sn p
h comp 5 a p
h comp 5 u p
h comp 5 sn p
h comp 4 a p
h comp 4 u p
h comp 4 sn p
h comp 3 a p
h comp 3 u p
h comp 3 sn p
h comp 2 a p
h comp 2 u p
h comp 2 sn p
h comp 1 a p
h comp 1 u p
h hsp rating 0 6
-.802**
.811**
.525**
-.908**
h hsp rating 7 8
.911**
-.833**
-.776**
h hsp rating 9 10
Table 94
-.776**
.544**
h recmnd dn
-.952**
h recmnd py
h recmnd dy
307
308
Appendix F
309
Appendix G
(a) Component 2
(b) Component 3
(c) Component 4
(d) Component 5
310
Appendix H
311
312
313
314
315
316
317
318
319
320
321
322