Hospital Readmission Rates in Medicare Advantage Plans
Published Online: February 27, 2012
Jeff Lemieux, MA; Cary Sennett, MD; Ray Wang, MS; Teresa Mulligan, MHSA; and Jon Bumbaugh, MA
In April 2009, Dr Stephen Jencks and his colleagues published a detailed study of hospital readmissions within Medicare’s traditional fee-for-service (FFS) program.1 The Jencks et al study was notable both for the high rates of readmissions found—a 19.6% 30-day readmission rate in 2004—and for the fi nding that in approximately half of those readmissions, there was no physician visit in the interim. The Jencks et al study also implied that there has been no substantial improvement in FFS readmission rates in the 23 years since the 1984 benchmark readmissions study by Anderson and Steinberg, who studied FFS data from 1981.2 The lack of physician visits following hospital discharge suggests opportunities to improve coordination of follow-up care. Jencks et al estimated that the cost of unplanned rehospitalizations exceeded $17 billion in FFS Medicare in 2004. (We use the terms “readmissions” and “rehospitalizations” synonymously throughout this report; likewise the terms “hospital discharges” and “admissions” may be used interchangeably where, for counting purposes, both terms refer to the same number of hospitalizations).
In general, Medicare’s FFS program pays hospitals a fi xed amount for each admission based on the diagnosis-related group (DRG) code and does not routinely pay for transitional care programs, provider-to-provider communications and care coordination, 24-hour nurse help lines, pharmacy reconciliation efforts, and other initiatives that can help prevent readmissions.3,4 (For a survey of the recent literature on methods of reducing readmissions, see the study by Boutwell and Hwu.3 The health insurance industry also has published a set of examples used by Medicare Advantage [MA] plans to reduce readmissions.4) The Patient Protection and Affordable Care Act of 2010 requires the Centers for Medicare & Medicaid Services (CMS) to establish a hospital readmissions reduction program in Medicare FFS beginning in 2013. When implemented, the program would reduce payments to hospitals based on their readmission rates. In April 2011, CMS announced a new “partnership for patients,” with the goals of reducing FFS readmission rates by 20% and hospitalacquired conditions or adverse events among FFS patients by 40%.5
Alternatively, capitated reimbursement (a fixed, risk-adjusted amount per enrollee) gives MA plans strong financial incentives to attempt to reduce avoidable hospitalizations and readmissions via case management or network contracting arrangements. Several industry studies using hospital discharge data have been published in recent months that estimate readmission rates for both FFS and MA in particular states. (An example is the 2010 study by America’s Health Insurance Plans Center for Policy and Research. 6) The discharge data are collected by states directly or in cooperation with the Agency for Healthcare Research and Quality (the agency is the source of most of the data used in the industry readmission studies7). However, these studies are currently limited to a handful of states where the discharge data contain reliable (de-identifi ed) patient indicators that can be used to track readmissions. (For a discussion of analytic and data issues surrounding the use of hospital discharge data sets for readmissions analysis, see the working paper by America’s Health Insurance Plans Center for Policy and Research.8) Moreover, state discharge data sets do not capture readmissions that occur out of state, and the distinctions between MA and FFS as the patients’ source of coverage are imprecise in some states.
The primary aim of this report is to create a national benchmark for readmission rates among MA patients using the same methods described by Jencks et al.1 Establishing a benchmark measurement will make it possible to track progress in lowering MA readmission rates over time. A second goal is to develop direct comparisons of MA and FFS readmission rates and discuss some analytic issues surrounding comparisons. We analyzed the possible impact of differences in geographic location, time frame, age, entitlement (disability) status, and the diagnosis and complications present at admission. The comparisons between the FFS and MA systems can be instructive for discerning the impact of the different incentive systems.
DATA AND METHODS
The MA data for this study are from the commercially available MedAssurant Medical Outcomes Research for Effectiveness and Economics Registry (MORE² Registry), which is a clinically enriched super set of de-identifi ed, longitudinal, patient-level administrative claims data. The MA data included approximately 5.6 million observations (enrollee-years) and 2.4 million individuals in the 3-year (2006-2008) period from 11 MA plans (de-identifi ed and aggregated by MedAssurant). There were 2.4 million individuals in the MA sample in 2008, approximately one-fourth of the MA population of 9.4 million in that year. We estimated that 45% of the patients in the MA sample were enrolled in for-profi t plans; 55% were in nonprofi t plans. The FFS data in this study were from the Medicare 5% sample claims and administrative files.
We used the same calculation methods described in the Jencks et al study1 for the MA data in the MedAssurant data set, based on the “all cause” concept and following the Jencks method of excluding readmissions identifi ed with the DRG for rehabilitation. This method tracks “initial” or “index” admissions from the fourth quarter of the prior year. It excludes patients who died or switched coverage types during the study period. Computationally, we believe our results for 2006-2008 using this method are directly comparable to the Jencks resultsfor 2004.
The MedAssurant MA sample is demographically and geographically diverse. (MedAssurant has entered into an agreement with the National Committee for Quality Assurance to provide data and measures of preventable hospital admissions and readmissions for the Medicare population, under contract with CMS.9) We compared the age and sex distribution of the 2006-2008 MedAssurant data with national MA demographic data from Medicare’s 5% sample administrative files for 2006-2008. The percentages of enrollees by age and sex in the MedAssurant MA sample were similar to those in the national data in most respects. The MedAssurant MA sample included proportionately more enrollees 80 years and older but slightly fewer in the 55 to 69 year range. The net impact on benchmark readmission rates of these differences in age distribution is likely to be small, in part because the differences were small and in part because patients 80 years and older and patients under age 65 years both tend to have higher-than-average readmission rates. eAppendix A (available at www.ajmc.com) shows the MedAssurant and national MA enrollment data by age cohort and sex.
While the MedAssurant data set has geographic representation in all 50 states, the distribution by state of MA enrollees is not identical to the national distribution of MA enrollees. In general, the MedAssurant MA sample had a somewhat higher share of enrollment in the South and a lower share of enrollees in the West. This could be an important difference, since readmission rates tend to be higher in the South than in the West.
To test for this possibility, we computed readmission rates by state from the MedAssurant MA sample and “weighted” them according to the statewide MA enrollment data. The resulting nationally weighted readmission rate was only slightly lower—about 0.1 percentage point—than the rate we computed from the MedAssurant data. Although we have not performed a detailed analysis of substate distributions, our preliminary conclusion is that any differences in the geographic dispersion of the MedAssurant and overall MA enrollment are not large enough to substantially change the estimated benchmark MA readmission rate.
We also tested whether geographic differences in samples would have an impact on the MA versus FFS comparisons. Our preliminary fi nding was that no adjustments were needed, at least based on comparisons of the samples on a state-by-state basis. We also performed some preliminary investigations of differences in enrollment by county within states. The distribution of MA enrollment by county appeared to be relatively similar to that of FFS in several states examined. If anything, MA enrollees may be clustered more in urban areas, which tend to have higher-than-average readmission rates within the states analyzed. Thus, any such within-state geographic adjustment would likely favor MA, not FFS. eAppendix B (available at www.ajmc.com) explains the calculations that led to these conclusions.
Risk Adjustment in MA Versus FFS Comparisons
There is no perfect way to measure how differences in risk of readmission could affect the comparisons between MA and FFS patients. We developed a measure of risk of readmission using expected readmission rates for each type of hospital discharge (DRGs) and measured differences in MA and FFS discharge distributions against this expectation. However, it is possible that additional unmeasured or unmeasurable factors could have affected the results.
PDF is available on the last page.