106

Journal for Healthcare Quality

Patient Factors Predictive of Hospital Readmissions Within 30 Days Eugene Kroch, Michael Duan, John Martin, Richard A. Bankowitz

Background: Under the Affordable Care Act, the Congress has mandated that the Centers for Medicare and Medicaid Services reduce payments to hospitals subject to their Inpatient Prospective Payment System that exhibits excess readmissions. Using hospital-coded discharge abstracts, we constructed a readmission measure that accounts for cross-hospital variation that enables hospitals to monitor their entire inpatient populations and evaluate their readmission rates relative to national benchmarks. Methods: Multivariate logistic regressions are applied to determine which patient factors increase the odds of a readmission within 30 days and by how much. This study uses deidentified discharge abstract data from a database of approximately 15 million inpatient discharges representing 611 acute care hospitals from Premier healthcare alliance over a 2-year period (2008q4– 2010q3). The hospitals are geographically diverse and represent large urban academic centers and small rural community hospitals. Results: This study demonstrates that meaningful riskadjusted readmission rates can be tracked in a dynamic database. The clinical conditions responsible for the index admission were the strongest predictive factor of readmissions, but factors such as age and accompanying comorbid conditions were also important. Socioeconomic factors, such as race, income, and payer status, also showed strong statistical significance in predicting readmissions. Conclusions: Payment models that are based on stratified comparisons might result in a more equitable payment system while at the same time providing transparency regarding disparities based on these factors. No model, yet available, discriminates potentially modifiable readmissions from those not subject to intervention highlighting the fact that the optimum readmission rate for any given condition is yet to be identified.

Keywords hospital readmissions risk adjustment benchmarking performance measurement Journal for Healthcare Quality Vol. 38, No. 2, pp. 106–115 © 2015 National Association for Healthcare Quality

Introduction and Objectives In June 2009, the Centers for Medicare and Medicaid Services (CMS) began reporting readmission measures on Hospital Compare (www.hospitalcompare.hhs.gov) for acute myocardial infarction (AMI), heart failure (HF), and pneumonia (PN). More recently with the passage of the Affordable Care Act, the Congress has mandated that CMS reduce payments to Inpatient Prospective

Payment System hospitals who exhibit excess readmissions in these same three clinical conditions effective for discharges beginning on October 1, 2012. For discharges occurring in that same time period, CMS has also begun to report a hospital-wide all-cause readmission rate for patients readmitted to the hospital within 30 days for any reason. The policy rationale is for hospital-wide payment adjustment to be based more broadly than for reducing readmissions among a small number of selected diseases. To assess which types of measures serve the policy best, we need to know more about what factors drive readmission rates, especially drivers that include demographic factors, the presence of certain clinical disease states, race, and socioeconomic status. The inclusion of some factors, like race and socioeconomic status, in risk adjustment can raise concerns about “adjusting away” differences that could mask disparities; nonetheless, providers, payers, and policy makers would benefit from a better understanding of the factors that predict readmission especially on a hospitalwide level. Using hospital-coded discharge abstracts, we construct a readmission measure that accounts for cross-hospital variation that enables hospitals to monitor their entire inpatient populations and evaluate their readmission rates relative to national benchmarks. We used multivariate logistic regression to determine which patient factors increase the odds of a readmission within 30 days and by how much.

Methods Definitions of Index Admission and Readmission In a time-varying data environment that supports time series analysis (in contrast to

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

Vol. 38 No. 2 March/April 2016

fixed cross sections of hospitals), the model specification does not assign any particular admission as the “ultimate” index admission. Rather, any hospitalization can be both an index admission and a readmission relative to the previous index discharge. This approach can be implemented on all inpatient data even if the database is continually updated with additional discharges. An index admission was defined as any inpatient hospitalization that was eligible to have subsequent admission to the same hospital (because of data limitations that did not allow us to track patients from one hospital to another). Patients who died, were transferred to other hospitals or left against medical advice were ineligible. A readmission was defined as an inpatient hospitalization, regardless of the cause, within 30 days of the previous discharge from the same hospital. Under this definition, a hospitalization could have been an index admission, readmission, or both, depending on its time interval to the previous hospitalization and the subsequent one. The definitions are illustrated using patient scenarios in Table 1. Patient 1 has two index admissions and zero readmissions because the time interval between the two hospitalizations is more than 30 days. Patient 2 has two index admissions and one readmission because he returned within the 30day window. The third hospitalization does not count as a readmission because it is beyond the 30-day window, and it does not count as an index admission because the

