Professional Documents
Culture Documents
Gut 2011 Morris 806 13
Gut 2011 Morris 806 13
com
Colon
ABSTRACT
Objectives To assess the variation in risk-adjusted 30day postoperative mortality for patients with colorectal
cancer between hospital trusts within the English NHS.
Design Retrospective cross-sectional population-based
study of data extracted from the National Cancer Data
Repository.
Setting All providers of major colorectal cancer surgery
within the English NHS.
Participants All 160 920 individuals who underwent
major resection for colorectal cancer diagnosed between
1998 and 2006 in the English NHS.
Main outcome measures National patterns of 30-day
postoperative mortality were examined and logistic
binary regression was used to study factors associated
with death within 30 days of surgery. Funnel plots were
used to show variation between trusts in risk-adjusted
mortality.
Results Overall 30-day mortality was 6.7% but
decreased over time from 6.8% in 1998 to 5.8% in 2006.
The largest reduction in mortality was seen in 2005 and
2006. Postoperative mortality increased with age (15.0%
(95% CI 14.1% to 15.9%) for those aged >80 years),
comorbidity (24.2% (95% CI 22.0% to 26.5%) for those
with a Charlson comorbidity score $3), stage of disease
(9.9% (95% CI 9.3% to 10.6%) for patients with Dukes D
disease), socioeconomic deprivation (7.8% (95% CI 7.2%
to 8.4%) for residents of the most deprived quintile) and
operative urgency (14.9% (95% CI 14.2% to 15.7%) for
patients undergoing emergency resection). Risk-adjusted
control charts showed that one trust had consistently
significantly better outcomes and three had significantly
worse outcomes than the population mean.
Conclusions Significant variation in 30-day
postoperative mortality following major colorectal cancer
surgery existed between NHS hospitals in England
throughout the period 1998e2006. Understanding the
underlying causes of this variation between surgical
providers will make it possible to identify and spread
best practice, improve outcomes and, ultimately, reduce
30-day postoperative mortality following colorectal
cancer surgery.
INTRODUCTION
Colorectal cancer is the third most common cancer
in the UK and, with more than 35 000 new cases
diagnosed annually,1 improving outcome is important. International comparisons show that survival
from colorectal cancer in the UK is relatively
poor.2e4
Surgery is the mainstay of colorectal cancer
treatment and is generally undertaken within
6 months of diagnosis. International variation in
806
Colon
vary between individuals. A young patient with an early-stage
tumour and no comorbid disease will bear a very different risk
from that of an elderly frail patient with advanced disease.
Robust comparison of postoperative mortality between
providers requires analyses to be risk-adjusted to ensure that
the impact of relevant differences between populations (such as
patient age, comorbidity and stage of disease) is taken into
account. Second, the annual number of patients operated upon
for colorectal cancer varies between institutions. Greater variability in postoperative mortality will arise by chance in institutions with smaller annual case loads compared with units
managing larger numbers. Appropriate adjustment for differences in hospital case load is also vital if valid institutional
comparisons are to be made.
National risk-adjusted outcome comparisons require national
data but, until now, such data have not been available.
Numerous routine data sources exist that contain information
about different aspects of colorectal cancer care, but none
contain all the data required to enable risk-adjusted comparisons
of postoperative mortality. For example, cancer registry data10
contain detailed tumour incidence and outcome information but
little data on treatment. In contrast, Hospital Episode Statistics
(HES)11 contain detailed treatment information but little
information on the characteristics of the tumours. The National
Cancer Intelligence Network (NCIN) has linked these data
sources to create the National Cancer Data Repository
(NCDR).12 This resource allows the main processes and
outcomes of care to be tracked for every NHS cancer patient in
England.
The NCDR contains case-mix information of reasonable
quality and good data on important prognostic factors such as
stage, age and comorbidity. Not all the relevant data items are
complete for each patient, however, and such missing information can restrict the interpretability of institutional comparisons. Techniques such as multiple imputation have the potential
to overcome some of the problems that arise from missing
data.13 14
This study seeks to make use of the available data within the
NCDR to scrutinise risk-adjusted surgical outcomes for patients
with colorectal cancer at a population level. It seeks to monitor
national patterns and trends of 30-day postoperative mortality
following major resection of colorectal cancer and, using
multiple imputation13 14 and funnel plots,8 9 to produce robust
comparisons of the performance of all NHS hospital trusts in
England.
