Risk Factors For 30-Day Hospital Readmission Following
Risk Factors For 30-Day Hospital Readmission Following
Patient readmission within 30 days from discharge has been perceived by the Centers for Medicare and Med-
ical Services as an indicator of poor healthcare quality for specific high-cost medical conditions. Patients who
undergo allogeneic hematopoietic cell transplantation (allo-HCT) are often being readmitted. Our study
identified the risk factors for 30-day readmission among 618 adult recipients of myeloablative allo-HCT
from 1990 to 2009. Two hundred forty-two (39%) of 618 patients (median age 5 42 years [range: 18-66])
were readmitted a median of 10 days (range: 1-30) from their hospital discharge. Median duration of read-
mission was 8 days (range: 0-103). Infections (n 5 68), fever with or without identified source of infection
(n 5 63), gastrointestinal complications (n 5 44), graft-versus-host disease (GVHD) (n 5 38), and other
reasons (n 5 29) accounted for 28%, 26%, 18%, 16%, and 12% of readmissions, respectively. During their
index admission, patients who were subsequently readmitted had more documented infections (P \.001),
higher hematopoietic cell transplantation comorbidity index (HCT-CI) (P \ .01), total body irridiation
(TBI)-based conditioning (P \.001), unrelated donor (P \.001), and peripheral stem cell (P 5 .014) trans-
plantation. In multivariable analysis, HCT-CI (odds ratio [OR] 5 1.78; 95% confidence interval [CI], 1.25-
2.52), TBI-based preparative regimen (OR 5 2.63; 95% CI, 1.67-4.13), and infection during admission for
allo-HSCT (OR 5 2.00; 95% CI, 1.37-2.92) predicted 30-day readmission. Thirty-day readmission itself
was an independent predictor of all-cause mortality (hazard ratio [HR]Adj 5 1.66; 95% CI, 1.36-2.10).
Our data emphasize the importance of a risk-standardized approach to 30-day hospital readmission if it is
used as a quality-of-care metric for bone marrow transplantation.
Biol Blood Marrow Transplant 18: 874-880 (2012) Ó 2012 American Society for Blood and Marrow Transplantation
KEY WORDS: 30-Day readmission, Risk factors, Allogeneic stem cell transplantation, allo-HCT
874
Biol Blood Marrow Transplant 18:874-880, 2012 Risk Factors for 30-Day Hospital Readmission 875
healthcare features (such as having Medicare insurance patient characteristics was collected retrospectively
and being discharged to a skilled nursing facility) [10]. from the Unified Transplant Database (UTD) at the
In a study conducted among adults admitted to a gen- Cleveland Clinic Taussig Cancer Institute.
eral medicine service, Charlson comorbidity index, Information on readmissions including the reasons
SF12 physical component score, insurance status, for readmissions were captured and collected prospec-
social support, previous admissions within a year, tively from patients’ individual transplant coordinators
and length of hospital stay during index admission and/or from on-call physician documentations in
were all found to be independent predictors for early either electronic or paper charts. In addition, further
readmissions [11]. Many readmissions might be pre- detailed information was also collected during daily in-
vented by the use of alternative nonhospital services, as patient rounds by our dedicated UTD data-collecting
most readmissions are because of chronic medical con- personnel. We collected and analyzed the following
ditions [7,12]. Transitional care programs have thus variables: patient age, gender, race, type of hematologic
been proposed to improve hospital discharge planning malignancy (myeloid, lymphoid, or other), pretrans-
for the patients at highest risk for readmissions, plantation disease status (remission or no remission),
including those requiring continuous complex medical number of prior chemotherapy regimens, prior radia-
care [13-15]. On the other hand, some studies have tion treatment, hematopoietic cell transplantation-
demonstrated a correlation between increased use of specific comorbidity index (HCT-CI), type of
hospital resources and lower patient mortality rate preparative regimen, source of hematopoietic cells,
[8,16]. Mortality was found to be lower among donor type, time until neutrophil count recovery, in-
patients with heart failure who were readmitted within fection and acute graft-versus-host disease (aGVHD)
30 days of discharge compared with those who were during hospital stay, and length of hospital stay.