107

patient died, leaving no potential to return to the hospital. Patient 3 has three index admissions, the latter two of which count as readmissions because they were both within the 30-day window. Aggregating across these three patients gives a total of three readmissions (one from patient 2 and two from patient 3) associated with seven index admissions (two from patient 1, two from patient 2, and three from patient 3) for an overall readmission rate of 3/7 = 43%.

Time Frame of Defining Readmissions The time frame of tracking readmission can be set shorter or longer than 30 days. In this study, 30-day cutoff is chosen because it is conventionally referenced in many existing readmission reports. There is no evidence suggesting that 30-day cutoff is superior to 15- or 90-day cutoff from a clinical perspective except to note that attributing the readmission to hospital performance becomes harder as the length of the window increases. A hospital has more influence over events within 30 days of discharge than as far out as 90 days, and readmission rates increase with the length of the window period (Jencks et al., 2009).

Modeling Readmission Risk From Patient Characteristics (Risk Adjustment) An index discharge that is followed by hospitalization within 30 days establishes a readmission, which allows the readmission to

Table 1. Examples of Defining Index Admission and Readmission First Hospitalization, Admit Date and Discharge Date

Second Hospitalization, Admit Date and Discharge Date

Patient 1

1/15 and 1/19

3/20 and 3/28

Patient 2

2/10 and 2/15

3/8 and 3/12

Patient 3

2/25 and 3/1

3/5 and 3/8

Third Hospitalization, Admit Date and Discharge Date

Index Readmission Admission Readmission Rate First and second

None

5/30 and 6/5 (died)

First and second

Second

3/15 and 3/20

First, Second and second, third and third

3/7 = 43%

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

108

Journal for Healthcare Quality

be tracked as a binary outcome (1, 0) of an index admission. The appropriate multivariate model specification, then, is a limited dependent variable logit, which estimates the likelihood that an index admission would have had a readmission based on the patient characteristics of the index admission. The patient risk model is specified as the following:   Pk 5 xik bk   ;   "ik ; LN 1 2 Pk where Pk is the probability of patients with diagnosis k having a readmission, xik is a vector of patient characteristics and socioeconomic factors for patient i, and bk is the diagnosis-specific marginal effect of the independent variables on the readmission log odds. The model was applied to 142 strata based on clinical disease groupings (Kroch et al., 2010 and Pauly et al., 1996). Those strata were mainly determined by principal diagnosis at the three-digit level. Major diseases had their own strata defined by diagnosis codes at the three-digit level, for example, AMI is stratified with International Classification of Disease, Ninth Revision (ICD-9) diagnosis code 410. Similar diagnosis codes were put into the same stratum, for example, ICD-9 diagnosis codes 204–208 are grouped into a Leukemia stratum. Other minor diagnoses were rolled up into broad diagnosis groups, for example, ICD-9 diagnosis codes 680–709 are rolled up into Diseases of Skin and Subcutaneous Tissue. The following patient characteristics and socioeconomic factors comprised the set of nonmodifiable risk factors as regressors. In each disease stratum, the model used stepwise selection to remove variables that did not meet statistical significance level at the 90% (alpha = 0.10) confidence level. That variable selection constraint helps to prevent overfitting and generates reliable parameter estimates suitable for out-of-sample prediction. • Age (quadratic form). • Principal diagnosis (terminal digit International Classification of Disease,