METHODS
The NCDR consists of pooled data from the eight populationbased cancer registries that cover England, linked (using all or
combinations of the identiers of NHS number, date of birth,
postcode at diagnosis and sex) to an extract of HES including
episodes of inpatient care for individuals who presented in any
NHS hospital with a diagnostic code for cancer between April
1997 and June 2007.
Information was extracted from this resource on all individuals who underwent a major resection for a primary colorectal
cancer diagnosed between 1 January 1998 and 31 December
2006. Information on age, sex, Dukes stage, NHS number,
postcode at diagnosis, dates of diagnosis and, where relevant,
date of death, was extracted from the registry dataset for all
colorectal cancers (ICD1015 C18eC20) while information about
patient management was derived from HES. For each colorectal
cancer patient in the registry extract who could be linked to the
Gut 2011;60:806e813. doi:10.1136/gut.2010.232181
Colon
Table 1
Total
Age
#50
9552
51e60
22 436
61e70
43 695
71e80
57 373
>80
27 864
Sex
Male
88 789
Female
72 131
Year of diagnosis
1998
18 018
1999
18 076
2000
18 075
2001
17 296
2002
17 336
2003
17 498
2004
17 869
2005
18 421
2006
18 331
Cancer site
Colon
104 023
Rectosigmoid
13 555
Rectum
43 342
Charlson comorbidity score
0
137 924
1
13 618
2
6551
$3
2827
Dukes stage
A
17 151
B
53 711
C
51 390
D
14 234
Unknown
24 434
IMD income category
Most affluent
31 538
2
35 139
3
34 409
4
31 889
Most deprived
27 541
Unknown
404
Operation type
Elective
145 480
Emergency
15 440
Stage at diagnosis
Multilevel
%
(95% CI)
112
438
1669
4344
4141
1.2
2.0
3.8
7.6
15.0
1.2
1.9
3.9
7.6
15.0
(1.0 to 1.4)
(1.7 to 2.2)
(3.5 to 4.2)
(7.1 to 8.0)
(14.1 to 15.9)
6037
4667
6.8
6.5
6.8
6.5
(6.4 to 7.3)
(6.1 to 6.9)
1231
1276
1249
1195
1213
1155
1230
1086
1069
6.8
7.1
6.9
6.9
7.0
6.6
6.9
5.9
5.8
6.9
7.1
6.9
6.9
7.0
6.6
6.9
6.0
5.9
(6.3
(6.5
(6.4
(6.4
(6.5
(6.0
(6.4
(5.5
(5.4
7933
748
2023
7.6
5.5
4.7
7.7
5.6
4.6
(7.3 to 8.2)
(5.1 to 6.2)
(4.3 to 5.0)
7333
1754
946
671
5.3
13.0
14.0
24.0
5.4
13.1
14.7
24.2
(5.0 to 5.7)
(12.2 to 14.1)
(13.2 to 16.3)
(22.0 to 26.5)
606
3122
3247
1287
2442
3.5
5.8
6.3
9.0
10.0
4.2
6.2
7.1
9.9
e
(3.7
(5.8
(6.7
(9.3
e
to
to
to
to
to
to
to
to
to
to
to
to
to
to
to
to
to
to
7.5)
7.7)
7.5)
7.5)
7.5)
7.1)
7.4)
6.4)
6.4)
4.7)
6.6)
7.6)
10.6)
1790
2113
2320
2358
2122
1
5.7
6.0
6.7
7.4
7.7
0.2
5.7
6.0
6.8
7.3
7.8
e
(5.3
(5.5
(6.4
(6.9
(7.2
e
6.1)
6.5)
7.2)
7.8)
8.4)
8401
2303
5.8
15.0
5.8
14.9