not [5]. Successful inpatient care with low in-hospital Myeloid malignancies were acute myelogenous leuke-
mortality among patients with heart failure may mia, myelodysplastic syndromes, myeloproliferative
lead to a larger living patient population at risk for neoplasms, and chronic myelogenous leukemia. Lym-
readmission [5]. phoid malignancies included acute lymphoblastic
Patients treated with allogeneic hematopoetic cell leukemia, biphenotypic leukemia, lymphomas, multi-
transplantation (allo-HCT) have high rates of hospital ple myeloma, and plasma cell leukemia. Other condi-
readmission [17]. However, neither the factors pre- tions requiring allo-HCT included aplastic anemia
dicting 30-day readmission nor the impact of readmis- and paroxismal nocturnal hemoglobulinemia. Infec-
sion on overall survival (OS) has been examined in the tions were defined as any documented bacterial, viral,
context of allo-HCT. We designed this single-center or fungal infections with isolation of a specific microor-
retrospective study to identify the reasons and risk ganism. The diagnosis of pneumonia was made when
factors for 30-day readmission among the recipients both radiologic and clinical findings of pneumonia
of myeloablative allo-HCT. We also evaluated the were present. This study was approved by the
impact of those factors and of readmission on survival. Cleveland Clinic institutional review board.
Statistical Analysis
MATERIALS AND METHODS Study variables were compared between the pa-
tients who were readmitted and those who were not.
Patients and Setting Chi-square and Wilcoxon rank sum tests were used
A total of 748 adult patients underwent allo-HCT to compare categoric and continuous variables, respec-
between 1990 and 2009 at the Cleveland Clinic tively. Univariate and multivariable risk factors for
Taussig Cancer Institute. Of these, 121 died during readmission were assessed using logistic regression
the transplant admission and 9 died within 30 days of analysis. As multiple preparative regimens were
discharge without a readmission. These 130 patients employed, recursive partitioning analysis was used to
were excluded, and the analysis was based on the categorize these regimens. Recursive partitioning
remaining 618 patients. All patients received their allo- analysis indicated that the most important aspect of
grafts in an inpatient bone marrow transplant unit. The the preparative regimen associated with readmission
patients were kept hospitalized until they became free was whether or not it included total body irridiation
of fever, infections, or other serious complications, (TBI). Multivariable analysis was performed using
and their absolute neutrophil count recovered above a stepwise procedure with a variable entry criterion
500 cells/mL. All patients received discharge teaching of P \ .10 and a variable retention criterion of P \
instructions on common complications after allo- .05. Model-based probabilities of readmission were
HCT, and the appropriate follow-up appointments calculated for all combinations of variables in the final
were arranged by individually assigned transplant co- multivariable logistic model. The number of risk fac-
ordinators upon their hospital discharge. Information tors was calculated for each patient, and the percentage
on hospital readmissions, demographic, and clinical readmitted was compared relative to the number of
876 N. Bejanyan et al. Biol Blood Marrow Transplant 18:874-880, 2012
risk factors using the Cochran-Armitage trend test. Table 1. Patient Characteristics
Survival after the transplantation discharge was esti- Readmission Not Readmitted
mated using the Kaplan-Meier method and compared Variable (n 5 242) (n 5 376) P Value
between patients who were readmitted within 30 days
Age, median 42 (18-66) 40 (18-68) .32
of discharge and those who were not using the log-rank Male 130 (54) 325 (53) .65