• •





• •



• • •

Ninth Revision, Clinical Modification [ICD-9-CM] code, where statistically significant). Comorbidity score (the severity weighted sum of secondary diagnoses present on admission based on Brailer and colleagues (1996)). Cancer status (benign, malignant, carcinoma in situ, history of cancer, derived from secondary diagnoses). Chronic disease and disease history (terminal digit ICD-9-CM diagnosis codes, such as diabetes, renal failure, hypertension, chronic gastrointestinal disease, chronic cardiopulmonary disease, obesity, and history of substance abuse). Relevant procedure (terminal ICD9-CM procedure codes, where clinically relevant and statistically significant). Admission source (physician referral, clinic referral, health maintenance organization (HMO) referral, transfer from a hospital, skilled nursing facility [SNF] or another healthcare facility, emergency room, court/law enforcement, newborn– normal delivery, premature delivery, sick baby, or extramural birth, unknown/other). Admission type (emergency, urgent, elective, newborn, delivery, unknown/ other). Payer class (Self-pay, Medicaid, Medicare, Blue Cross/Blue Shield, commercial, HMO, Workman’s Compensation, Civilian Health and Medical Program of the Uniformed Services/Federal Employees Health Benefits/Other Federal Government, unknown/other). Discharge disposition (home or self care, SNF, intermediate care facility, other type of institution, home under care of organized home health service, discharged home on intravenous medications, unknown/other). Birth weight (quadratic form for neonatal model only). Sex (female, male, unknown). Race (White, Black, Asian-Pacific Islander, unknown).

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

Vol. 38 No. 2 March/April 2016

• Income (median household income within a zip code reported by U.S. Census Bureau). • Distance traveled (the centroid-tocentroid distance between the zip code of the household and the zip code of the hospital or provider represented as a relative term).

109

centers and small rural community hospitals. Although hospitals’ characteristics are not provided by all hospitals in the data set, we had complete observations for more than 535 of the 611 in our database. Table 2 summarizes how the sample characteristics compare with the rest of the country. Table 3 lists 10 of the most common principal diagnoses that are associated with readmissions in the study sample. They fall within 26 hospital service lines based on major diagnostic categories that span Medicare Severity Diagnosis Related Groups and align with physician specialties, such as cardiology and orthopedics (AHRQ, 2012; Jencks et al., 2009). Patients may return to hospital for many reasons. Some patients come back for scheduled treatment, for example, chemotherapy, but most readmissions are unscheduled. They may or may not be related to the index admission. Table 3 documents how the readmission rates depend on defining readmissions as allcause or more narrowly in a related service line or the same service line. When readmissions are restricted to subsequent hospitalizations that are related to or the same

Note that the comorbidity score excludes chronic disease and disease history, which are introduced separately and are mutually exclusive in the model specification.

Results Sample Data and Readmission Rates This study used deidentified discharge abstract data from a database of approximately 15 million inpatient discharges representing about 611 acute care hospitals from Premier healthcare alliance over a 2-year period from the last quarter of 2008 to the third quarter of 2010 (2008q4– 2010q3). The hospitals are geographically diverse and represent large urban academic

Table 2. Hospitals in the Data Set Hospital Characteristics Bed size

Urban/rural Geographic location

Teaching status

Description

Sample Count

Sample Percent

National Percent*

,100 100–199 200–299 300–399 $400 Rural Urban Midwest Northeast South West Yes No

91 121 108 94 155 131 438 121 77 266 105 86 476

16 22 19 17 28 23 78 22 14 46 19 15 85

38 29 14 8 11 27 73 23 15 43 19 30 70

Teaching status is defined as membership status of Council of Teaching Hospitals and Health Systems. Formal tests yield significant differences for size (p , .01) and teaching status (p , .01). * This information is public data from CMS (Center for Medicare and Medicaid Services) 2013.