(5.4 to 6.2)
(14.2 to 15.7)
Characteristic
Age at diagnosis
#50
51e60
61e70
71e80
$81
Sex
Male
Female
Operation
Elective
Emergency
IMD income category
Most affluent
2
3
4
Most deprived
Unknown*
Cancer site
Colon
Rectosigmoid
Rectum
Year of diagnosis
1998
1999
2000
2001
2002
2003
2004
2005
2006
Charlson comorbidity score
0
1
2
3
Dukes stage distribution
Before imputation (all)
Before imputation
(staged cases only)
After imputation
Across imputed cases
Unknown
9.4
11.0
11.4
11.0
9.0
25.7
28.3
31.2
35.3
39.7
33.8
33.8
32.6
31.1
30.5
12.8
11.2
9.4
8.2
6.1
18.2
15.7
15.4
14.6
14.7
11.1
10.1
32.9
34.0
31.7
32.2
8.8
8.9
15.5
14.9
11.6
1.8
33.5
31.9
31.4
37.3
8.3
14.2
15.2
14.8
11.5
11.2
10.4
10.2
9.8
11.4
32.3
33.3
34.2
33.5
33.5
34.7
32.0
32.1
31.8
32.0
31.7
36.6
8.7
8.6
8.7
8.8
9.7
11.6
15.5
14.8
14.9
15.6
15.4
5.7
7.6
11.8
17.6
36.7
30.7
26.3
32.1
33.3
31.0
9.9
10.0
6.0
13.6
14.3
19.2
8.5
10.1
10.4
11.3
10.9
11.2
10.4
11.5
11.8
31.0
30.9
33.1
34.7
34.5
33.9
31.8
35.1
35.5
28.2
30.3
31.8
32.8
34.2
32.6
30.4
34.5
32.6
8.6
8.9
8.8
9.9
9.2
8.7
8.4
8.7
8.4
23.7
19.9
15.9
11.4
11.2
13.6
19.0
10.2
11.7
10.9
8.2
10.6
9.2
33.1
34.7
35.5
37.4
31.8
33.7
31.9
31.9
9.0
8.7
7.6
6.1
15.3
14.7
14.5
15.5
10.7
12.6
33.4
39.4
31.9
37.7
8.9
10.4
15.2
e
12.7
13.5
39.5
40.0
37.4
36.3
10.4
10.1
e
e
*Index of Multiple Deprivation (IMD) income category information was missing in only
a small proportion of cases (n404, 0.25%).
Colon
Table 3
Characteristic
OR
(95% CI)
p Value
OR
(95% CI)
p Value
1.08
0.99
(1.07 to 1.08)
(0.98 to 1.00)
<0.001
<0.001
1.08
0.97
(1.08 to 1.08)
(0.97 to 0.98)
<0.001
<0.001
1.00
0.81
e
(0.77 to 0.85)
<0.001
1.00
0.83
e
(0.79 to 0.86)
1.00
2.61
e
(2.46 to 2.77)
<0.001
1.00
2.67
e
(2.53 to 2.82)
1.00
1.28
1.53
2.63
e
(1.17 to 1.4)
(1.39 to 1.68)
(2.37 to 2.93)
<0.001
1.00
1.23
1.54
2.50
e
(1.12 to 1.35)
(1.40 to 1.69)
(2.24 to 2.78)
1.00
1.04
1.13
1.24
1.37
e
(0.96
(1.05
(1.15
(1.26
<0.001
1.00
1.03
1.11
1.22
1.32
e
(0.96
(1.04
(1.13
(1.23
1.00
0.83
0.92
e
(0.76 to 0.91)
(0.86 to 0.98)
<0.001
1.00
0.88
0.94
e
(0.82 to 0.96)
(0.89 to 0.99)
1.00
2.12
2.46
4.51
e
(1.99 to 2.26)
(2.26 to 2.68)
(4.06 to 5.01)
<0.001
1.00
2.05
2.43
4.38
e
(1.94 to 2.18)
(2.25 to 2.62)
(3.98 to 4.82)
to
to
to
to
1.12)
1.22)
1.34)
1.49)
<0.001
<0.001
<0.001
<0.001
to
to
to
to
1.10)
1.19)
1.30)
1.42)
0.0021
<0.001
*Analyses based on only those individuals for whom all case-mix variables were available (n136 105).
yAnalyses based on all individuals with missing case-mix data being imputed (n160 290).