test. Univariate and multivariable risk factors for mor- Race
Caucasian 215 (89) 342 (91) .27
tality were assessed using Cox proportional hazards
African American 23 (10) 24 (6)
analysis. Multivariable analysis was done using a step- Other 4 (2) 10 (3)
wise procedure as described earlier. Analyses were Malignancy type
Myeloid 161 (67) 275 (73) .06
done using SASÒ software (SAS Institute, Inc., Cary,
Lymphoid 73 (30) 83 (22)
NC). All statistical tests were 2 sided, and P \ .05 Other* 8 (3) 18 (5)
was used to indicate statistical significance. Disease status
Active disease 141 (58) 206 (55) .40
Remission 101 (42) 170 (45)
Prior chemoRx
RESULTS 0/hydrea§ 23 (9) 50 (13) .06
1-2 159 (66) 257 (69)
Patient Characteristics $3 60 (25) 66 (18)
Prior RT 28 (12) 34 (9) .31
A total of 242 of the 618 allograft patients (39%) HCT-CI‡
were readmitted within 30 days after hospital dis- High/intermediate 150 (64) 177 (49)
Low 84 (36) 187 (51)
charge following their index hospitalization for allo- TBI-containing prep 64 (26) 40 (11) <.001
HCT. Baseline characteristics of both the readmitted Stem cell source
and the not readmitted subsets are summarized in Bone marrow 203 (84) 344 (92) .014
Peripheral stem cell cord 31 (13) 24 (6)
Table 1. Compared with the group that was not read- Cord 8 (3) 8 (2)
mitted, the readmitted group was enriched for patients Donor, unrelated 112 (46) 113 (30) <.001
who had higher HCT-CI (64% versus 49%, P \ .01), Infection† 99 (41) 80 (21) <.001
aGVHD, grade II-IV† 40 (17) 52 (14) .36
TBI-containing preparative regimens (26% versus Length of stay, days† 32 (17-106) 31 (18-91) .96
11%, P \ .001), unrelated donors (46% versus 30%, Days to ANC >500 15 (6-66) 16 (7-75) .10
P \ .001), peripheral blood stem cell source (13% ver- Days to readmission 10 (1-30) —
sus 6%, P 5 .014), and documented infectious compli- ANC indicates absolute neutrophil count.
cations (41% versus 21%, P \ .001) during their index Note: Values in brackets correspond to percentage or range for median
admission for transplantation. Grade 2 or greater values.
*Other: aplastic anemia, peroxismal nocturnal hemoglobulinemia.
aGVHD developed in 15% of patients receiving allo- †During index admission for allo-HCT.
HCT. Median neutrophil count recovery was 16 ‡Twenty subjects had missing HCT-CI data.
days (range: 6-75) and median length of hospital stay §Fifty-three subjects had no prior chemotherapy: chronic myeloid leuke-
32 days (range: 17-106) for all patients. Median days mia (n 5 17), aplastic anemia (n 5 16), myelodysplastic syndromes/
chronic myelomonogotic leukemia (n 5 9), secondary acute myeloge-
from hospital discharge to readmission were 10 (range: nous leukemia (n 5 4), myelofibrosis (n 5 4), myeloproliferative disor-
1-30), and the median duration of the readmission ders (n 5 1), paroxismal nocturnal hemoglobulinemia (n 5 1), and Ph+
hospital stay was 8 days (range: 0-103). acute lymphoblastic leukemia (n 5 1).
Age
Per 10 year increase 1.06 0.92-1.22 .42 — — —
Gender
Male/female 1.08 0.78-1.49 .65 — — —
Race
African American/other 1.54 0.85-2.80 .16 — — —
Malignancy type
Lymphoid/myeloid 1.50 1.04-2.17 .031 — — —
Other/myeloid 0.76 0.32-1.78 .53
Disease status
Active/remission 1.15 0.83-1.60 .40 — — —
Number of prior chemoRx
Per 1 regimen increase 1.20 1.05-1.37 .007 — — —
Prior RT
Yes/No 1.32 0.78-2.23 .31 — — —
HCT-CI
High+Int/Low 1.89 1.35-2.64 <.001 1.78 1.25-2.52 .001
Prep regimen
TBI/No TBI 3.02 1.96-4.66 <.001 2.63 1.67-4.13 <.001
Donor type
Unrelated/related 2.00 1.43-2.80 <.001 — — —
Infection*
Yes/No 2.56 1.79-3.66 <.001 2.00 1.37-2.92 <.001
aGVHD*
Per 1 grade increase 1.13 0.94-1.35 .19 — — —
Days to ANC >500
Per 1 day increase 0.98 0.95-1.01 .16 — — —
Length of stay, days*
Per 1 week increase 1.05 0.96-1.15 .26 — — —
30-day readmission: HCT-CI (OR 1.78; P 5 .001), sion was 10.0 months and 42.8 months, respectively
preparative regimen (OR 2.63; P\.001), and infection (P \ .001).