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

110

Journal for Healthcare Quality

Table 3. Readmission Rates With Alternative Definitions Readmission Rate Disease Group

All Cause, %

Related Service Line, %

Same Service Line, %

17.0 10.6 11.6 20.2 9.6

12.7 7.0 6.0 16.7 6.9

10.4 2.7 2.9 13.3 2.6

12.1 18.3 17.4 5.2 61.9

9.7 16.4 14.4 2.8 51.8

7.3 14.7 11.1 1.3 49.8

9.8

6.6

5.3

Septicemia AMI Cardiac dysrhythmias HF Occlusion of cerebral arteries (stroke) Pneumonia COPD Renal failure Intervertebral disc disorders Chemotherapy, radiotherapy, and aftercare All inpatients

Service lines are based on the 26 major diagnostic categories from the Medicare Severity Diagnosis Related Groups system. AMI, acute myocardial infarction; COPD, chronic obstructive pulmonary disease; HF, heart failure.

as the index discharge’s service line, readmission rates drop for most diseases. Moreover, a readmission to the related or the same service line does not necessarily indicate the previous hospitalization led to the readmission. The principal diagnosis of that readmission is often assigned for other clinical considerations even when there are clear aspects of treatment that are related to the index diagnosis. Establishing a casual relationship between an index discharge and a readmission would require comprehensive documentation of treatments and precise clinical insights. In claim-based administrative database, such level of documentation is generally absent (Dharmarajan et al., 2013). It also appears problematic to remove certain populations from the readmission measure due to the “planned” nature of readmissions. As shown in Table 3, there is 12% gap between all-cause and same service line readmissions for patients under chemotherapy and radiotherapy. That suggests even those patients may return to hospital for “unplanned” reasons. Hence, a hospital-wide readmission measure

should include all inpatients and planned readmissions can be addressed in proper risk adjustment.

Discriminatory Power of the Risk Model Across 142 disease strata, the model’s ability to discriminate patients with high risk for readmissions from patients with low risk varies widely as evidenced by c-statistics (area under the receiver operating characteristic curve) that range from 0.55 (organ transplant) to 0.85 (obstetrics) with a median value of 0.66 (osteoarthrosis). The higher the c-statistic, the better the discriminatory power, with 1 being perfect discrimination and 0.5 random sorting. As a comparison, most mortality risk models reach a c-statistic in the 0.7 to 0.85 range (Jarman et al., 1999; McNamara et al., 2012). For the three conditions tracked on Hospital Compare (AMI, HF, and PN), our model compares favorably with the CMS model. For AMI, our model had a c-statistic of 0.70, whereas the CMS model had a c-statistic of 0.63. For PN, the results were 0.68 versus 0.63, and for HF,

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

Vol. 38 No. 2 March/April 2016

both methods yielded a very similar result: 0.61 versus 0.60. However, Hospital Compare report is solely based on Medicare patients who are more homogenous than Premier’s all-payer data. That could adversely affect the discriminatory power of a model. Disease stratification, alone, provides a c-statistic of 0.68, which rises to 0.77 for the full model, once the discriminatory power of the model within each disease stratum is taken into account.

Coefficients and Odds Ratios Table 4 is a partial list of risk factors with their coefficients and odds ratio for AMI patients. Each coefficient represents the marginal effect of risk factor on the log

111

odds of readmission. By computing eb, the coefficient is transformed into an odds ratio, which can be interpreted as probability of readmission. For example, the odds ratio of 1.434 (discharge status: SNF vs. home) tells us that holding all other patient characteristics constant, the odds of returning to hospital are 43% higher among patients who were discharged to SNF than those discharged home. The results shown in Table 4 not only support the hypothesis that patient attributes did affect the likelihood of readmission but also indicate the magnitude of such effects. Because the model was stratified by disease, each stratum had its unique set of coefficients and odds ratios. A number of patient risk factors had

Table 4. Partial List of Risk Factors Among AMI Patients Risk Factor Age Sex Race Income Travel distance Cancer status Admission source Admission type Payer type Discharge status Procedure Procedure Procedure Chronic condition Chronic condition Chronic condition Chronic condition Chronic condition

Coefficient

p

Odds Ratio, Estimate (95% Confidence Limits)