IMD, Index of Multiple Deprivation.
RESULTS
One hundred and sixty thousand, nine hundred and twenty
individuals were identied with a diagnosis of colorectal cancer
between 1998 and 2006 and who subsequently underwent
a major resection for their disease. They were treated by 150
different hospital teams within 28 cancer networks. Of these,
10 704 (6.7%) died within 30 days of the resection. The characteristics of the study population are presented in table 1. The
distribution of stage before and after imputation and among the
imputed cases was very similar (table 2).
Table 1 shows the characteristics of the population and their
relationship to 30-day postoperative mortality. Owing to the
high numbers of individuals included in the study, most of the
differences across groups are statistically signicant. Analyses
undertaken on the imputed dataset and accounting for the
clustering of patients with trusts showed that the 30-day
postoperative mortality declined slightly from 6.9% (95% CI
Gut 2011;60:806e813. doi:10.1136/gut.2010.232181
Colon
p<0.001). The odds of death were lower in women than men
(OR 0.83, 95% CI 0.79 to 0.86) and lower for patients with rectal
tumours than for those with colonic tumours (OR 0.94, 95%
0.89 to 0.99).
The odds of death within 30 days of surgery for the various
case-mix factors are shown in table 4. A strong deprivation effect
was apparent even after adjustment for the case-mix factors
thought to differ between socioeconomic groups such as stage,
comorbidity and emergency presentation.
For patients diagnosed during 1998e2002, unadjusted 30-day
postoperative mortality was above the 99.8% control limit for
eight trusts (gure 1A), indicating that their surgical mortality
was signicantly higher than expected. A further 20 trusts were
above the 95% control limit. After inclusion of all risk factors in
the model (gure 1B), eight trusts remained above the upper
99.8% limit while 15 trusts were above the 95% control limit. In
the unadjusted model six trusts had signicantly lower 30-day
postoperative mortality than expected (ie, they were below the
lowest 99.8% control limit) and 19 more trusts were below the
95% control limit. In the risk-adjusted model, ve and 17 trusts
respectively remained below the lower 99.8% and 95% control
limits.
Similar results were observed for patients diagnosed during
2003e2006 (gure 2). In the risk-adjusted model, postoperative
mortality in ve trusts was above the upper 99.8% control limit
while a further 11 were above the 95% limit. Three trusts had
signicantly better outcomes than expected, below the 99.8%
limit, and nine trusts were below the 95% limit.
Three trusts appeared above (and one trust below) the 99.8%
control limits in both time periods indicating consistently
outlying 30-day postoperative mortality.
DISCUSSION
This retrospective population-based study is the rst to provide
a comprehensive national perspective on the 30-day postoperative mortality associated with colorectal cancer surgery
across England. Overall, 6.7% of the study population died
within 30 days of surgery amounting to 10 704 deaths. There
was signicant variation across the population with postoperative mortality greater in the elderly, men, the socioeconomically deprived, those with advanced stage disease at
diagnosis or with additional comorbidities and among those
operated upon as an emergency. Signicant variation, independent of case-mix, was also observed between hospital trusts.
One trust had postoperative mortality signicantly lower and
three signicantly higher than could be explained by the casemix information available in both time periods examined. These
hospitals were all district general hospitals and two of those
with signicantly worse outcomes than expected had Foundation status.
The postoperative mortality of 6.7% seen in this study is
notably higher than that previously reported for the UK. Data
submitted to the most recent National Bowel Cancer Audit
Programme (NBOCAP) report recorded postoperative mortality
of 4.7% for all surgical cases.5 Submission to this audit is,
however, voluntary, resulting in incomplete case ascertainment.
In addition, it is not possible to calculate postoperative
mortality across all surgically resected cases submitted due to
incomplete or inaccurate reporting of dates of surgery and (prior
to 2009) death. In consequence, it is likely that, due to underreporting, the results from the NBOCAP audits are biased.