during index admission for allo-HCT (OR 2.0; P \ Univariate survival analysis identified 30-day read-
.001). mission (hazard ratio [HR] 1.81; 95% CI, 1.47-2.23;
Model-based probabilities for readmission were P \ .001), patient age at transplantation (HR 1.11;
calculated from 3 risk factors identified by multivari- 95% CI, 1.01-1.22; P 5 .035), disease status (HR
able logistic regression analysis (Table 3). The lowest 1.41; 95% CI, 1.15-1.74, P 5 .001), number of prior
risk for readmission (24%) was among the patients chemotherapy regimens (HR 1.14; 95% CI, 1.05-
with low HCT-CI who received non-TBI-based regi- 1.23; P 5 .002), HCT-CI (HR 1.77; 95% CI, 1.42-
mens and had no documented infection during index 2.20; P \ .001), unrelated donor types (HR 1.27;
transplantation admission. In contrast, the greatest 95% CI, 1.02-1.56; P 5 .029), infections (HR 1.32;
risk for readmission (75%) was observed in the pres- 95% CI, 1.06-1.65; P 5 .012), aGVHD (HR 1.20;
ence of all 3 risk factors; those patients with high 95% CI, 1.07-1.33; P 5 .001), and length of hospital
HCT-CI who received a TBI-based regimen and stay (HR 1.18; 95% CI, 1.12-1.24; P\.001) during in-
had infectious complications during their hospital dex admission for transplantation as significant risk
stay. Estimated readmission percentages in the pres- factors for all-cause mortality (Table 4). In the ad-
ence of 0, 1, 2, or all 3 risk factors demonstrated step- justed Cox analysis, however, only 5 of these factors
wise increments with 23.3%, 38.5%, 54.1%, and
76.7% rates of readmission, respectively (P \ .001).
Table 3. Predicted Risk of 30-Day Readmission
DISCUSSION
30-day readmission
Yes/No 1.81 1.47-2.23 <.001 1.66 1.36-2.10 <.001
Age
Per 10 year increase 1.11 1.01-1.22 .035 — — —
Gender
Male/female 1.22 0.99-1.50 .06 — — —
Race
African American/other 1.20 0.83-1.72 .33 — — —
Malignancy type
Lymphoid/myeloid 1.22 0.97-1.52 .09 — — —
Other/myeloid 0.35 0.16-0.78 .011
Disease status
Active/remission 1.41 1.15-1.74 .001 1.42 1.14-1.78 .002
Number of prior chemoRx
Per 1 regimen increase 1.14 1.05-1.23 .002 1.13 1.04-1.22 0.004
Prior RT
Yes/No 1.43 1.04-1.95 .025 — — —
HCT-CI
High + Int/Low 1.77 1.42-2.20 <.001 1.54 1.23-1.93 <.001
Prep regimen
TBI/no TBI 1.08 0.82-1.41 .58 — — —
Donor type
Unrelated/related 1.27 1.02-1.56 .029 — — —
Infection*
Yes/No 1.32 1.06-1.65 .012 — — —
aGVHD*
Per 1 grade increase 1.20 1.07-1.33 .001 — — —
Days to ANC >500
Per 1 day increase 1.01 0.99-1.02 .32 — — —
Length of stay, days*
Per 1 week increase 1.18 1.12-1.24 <.001 1.16 1.10-1.22 <.001
more prevalent among recipients of allo-HCT. We higher rates of 30-day readmissions. This hypothesis,
have found a similar pattern of readmission within however, will require further validation in independent
the 30 days of discharge. Uniformity of readmission dataset(s).
patterns across geographic regions favors the notion As was expected, HCT-CI was strongly associated
that the frequency of readmission is not specific for with both 30-day readmissions and postdischarge mor-
individual transplant centers but is rather related to tality. In this regard, our data corroborates the findings
the allo-HCT patient population. from other studies that consistently report increased
We identified 3 risk factors predictive for 30-day risks of readmissions in patients with underlying
hospital readmissions including HCT-CI, TBI-con- chronic medical conditions [7,10-12]. A reasonable
taining regimens, and infections during index trans- strategy to minimize hospital readmissions because
plant admission. Two of these risk factors (infections of underlying chronic medical conditions would
and use of TBI) are potentially modifiable, but preven- involve scheduling early postdischarge outpatient
tion of infections during index admission may be par- follow-ups with appropriate subspecialty physicians.