Age (squared), in 1,000 Female Black vs. white $ in 1,000 Relative to the mean of all patients Malignant cancer flag Physician referral vs. ER

0.069 0.155 0.092 20.002 20.105

,.0001 ,.0001 ,.0001 .0003 ,.0001

1.072 (1.062–1.082) 1.168 (1.136–1.202) 1.096 (1.050–1.144) 0.998 (0.997–0.999) 0.900 (0.881–0.919)

0.232 20.294

,.0001 ,.0001

1.261 (1.181–1.346) 0.745 (0.713–0.779)

Elective vs. emergency Commercial vs. Medicare SNF vs. home

20.100 20.273 0.360

.0024 ,.0001 ,.0001

0.905 (0.849–0.965) 0.761 (0.713–0.813) 1.434 (1.374–1.496)

CABG Repair of noncoronary vessel Pulsation balloon implant DM, w/cc

0.188 0.235 0.174 0.350

,.0001 .0089 ,.0001 ,.0001

1.207 (1.154–1.262) 1.265 (1.061–1.509) 1.190 (1.112–1.273) 1.419 (1.350–1.492)

Anemia

0.319

.0457

1.375 (1.006–1.880)

Ischemic heart disease, chronic Obstruction, chronic airway NEC Renal failure, chronic

0.174

,.0001

1.190 (1.146–1.235)

0.183

,.0001

1.201 (1.160–1.243)

0.350

,.0001

1.420 (1.299–1.552)

Risk Factor Description

CABG, coronary artery bypass graft; DM, diabetes mellitus; ER, emergency room; NEC, not elsewhere classified; SNF, skilled nursing facility.

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

112

Journal for Healthcare Quality

statistically significant impacts on the likelihood of readmission across disease strata. Age was a prime example, and comorbid conditions increased the probability of readmission as well. A few common chronic conditions, listed in Table 4, are strongly associated with readmissions. A number of others could not be included in the table due to space limitations. Patient demographic and socioeconomic factors are significant in many strata. Relative distance is found to be significant in 121 strata out of 142 making it one of the most frequent risk factors. Admission source, admission type, and discharge status all affected the probability of readmission. Patients who were discharged to other facilities, for example, SNF, are much more likely to be readmitted. Yet, relatively low c-statistics within each regression stratum indicated the absence of the ideal set of risk factors. Our measure of hospital-wide readmission did not exclude patients on the basis of “planned” readmissions. Instead, we created risk-adjustment models to capture these effects accounting separately for those disease strata where

readmissions are commonly planned, such as chemotherapy and radiotherapy. There were also multiple strata, representing lung, liver, and other types of cancer. For patients with cancer, but having other clinical condition as principal diagnosis, we included a cancer status flag in the model. As shown among AMI patients, cancer status did increase the odds of readmission. For example, AMI patients with cancer were 26% more likely to return to hospital.

Hospital Performance The stratified logit model predicts the probability of a readmission based on patient characteristics upon index discharge. The predicted values can be aggregated in flexible ways. By comparing the predicted value to the observed outcome at the hospital level, we were able to measure a hospital’s performance relative to peers. The risk adjustment on patient characteristics eliminated the differences caused by case mix. If a hospital’s observed readmission rate was statistically significantly higher than the predicted rate

Table 5. Range of Risk-Adjusted Readmission Rate and Underperformers

Disease Group Septicemia AMI Cardiac dysrhythmias HF Occlusion of cerebral arteries (stroke) Pneumonia COPD Renal failure Intervertebral disc disorders Chemotherapy, radiotherapy, and aftercare All inpatients

No. of Hospitals

Risk-Adjusted Readmission Rate, % P25 Median P75

Percent of Underperformers

489 369 498 533 430 572 522 424 347 155

14.2 8.9 9.7 17.2 8.0 10.0 15.1 14.7 4.0 57.8

16.4 10.6 11.1 19.6 9.6 11.7 17.5 17.3 5.1 61.8

18.9 12.2 12.8 21.9 11.1 13.4 20.0 19.7 6.2 65.9

12 14 10 15 7 9 12 12 6 21

611

8.4

9.4

10.3

27

Only hospitals with at least 100 cases in a given disease group are included. Underperformance was defined as having had a statistically significant positive deviation between observed and expected rates of readmissions for the respective populations. Risk-adjusted readmissions are derived from the observed-to-expected ratio multiplied by the grand mean of the reference population. AMI, acute myocardial infarction; COPD, chronic obstructive pulmonary disease; HF, heart failure.

