Hospitals By Christopher Ody, Lucy Msall, Leemore S. Dafny, David C. Grabowski, and David M. Cutler doi: 10.1377/hlthaff.2018.05178 HEALTH AFFAIRS 38, NO. 1 (2019): 36–43 ©2019 Project HOPE— Decreases In Readmissions The People-to-People Health Foundation, Inc. Credited To Medicare’s Program To Reduce Hospital Readmissions Have Been Overstated Christopher Ody (c-ody @kellogg.northwestern.edu) is ABSTRACT Medicare’s Hospital Readmissions Reduction Program (HRRP) a research assistant professor has been credited with lowering risk-adjusted readmission rates for in the Kellogg School of Management, Northwestern targeted conditions at general acute care hospitals. However, these University, in Evanston, Illinois. reductions appear to be illusory or overstated. This is because a concurrent change in electronic transaction standards allowed hospitals Lucy Msall is a PhD candidate in the Booth School of to document a larger number of diagnoses per claim, which had the Business, University of effect of reducing risk-adjusted patient readmission rates. Prior studies of Chicago, in Illinois. the HRRP relied upon control groups’ having lower baseline readmission Leemore S. Dafny is the MBA rates, which could falsely create the appearance that readmission rates Class of 1960 Professor of Business Administration at are changing more in the treatment than in the control group. Harvard Business School, in Accounting for the revised standards reduced the decline in risk-adjusted Boston, Massachusetts. readmission rates for targeted conditions by 48 percent. After further David C. Grabowski is a adjusting for differences in pre-HRRP readmission rates across samples, professor in the Department of Health Care Policy, Harvard we found that declines for targeted conditions at general acute care Medical School, in Boston. hospitals were statistically indistinguishable from declines in two control David M. Cutler is the Otto samples. Either the HRRP had no effect on readmissions, or it led to a Eckstein Professor of Applied systemwide reduction in readmissions that was roughly half as large as Economics in the Department of Economics at Harvard prior estimates have suggested. University and a research associate at the National Bureau of Economic Research, both in Cambridge, Massachusetts. n March 2010 the Affordable Care Act clined after the HRRP was established. Declines (ACA) established the Hospital Read- were larger for targeted conditions than for non- missions Reduction Program (HRRP) to targeted conditions, for Medicare patients than incentivize hospitals to reduce readmis- for other patients, and for hospitals that were sions among Medicare beneficiaries. subject to the HRRP than for hospitals that were IThe program penalized general acute care hos- not.1–4 Readmission rates also declined for non- pitals having higher-than-anticipated thirty-day targeted conditions, which could have been a risk-adjusted readmission rates for targeted con- result of spillover effects of the program or of ditions. In October 2012 the program began pe- unrelated changes.5,6 Most of the declines oc- nalizing hospitals for three targeted conditions: curred during the period between the enactment acute myocardial infarction, heart failure, and of the ACA (March 2010) and the month when pneumonia. The program targeted additional hospitals first faced penalties (October 2012). conditions in more recent years, and the penalty This study presents new evidence on why risk- increased from a maximum of 1 percent of Medi- adjusted readmission rates have decreased since care reimbursements in 2012 to a maximum of the HRRP was established and why reductions 3 percent starting in 2015. were larger for patients with targeted conditions A number of studies have documented that treated at general acute care hospitals than for thirty-day risk-adjusted readmission rates de- other patients. Andrew Ibrahim and coauthors 36 Health Affairs January 2019 38:1 Downloaded from HealthAffairs.org on August 13, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. have identified one concern with evidence for treated at targeted hospitals had higher readmis- behavioral changes by hospitals: The majority sion rates than those who had nontargeted con- of the decrease was generated by increased pa- ditions or were treated at nontargeted hospitals. tient risk scores, rather than by actual lower re- Many factors (such as coding) that might affect admission rates.4 However, the study did not readmission rates will have a more pronounced determine why patient risk scores increased, effect for conditions with higher initial readmis- and the authors noted that the changes could sion rates.When we accounted for differences in have