The postoperative mortality of English patients with colorectal cancer determined in this study is also signicantly higher
Table 4 Results of the additive logistic regression models (based on the imputed dataset) investigating the odds of death within 30 days of surgery
Characteristic
OR
(95% CI)
OR
(95% CI)
OR
(95% CI)
OR
(95% CI)
1.08
1.08 to 1.08
1.08
1.07 to 1.08
1.08
1.08 to 1.08
1.08
1.08 to 1.08
1.00
0.78
0.98
e
0.75 to 0.81
0.97 to 0.98
1.00
0.82
0.97
e
0.79 to 0.86
0.96 to 0.98
1.00
0.82
0.97
e
0.79 to 0.86
0.96 to 0.98
1.00
0.83
0.97
e
0.79 to 0.86
0.97 to 0.98
1.00
0.77
0.70
e
0.71 to 0.84
0.67 to 0.74
1.00
0.82
0.77
e
0.76 to 0.89
0.73 to 0.81
1.00
0.83
0.82
e
0.76 to 0.90
0.78 to 0.86
1.00
0.88
0.94
e
0.82 to 0.96
0.89 to 0.99
1.00
1.03
1.14
1.26
1.42
e
0.97
1.07
1.18
1.32
1.00
1.03
1.12
1.22
1.35
e
0.96
1.05
1.14
1.26
1.10
1.20
1.31
1.45
1.00
1.03
1.12
1.22
1.35
e
0.96
1.05
1.14
1.25
1.10
1.20
1.31
1.45
1.00
1.03
1.11
1.22
1.32
e
0.96
1.04
1.13
1.23
1.00
2.04
2.34
4.13
e
1.93 to 2.16
2.17 to 2.53
3.76 to 4.54
1.00
2.02
2.38
4.23
e
1.91 to 2.14
2.20 to 2.56
3.85 to 4.65
1.00
2.05
2.43
4.38
e
1.94 to 2.18
2.25 to 2.62
3.98 to 4.82
1.00
1.31
1.70
2.86
e
1.20 to 1.44
1.55 to 1.87
2.57 to 3.19
1.00
1.23
1.54
2.50
e
1.12 to 1.35
1.40 to 1.69
2.24 to 2.78
1.00
2.67
e
2.53 to 2.82
to
to
to
to
1.11
1.22
1.35
1.52
to
to
to
to
to
to
to
to
to
to
to
to
1.10
1.19
1.30
1.42
810
Colon
than that reported from other countries. Thirty-day postoperative mortality from population-based studies in Scandinavia, Canada and the USA ranged from 2.7% (for rectal cancers
alone) to 5.7%.20e25 While there are undoubtedly big differences
between the populations in these international studies that
make comparison with the UK difcult, the postoperative
mortality from these reports is consistently below the 6.7%
found in this study. This suggests that either the NHS may have
fundamentally worse postoperative outcomes than some other
comparable health services or the operative risk of patients
differs between countries. Understanding and minimising these
differences could signicantly reduce the number of premature
deaths caused by this disease across the country.
A strong relationship between socioeconomic deprivation and
postoperative mortality was observed, with those residing in
more deprived areas having a signicantly greater risk of death
within 30 days of surgery than those residing in more afuent
areas. This effect remained despite adjustment for stage of
disease, comorbidity and urgency of surgical resection. This
nding mirrors other studies that have shown socioeconomic
gradients in both the long-term and short-term outcomes of
colorectal cancer.22 26 27 In contrast, there is evidence to suggest
that this gradient disappears in a randomised trial setting where
patients are given equal treatment,28 although it is possible that
811
Colon
factor is confounding our results. These analyses do, however,
include adjustment for many of the most important factors
known to inuence outcome such as age, comorbidity, stage of
disease and socioeconomic deprivation and, as such, the results
should not be dismissed. Previous studies have shown that
routine data can be used to identify divergent practice,31 32 and
the linked data upon which this study is based are much more
comprehensive than any previously available. Furthermore, it is
hoped that the NBOCAP data will soon be incorporated into the
NCDR. These data contain information such as anaesthetic risk
scores that are not currently available in the NCDR but that
could signicantly inuence postoperative outcomes. The
availability of such data could help to rene the models further
in the future.