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

Vol. 38 No. 2 March/April 2016

(p , .05), it was considered an underperformer. Table 5 shows that the model was able to identify a meaningful number of significant underperformers in key disease grouping. Overall, about 27% of 611 hospitals were identified as underperforming in the study period. At the individual disease stratum level, the sample size (number of cases) is smaller, so the power of the model declines, resulting in a smaller proportion of hospitals determined to be underperformers.

Discussion Payment models that are based on stratified comparisons might result in a more equitable payment system while at the same time providing transparency regarding disparities based on these factors. No model, yet available, discriminates potentially modifiable readmissions from those not subject to intervention, highlighting the fact that the optimum readmission rate for any given condition is yet to be identified. Race, income, distance traveled, and payer status had statistically significant impacts on the likelihood of a patient being readmitted within 30 days. Current CMS payment mechanisms make no adjustment for these factors following the principle that disparities related to socioeconomic factors should not be masked and “adjusted away.” Nonetheless, hospitals operate in an environment that reflects the realities of these factors. They serve vastly different patient populations depending on their location, their payer mix, and the composition of the surrounding communities. Although it would be wrong to assume that hospitals serving a disproportionate share of a vulnerable population are not capable of making improvement in their rates of readmission, it would also be incorrect, in an environment that ties payment to observed rates, to ignore the impact of socioeconomic factors beyond the control of the hospital. In fact, such an approach could have the unintended consequence of diverting resources away from hospitals that provide the indigent care. Although this model and others have used patient factors to accurately detect

113

patients at increased risk for readmission, these models cannot reveal underlying reasons for the return to the hospital (Corrigan and Martin, 1992; Hernandez et al., 2010; Krumholz et al., 2011). The often cited hypothetical case of an HF patient returning to the hospital for an “unrelated” condition because of an automobile accident or other trauma is most likely a rare event, and one that, if truly random, would be accounted for in statistical models. However, patients may return for many possible reasons, as revealed in Table 3, whether planned or not. Treatment failures including complications of care, premature discharges, lack of adequate follow-up, and failures of coordination could be addressed by the healthcare system and therefore can and should be compared (Kansagara et al., 2011). Unfortunately, mixed in with these readmissions are those that return due to lack of adherence based on healthcare literacy or other factors and readmissions that return because there is simply no other access to care in the particular community (Weiss et al., 2010). Not to be discounted are patients who are readmitted for patient- and family-centered reasons rather than imposing repeated trips to the emergency department or observation unit (Joynt et al., 2011). None of the payment models make any effort to separate out readmissions amenable to modification from those that are not. In fact, the “optimal” readmission rate is not yet known (Lindenauer et al., 2011), although the tacit assumption of the current model is that “less is always better.”

Limitations The model was restricted to inpatient data for patients returning to the same hospital, which underpredicted the number of readmissions and the readmission rate as we did not capture those readmissions to other facilities. However, with limitations of the data and patient privacy issues, it was not possible to obtain data that sufficiently captured the inpatient visit in relation to longitudinal care. The model was based on administrative data that suffer from