resulted from either increased patient risk baseline readmissions across these samples, we or increased coding of diagnoses. The HRRP no longer found evidence to suggest that de- bases patient risk scores on age, sex, and co- clines have been larger for patients with targeted morbidities calculated using patient diagnoses conditions and those admitted to targeted hos- from inpatient and outpatient claims for the pitals than for patients with nontargeted condi- twelve months before hospitalization for the tar- tions or at nontargeted hospitals. These findings geted condition. Crucially, the HRRP risk scores call into question the relatively broad-based exclude many diagnoses coded only during the consensus that the HRRP has meaningfully de- targeted admission. As a result, it is unclear how creased risk-adjusted readmission rates. much hospitals could have manipulated patient risk scores. To “game” the program’s risk adjust- ment, hospitals would need to code patient diag- Study Data And Methods noses more aggressively for care received before Data Sources And Study Variables We con- the program’s targeted admission. structed the sample and calculated risk-adjusted We argue that the increased coding of patient readmission rates to match those used in prior risk scores has a more mundane explanation: HRRP studies.1,3,4 Specifically, we began with Between the March 2010 establishment of the data from Medicare’s 100 percent Research Iden- HRRP and the October 2012 introduction of pen- tifiable Files, and we defined index admissions as alties, the Centers for Medicare and Medicaid hospitalizations that occurred in the period Jan- Services (CMS) changed the electronic transac- uary 2007–November 2014 among beneficiaries tion standards that hospitals use to submit Medi- who were enrolled in fee-for-service Medicare for care claims, allowing for an increased number of at least twelve months before the index admis- diagnosis codes. This change coincided with the sion, who were age sixty-five or older, and who time window in which risk-adjusted readmission had at least thirty days of Medicare coverage rates declined the fastest. In particular, before following their discharge. 2011 providers submitted claims using version We constructed three samples of index admis- 4010A of the electronic transaction standards. sions, one of which was targeted by the HRRP This version allowed a maximum of nine or and two of which were not. Past studies com- ten diagnosis codes (the tenth code was reserved pared changes in risk-adjusted readmission rates for coding an external cause of injury and was for the targeted sample against changes for these usually, but not always, blank). Starting in Jan- two ostensible control groups.1,4 Specifically, the uary 2011 CMS encouraged providers to submit first sample consisted of index admissions to claims using version 5010, and most hospitals targeted general acute care hospitals for the immediately complied. That version allowed a three conditions that were always targeted (acute maximum of twenty-five diagnosis codes.7–9 myocardial infarction, heart failure, and pneu- Some providers submitted bills using the new monia). The targeted conditions and targeted system in 2010, while others waited for the man- hospitals sample included 7,049,806 index ad- datory transition in January 2012. missions to 3,350 hospitals. The second sample, We document that around January 2011 the nontargeted conditions, was composed of ad- share of inpatient claims with nine or ten diag- missions to targeted general acute care hospitals noses plummeted and the share with eleven or for conditions that were never included in the more rose sharply. Accounting for this change HRRP but were included in CMS’s thirty-day all- reduces the decline in risk-adjusted readmission cause hospital readmission measure.10 This sam- rates for patients in HRRP-targeted conditions at ple included 40,148,231 index admissions to targeted hospitals by 48 percent. 3,467 hospitals. The third sample, nontargeted We then reexamined a second piece of the os- hospitals, was composed of admissions for the tensible evidence that the HRRP reduced risk- three always-targeted conditions to critical ac- adjusted readmission rates: Decreases have been cess hospitals, a group of hospitals that was larger for patients with targeted conditions not subject to the HRRP. That sample included treated at hospitals targeted by the program than 429,072 index admissions to 1,115 hospitals. for other patients. Before the implementation of We constructed risk-adjusted readmission the program, patients with targeted conditions rates for each of the three samples using a com- January 2019 38:1 Health Affairs 37 Downloaded