ticularly challenging for a number of reasons. First, the Model-based readmission probabilities calculated
increased rate of infectious complications in myeloa- based on the number of risk factors facilitate stratifica-
blative allo-HCT recipients is because of enhanced tion of patients according to their risk of 30-day post-
susceptibility to various pathogens up to a year follow- discharge readmission. According to our model, the
ing allo-HCT until full immune reconstitution is individuals at highest risk are those with high HCT-
achieved [19-21]. Second, discharging allo-HCT CI who received a TBI-based regimen and who devel-
recipients with indwelling central venous catheters re- oped infectious complications during their index
mains a common institutional policy across transplant admission. Identification of these patients before dis-
centers, including ours. Although this practice facili- charge may help improve the quality of their outpa-
tates frequent outpatient blood product transfusions tient care and prevent complications by scheduling
in the early posttransplantation period, it may increase closer follow-ups with their transplant physicians as
the risk for line infections. Third, in order to control well as other specialists.
GVHD, allograft recipients require prolonged immu- The study of heart failure patients across the U.S.
nosuppressive therapy, which itself leads to increased ranked and nonranked hospitals demonstrated that
susceptibility to infectious complications. Further- ranked hospitals had more patient readmissions, but
more, the rate of infections is particularly high in pa- less mortality, compared with nonranked hospitals,
tients who develop GVHD [22]. emphasizing that readmissions and OS possibly
Because the risk of developing life-threatening in- measure distinct quality metrics among heart failure
fections is higher for allo-HCT patients, our institu- patients [11]. In contrast to patients with heart failure,
tional approach has been to require these patients to the correlation between 30-day readmission and
reside within 1 hour of driving distance from the hos- all-cause patient mortality in our study appears to be
pital within first 100 days after allo-HCT. The pa- specific for patients with allo-HCT. Our findings
tients are also expected to have 24-hour caregivers support the notion that quality metrics are highly
and are instructed to notify their transplant nurse or disease-specific and therefore cannot be uniformly
overnight/weekend coverage physician for any fever generalized.
$100.4 F in order to facilitate direct hospital admis- Although the length of index hospital stay for allo-
sion to the bone marrow transplant unit. Such a low HCT was independently associated with overall
threshold for readmission has been used to maximize patient mortality, it was not found to be predictive
rapid detection and management of life-threatening for 30-day readmissions in our study. Prolonged hos-
complications of allo-HCT. Although this strategy pitalization during admission for allo-HCT is gener-
may lead to increased rates of hospital readmissions ally because of serious complications, which could
in patients with allo-HCT, it may also ensure the have possibly resulted in poor OS for the patients in
best quality of care. this study.
Although the type of preparative regimen had no One possible limitation of our study is its retro-
impact on patient survival, the use of TBI-containing spective design with its inherent potential to introduce
regimens was associated with higher rates of readmis- the ascertainment bias. However, our Bone Marrow
sion. A recent large meta-analysis involving over Transplant Database was created independently from
3000 patients from 18 trials compared the outcomes this study and continuous data quality checking has
of TBI/cyclophosphamide and busulfan/cyclophos- been set up to minimize potential bias.
phamide and demonstrated differences in types of In summary, 30-day hospital readmissions follow-
posttransplantation complications [23]. Because TBI- ing myeloablative allo-HCT was associated with poor
containing regimens were associated with 30-day OS. The risk-stratification approach to patient read-
readmission, but not the OS, it is possible that TBI- mission appears to be an essential component of the
induced complications manifested earlier leading to quality-of-care metric among patients with allo-HCT.
880 N. Bejanyan et al. Biol Blood Marrow Transplant 18:874-880, 2012
ACKNOWLEDGMENTS 12. Kirby SE, Dennis SM, Jayasinghe UW, Harris MF. Patient re-
lated factors in frequent readmissions: the influence of condi-
Financial disclosure: The authors have nothing to tion, access to services and patient choice. BMC Health Serv
Res. 2010;10:216.
disclose. 13. Coleman EA, Boult C. Improving the quality of transitional care
for persons with complex care needs. J Am Geriatr Soc. 2003;51:
REFERENCES 556-557.
14. Coleman EA, Berenson RA. Lost in transition: challenges and
1. Jencks SF, Williams MV, Coleman EA. Rehospitalizations opportunities for improving the quality of transitional care.
among patients in the Medicare fee-for-service program. N Ann Intern Med. 2004;141:533-536.