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

114

Journal for Healthcare Quality

imperfect standards for documentation and coding and introduce artificial variation across hospitals; nevertheless, we have no reason to suppose that these imperfections unduly biased inferences across segments of the inpatient population and the departments that serve them. Finally, although the discriminatory power for our model (as measured by cstatistics) is higher than what is currently available for public reporting, it is still not at the level that would be prudent for payment purposes. Under the method described in this article, the readmission rates of AMI, HF, and PN from the Premier data were different than those reported in the CMS Hospital Compare data (www.hospitalcompare.hhs. gov). The difference was likely attributed to at least three reasons. First, Premier’s data covered all payer types, whereas Hospital Compare was based on Medicare patients only, and Medicare patients are likely to have higher readmission rates due to their age. Second, Hospital Compare was able to track the same Medicare patients across multiple hospitals, whereas Premier’s data could track only readmissions to the same hospital (Nasir et al., 2010). Third, our readmission measure identified more index admissions and readmissions than Hospital Compare, where it is not possible to have multiple index admissions in a 30-day period, which drives the denominator of the ratio lower, causing the readmission rate to be higher.

Conclusions This study demonstrated that meaningful risk-adjusted readmission rates can be tracked in a dynamic database. The clinical conditions responsible for the index admission were the strongest predictive factors for readmissions, but factors such as age and accompanying comorbid conditions were also important. Socioeconomic factors, such as race, income, and payer status, also showed strong statistical significance in predicting readmissions. On the one hand, using them to adjust for patient risk might mask disparities in care, but equally valid (if not more so) is that failing to include them will attribute to hospitals’

influences beyond their control. Moreover, hospitals that serve disproportionate shares of the indigent population will appear to underperform and could suffer under payment rules that are based on such models. Payment models that are based on stratified comparisons might result in a more equitable payment system while at the same time providing transparency regarding disparities based on these factors. No model, yet available, discriminates potentially modifiable readmissions from those not subject to intervention highlighting the fact that the optimum readmission rate for any given condition is yet to be identified.

References AHRQ. Healthcare Cost and Utilization Project, National Estimates of 30-Day Readmission Rates for Specific Procedures and Diagnoses, HCUP Statistical Briefs #153 and #154. 2012. Available at: www.hcup-us.ahrq.gov/reports/statbriefs/ statbriefs.jsp. Accessed September 20, 2013. Brailer, D.J., Kroch, E.A., & Pauly, M.V. Comorbidity-adjusted complication risk: a new outcome quality measure. Med Care 1996;34:490–505. Corrigan, J.M., & Martin, J.B. Identification of factors associated with hospital readmission and development of a predictive model. Health Serv Res 1992;27:81–101. Dharmarajan, K., Hsieh, A.F., & Lin, Z., et al. Diagnoses and timing of 30-day readmissions after hospitalization for heart failure, acute myocardial infarction, or pneumonia. JAMA 2013;309:355–363. Hernandez, A.F., Greiner, M.A., & Fonarow, G.C., et al. Relationship between early physician follow-up and 30-day readmission among Medicare beneficiaries hospitalized for heart failure. JAMA 2010;303:1716–1722. Jarman, B., Gault, S., & Alves, B., et al. Explaining differences in English hospital death rates using routinely collected data. BMJ 1999;318:1515–1520. Jencks, S.F., Williams, M.V., & Coleman, E.A., et al. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med 2009;360:1418–1428. Joynt, K.E., Orav, E.J., & Jha, A.K. Thirty-day readmission rates for Medicare beneficiaries by race and site of care. JAMA 2011;305:675–681. Kansagara, D., Englander, H., & Salanitro, A., et al. Risk prediction models for hospital readmission: a systematic review. JAMA 2011;306:1688–1698. Kroch, E., Johnson, M., Martin, J., & Duan, M. Making hospital mortality measurement more meaningful: incorporating advance

