Total hip arthroplasty (THA) and total knee arthroplasty (TKA) are among the most common and costly elective surgical procedures1,2. Research studies have demonstrated variation in the quality of these procedures3-5. In response, public and private payers have focused on improving the quality and value of THAs and TKAs through initiatives such as public reporting, centers of excellence, reference pricing, and pay for performance.
Readmission rates are a key quality and efficiency measure used in such initiatives. Blue Cross and Blue Shield plans and other payers use readmission rates as part of their criteria for selecting centers of excellence6,7. Readmission rates following THA and TKA also have been added to the formula used by the Centers for Medicare & Medicaid Services (CMS) Hospital Readmissions Reduction Program (HRRP)8. As part of the HRRP, CMS reduces payments to hospitals that are identified as having excess readmissions. Furthermore, CMS compares each hospital’s readmission rate with the U.S. national rate on its Hospital Compare website9.
The specific methodology used to risk-adjust readmission rates affects the comparison of hospitals for public reporting and the determination of financial penalties in these payer-driven initiatives. A number of studies have documented disparities in readmission rates following THAs and TKAs based on patient race/ethnicity and socioeconomic status10-13. Currently, neither CMS nor any known private payer incorporates race/ethnicity or socioeconomic status into their risk-adjustment algorithms. Exclusion of these patient factors means that hospitals caring for a high proportion of patients who are minority or poor, or both, could be inaccurately designated as low-performing hospitals14. The American Academy of Orthopaedic Surgeons has voiced support for studying risk-adjustment methods that include race/ethnicity and socioeconomic status15. The National Quality Forum recently convened an expert panel that recommended incorporating socioeconomic status into risk-adjustment algorithms, and the U.S. Congress passed a law in 2014 calling for CMS to examine the effect of socioeconomic status on Medicare quality measures16,17.
Despite these recommendations, little is known about the extent to which race/ethnicity and socioeconomic status affect hospitals’ readmission rates and relative performance. In this study, we added these 2 patient factors to the CMS risk-adjustment algorithm to examine their impact on readmission rates following THAs and TKAs.
Materials and Methods
Data from the Agency for Healthcare Research and Quality (AHRQ) Healthcare Cost and Utilization Project (HCUP) 2011 all-payer (including the uninsured) State Inpatient Databases (SID) were used. We focused on 16 HCUP partner states that have reliable synthetic patient linkage numbers that can be used to track a person across hospitals within a state and that report the patient’s race/ethnicity18. These 16 states are geographically dispersed and account for 44% of the total U.S. resident population and 43% of all U.S. hospitalizations. We obtained information on hospital characteristics from the American Hospital Association (AHA) Annual Survey of Hospitals database.
The study population included adults, ≥18 years of age, who had had a non-revision (primary) THA or TKA at a community hospital that was not a rehabilitation hospital, long-term acute-care hospital, or hospital with limited reporting of patient race (for example, one in which a large percentage of discharge records were missing data on race or a large hospital at which patients were reported only as white non-Hispanic). We selected cases (the index admissions) for the period January to November 2011 to allow for 30-day follow-up. We excluded 6.5% of cases from the analysis because the procedure was a partial arthroplasty (0.001%), a second THA or TKA was performed within 30 days after discharge from an index admission (0.2%), the arthroplasty was nonelective (2.3%), bilateral surgery was performed during the index admission (2.8%), or race was missing from the record (1.2%). The inclusion and exclusion criteria described here are consistent with CMS methodology (see Appendix)19. These criteria yielded a relatively homogeneous clinical population that is likely at lower risk for readmission. For comparison, we also separately examined a subset of the study population: Medicare patients 65 years of age and older.
Risk-Adjusted Readmission Rates
The key measure in our analysis was a risk-adjusted readmission rate calculated for THA and TKA combined, which is consistent with the CMS approach. We defined readmission as a subsequent “unplanned” hospitalization (as defined by CMS specifications) occurring within 30 days after discharge from the index admission regardless of whether the patient was readmitted to the same or a different hospital. Planned hospitalization, according to CMS specifications, is a hospitalization for a procedure or principal diagnosis that is always planned (for example, an organ transplant procedure or maintenance chemotherapy) or potentially planned without an acute diagnosis19. Patients with ≥1 hospitalizations within the first 30 days were counted as having only 1 readmission.