Engl J Med. 2009;360:1418-1428. 15. Coleman EA, Min SJ, Chomiak A, Kramer AM. Posthospital
2. Ross JS, Chen J, Lin Z, et al. Recent national trends in readmis- care transitions: patterns, complications, and risk identification.
sion rates after heart failure hospitalization. Circ Heart Fail. Health Serv Res. 2004;39:1449-1465.
2010;3:97-103. 16. Ong MK, Mangione CM, Romano PS, et al. Looking forward,
3. Hernandez AF, Greiner MA, Fonarow GC, et al. Relationship looking back: assessing variations in hospital resource use and
between early physician follow-up and 30-day readmission outcomes for elderly patients with heart failure. Circ Cardiovasc
among Medicare beneficiaries hospitalized for heart failure. Qual Outcomes. 2009;2:548-557.
JAMA. 2010;303:1716-1722. 17. Grant M, Cooke L, Bhatia S, Forman S. Discharge and
4. Lindenauer PK, Bernheim SM, Grady JN, et al. The perfor- unscheduled readmissions of adult patients undergoing he-
mance of US hospitals as reflected in risk-standardized 30-day matopoietic stem cell transplantation: implications for devel-
mortality and readmission rates for Medicare beneficiaries oping nursing interventions. Oncol Nurs Forum. 2005;32:
with pneumonia. J Hosp Med. 2010;5:E12-E18. E1-E8.
5. Gorodeski EZ, Starling RC, Blackstone EH. Are all readmis- 18. Atkinson K, Farewell V, Storb R, et al. Analysis of late infections
sions bad readmissions? N Engl J Med. 2010;363:297-298. after human bone marrow transplantation: role of genotypic
6. Lindenauer PK, Normand SL, Drye EE, et al. Development, val- nonidentity between marrow donor and recipient and of non-
idation, and results of a measure of 30-day readmission following specific suppressor cells in patients with chronic graft-versus-
hospitalization for pneumonia. J Hosp Med. 2011;6:142-150. host disease. Blood. 1982;60:714-720.
7. Keenan PS, Normand SL, Lin Z, et al. An administrative claims 19. Linch DC, Knott LJ, Thomas RM, et al. T cell regeneration af-
measure suitable for profiling hospital performance on the basis ter allogeneic and autologous bone marrow transplantation. Br J
of 30-day all-cause readmission rates among patients with heart Haematol. 1983;53:451-458.
failure. Circ Cardiovasc Qual Outcomes. 2008;1:29-37. 20. Atkinson K. Reconstruction of the haemopoietic and immune
8. Mulvey GK, Wang Y, Lin Z, et al. Mortality and readmission systems after marrow transplantation. Bone Marrow Transplant.
for patients with heart failure among U.S. News & World 1990;5:209-226.
Report’s top heart hospitals. Circ Cardiovasc Qual Outcomes. 21. Storek J, Dawson MA, Storer B, et al. Immune reconstitution af-
2009;2:558-565. ter allogeneic marrow transplantation compared with blood
9. Krumholz HM, Merrill AR, Schone EM, et al. Patterns of hos- stem cell transplantation. Blood. 2001;97:3380-3389.
pital performance in acute myocardial infarction and heart fail- 22. Ochs L, Shu XO, Miller J, et al. Late infections after allogeneic
ure 30-day mortality and readmission. Circ Cardiovasc Qual bone marrow transplantations: comparison of incidence in re-
Outcomes. 2009;2:407-413. lated and unrelated donor transplant recipients. Blood. 1995;86:
10. Silverstein MD, Qin H, Mercer SQ, Fong J, Haydar Z. Risk fac- 3979-3986.
tors for 30-day hospital readmission in patients ./565 years of 23. Shi-Xia X, Xian-Hua T, Hai-Qin X, Bo F, Xiang-Feng T. Total
age. Proc (Bayl Univ Med Cent). 2008;21:363-372. body irradiation plus cyclophosphamide versus busulphan with
11. Hasan O, Meltzer DO, Shaykevich SA, et al. Hospital readmis- cyclophosphamide as conditioning regimen for patients with
sion in general medicine patients: a prediction model. J Gen leukemia undergoing allogeneic stem cell transplantation:
Intern Med. 2010;25:211-219. a meta-analysis. Leuk Lymphoma. 2010;51:50-60.