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

Vol. 38 No. 2 March/April 2016

directives and palliative care designations. Am J Med Qual 2010;25:24–33. Krumholz, H.M., Lin, Z., & Drye, E.E., et al. An administrative claims measure suitable for profiling hospital performance based on 30day all-cause readmission rates among patients with acute myocardial infarction. Circ Cardiovasc Qual Outcomes 2011;4:243–252. Lindenauer, P.K., Normand, S.L., & Drye, E.E., et al. Development, validation, and results of a measure of 30-day readmission following hospitalization for pneumonia. J Hosp Med 2011;6:142–150. McNamara, R., et al. Hospital 30-day riskstandardized acute myocardial infarction (AMI) mortality eMeasure. Submitted by Yale New Haven Health Services Corporation/Center for Outcomes Research & Evaluation (YNHHSC/CORE) to CMS under contract # HHSM-500-2008-00025I/ HHSM-500-T0001. 2012. Nasir, K., Lin, Z., & Bueno, H., et al. Is samehospital readmission rate a good surrogate for all-hospital readmission rate? Med Care 2010;48:477–481. Pauly, M.V., Brailer, D.J., & Kroch, E.A. The corporate hospital rating project: measuring hospital outcomes from a buyer’s perspective. Am J Med Qual 1996;11:112–122. Weiss, M., Yakusheva, O., & Bobay, K. Nurse and patient perceptions of discharge readiness in relation to postdischarge utilization. Med Care 2010;48:482–486.

Authors’ Biographies Eugene A. Kroch, Ph.D., is Executive Advisor at Booz Allen Hamilton, before which he was Vice President and Chief Scientist at Premier, Inc. Kroch has lectured and written extensively on economics, public policy, and health economics. He leads efforts to develop measures to track quality, cost, and access, conducting research on large-scale health databases. He is a senior fellow of the Leonard Davis Institute of Health Economics at the University of Pennsylvania and has served as an advisor to the U.S. Department of Health and Human Services in a number of capacities, including serving on expert panels for AHRQ, CMS, and ONC. He holds a bachelor’s degree in economics from the Massachusetts Institute of Technology and A.M. and Ph.D. degrees in economics from Harvard University, where he was a National Science Foundation Fellow. Michael Duan, MS, is Senior Statistician at Premier, Inc., where he leads methodological development of

115

hospital outcome measures and oversees the implementation of those methods in hospital Web applications. With over 10 years of experience in healthcare data, Duan bridges the gap between clinical topics and technical data languages. Duan has supported multiple Premier research initiatives and collaborations with healthcare research organizations. Duan earned a bachelor’s degree in economics from Sichuan University in China and a master’s degree in applied economics from Wright State University. John Martin, MPH, is Executive Director of the Premier Research Institute. With oversight of research activities and strategic product methodology development, Martin works with healthcare organizations, product development teams and field representatives on research and methodology issues. With over 15 years of healthcare research experience, Martin has worked with pharmaceutical, nonprofit, academic and governmental agencies on projects ranging from clinical research to outcomes and health economic research. He also serves on the Improving Health Systems Advisory Panel for PCORI. Martin earned his bachelor’s degree from Sterling College, a master’s degree in public health from the University of Kansas and is a doctoral candidate at Rutgers University, School of Public Health. Richard Bankowitz, MD, MBA, FACP, is EnterpriseWide Chief Medical Officer at Premier, Inc., where he engages physicians, provides thought leadership, and ensures value is delivered to Premier’s clinician constituency. A board-certified internist and a medical informaticist, Dr. Bankowitz has devoted his career to improving healthcare quality at the national level by promoting rigorous, data-driven approaches to quality improvement and by engaging senior clinicians and healthcare leaders. Dr. Bankowitz was named by Modern Healthcare as one of the top 25 clinical informaticists in the US for three consecutive years: 2010–2012, and in 2013 he was named by Modern Healthcare among the top 50 physician executives in the nation. Dr. Bankowitz received his bachelor’s degree from Union College and is a graduate of the University of Chicago Pritzker School of Medicine and the University of Chicago Graduate School of Business. The authors declare no conflict of interest. For more information on this article, contact Eugene Kroch at [email protected].

Copyright 2015 National Association for Healthcare Quality. Unauthorized reproduction of this article is prohibited.

Patient Factors Predictive of Hospital Readmissions Within 30 Days.

Under the Affordable Care Act, the Congress has mandated that the Centers for Medicare and Medicaid Services reduce payments to hospitals subject to t...
129KB Sizes 1 Downloads 11 Views