Currently, the NCDR is limited by the timeliness of the routine
data available. Efforts are being made across the NHS to increase
the timeliness of data it collects and it is a priority for the NCIN
to improve the temporality of the NCDR. In the future it is
therefore hoped that more timely reporting can be achieved.
Many factors may inuence 30-day postoperative mortality.
These may relate to the patient (eg, stage of disease or level of
comorbidity) or the institution offering care (such as the
specialisation of the operating team, the quality of postoperative
care or the availability of beds in high dependency and intensive
care units). Examining how these factors vary in relation to 30day postoperative mortality rates may provide evidence to help
explain the variability seen across English NHS trusts, among
socioeconomic groups and between countries. While this study
has identied providers with outlying 30-day postoperative
mortality, it is not possible to determine from the data available
what aspects of caredor, indeed, if the quality of care within
these unitsdis decient. The outlying status could be explained
by problems in data quality, chance or, as discussed previously,
case-mix factors not quantied in this study. Institutions with
outlying status should not, however, be ignored, but efforts
should be made to determine why they appear to have signicantly better or worse postoperative mortality than other units.
With this information it should then be possible to learn from
those achieving good outcomes by seeking the underlying causes,
adding to and spreading the adoption of best practice guidelines,33e36 improving poor outcomes and, ultimately, reducing
postoperative mortality following colorectal cancer surgery.
Cardiothoracic surgeons in the UK have openly reported
their surgical outcomes since 1998, and the publication of these
results has demonstrably improved outcomes for cardiothoracic
surgical mortality across the country.37 38 It is intended that
the development of the NCDR will enable national 30-day
postoperative mortality to be reported annually at both a trust
level anddas the NCDR also contains information about the
consultant overseeing each surgical eventdpotentially the
surgeon level (although it should be emphasised that postoperative mortality should be treated as a colorectal team or
trust event and the operating surgeon should not be vilied).
The NCIN plans to work collaboratively with the Association
of Coloproctology of Great Britain and Ireland to disseminate
these ndings to hospital trusts and cancer networks and use
them to inform care. The reduction in 30-day postoperative
mortality over the study period is welcomed, but our ndings
show that there is wide variation across the NHS and considerable scope for improvement. It is now time for colorectal
cancerdand subsequently other cancer teamsdto follow the
example of cardiothoracic surgeons in order to improve
outcomes. The NCDR provides the means by which this
process can start.
812
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
Colon
25.
26.
27.
28.
29.
30.
31.
Dimick JB, Cowan JA, Upchurch GR, et al. Hospital volume and surgical outcomes
for elderly patients with colorectal cancer in the United States. J Surg Res
2003;114:50e6.
Coleman MP, Babb P, Quinn MJ, et al. Socio-economic inequalities in cancer
survival in England and Wales. Cancer 2001;91:208e16.
Coleman MP, Rachet B, Woods LM, et al. Trends and socio-economic inequalities in
cancer survival in England and Wales up to 2001. Br J Cancer
2004;90:1367e73.
Nur U, Rachet B, Parmar MR, et al. No socioeconomic inequalities in colorectal
cancer survival within a randomised clinical trial. Br J Cancer 2008;99:1923e8.
Audit Commission. PbR Data Assurance Framework 2007/8. London: Audit
Commission, 2008.
Morris EJ, Jordan C, Thomas JD, et al. Comparison of treatment and outcome
information between a clinical trial and the National Cancer Data Repository. Br J
Surg 2011;98:299e307.
Aylin P, Alves B, Best N, et al. Comparison of UK paediatric cardiac surgical
performance by analysis of routinely collected data 1984e96: was Bristol an outlier?
Lancet 2001;358:181e7.
32.
33.
34.
35.
36.
37.
38.
813
doi: 10.1136/gut.2010.232181
Updated information and services can be found at:
http://gut.bmj.com/content/60/6/806
These include:
References
Email alerting
service
Topic
Collections
This article cites 26 articles, 8 of which you can access for free at:
http://gut.bmj.com/content/60/6/806#BIBL
Receive free email alerts when new articles cite this article. Sign up in the
box at the top right corner of the online article.
Notes