We calculated 2 separate sets of risk-adjusted readmission rates for THA and TKA combined. Our approach to these calculations is consistent with the methodology used by CMS as part of Hospital Compare and the HRRP. Additional information regarding the CMS methodology for calculating risk-standardized readmission rates and its application to the HCUP data are presented in the Appendix.
The standard CMS method for calculation of a risk-adjusted rate involves running a hierarchical logistic regression model and regressing the presence of a readmission on patient age, sex, presence of 29 specific comorbidities, and random effects for each hospital19. CMS defines comorbidities on the basis of the prior year of inpatient and outpatient data; however, the HCUP SID include only current-year inpatient data, so we modified the approach by using comorbidities at the time of the inpatient index surgery only.
We calculated the second set of risk-adjusted rates by repeating the same risk-adjustment approach but adding race/ethnicity and socioeconomic status into the hierarchical logistic regressions. We measured race/ethnicity using 3 patient-level binary variables—Hispanic, black non-Hispanic, and other non-Hispanic (for example, Asian, Pacific Islander, or Native American)—and excluding white non-Hispanic as the reference. We used the median household income for the patient’s residential ZIP code to measure socioeconomic status.
Hospital Impact Metrics
We used 3 approaches to determine the impact of adding race/ethnicity and socioeconomic status to the CMS risk-adjusted readmission-rate algorithm. First, we examined the absolute readmission rate for each hospital. Second, we used the CMS approach of categorizing hospitals by the statistical difference (better, worse, or no different) between their readmission rate and the population mean. To do this, we created confidence intervals using bootstrapping techniques consistent with the approach used by the CMS19. Hospitals were considered no different from the population mean if the 95% confidence interval for their readmission rate contained the overall observed population mean rate. Hospitals were considered better or worse than the population mean if their 95% confidence interval did not contain the observed population mean rate.
The third approach was to use the method of calculating a hospital’s excess readmission ratio that CMS employs to determine penalties in the HRRP. The excess readmission ratio is calculated as the ratio of the risk-adjusted predicted number of readmissions to the expected number of readmissions from the logistic regression models. From those models, a hospital’s predicted number of readmissions is based on its own readmissions performance after controlling for case mix; the hospital’s expected number of readmissions is based on the performance of all hospitals in the data set that have a similar mix of patients treated with THA and TKA. A ratio of <1 indicates better performance and a ratio of >1 indicates poorer performance. Hospitals with an excess readmission ratio of >1 are at risk of receiving a financial penalty from CMS8. The size of the penalty is determined by the aggregate payments for the excess readmissions across a number of conditions including THA and TKA.
First, we used scatterplots to compare hospitals’ readmission rates for the combined THA and TKA measure calculated using the standard CMS risk-adjustment methodology versus the method that included race/ethnicity and socioeconomic status.
Second, we statistically compared the 95% confidence intervals for the 2 different risk-adjusted hospital readmission rates (with and without adjustment for race/ethnicity and socioeconomic status) to determine whether they were better than, worse than, or no different from the population mean. We calculated the proportions of hospitals that experienced no change in designation, that had a favorable change (for example, from “no different” to “better than” the population mean), and that had an unfavorable change (for example, from “no different” to “worse than” the population mean).
Third, we compared hospitals’ excess readmission ratios with and without race/ethnicity and socioeconomic status included in the model. We calculated the proportions of hospitals that had a favorable change (that is, from an excess readmission ratio of >1 to a ratio of ≤1) and that had an unfavorable change (from an excess readmission ratio of ≤1 to a ratio of >1). The remaining hospitals were categorized as experiencing no change.
We calculated estimates for each of the 3 hospital impact measures separately for all payers and for Medicare only. We included all hospitals at which at least 1 THA or TKA had been performed, whereas CMS limits these calculations to hospitals with a minimum of 25 THAs or TKAs. We performed a sensitivity analysis by limiting our sample to the hospitals with at least 25 THAs or TKAs.
Of 1,194 hospitals that performed at least 1 THA or TKA across all payers, 24% were teaching hospitals, 5% were critical access hospitals, 24% were safety-net hospitals, and 17% had performed <25 THAs and TKAs combined (Table I). The largest proportions of hospitals were private/not-for-profit (61%), were located in the South (42%) and in a large metropolitan area (48%), and had a large number of beds (48%). The hospitals were reasonably well distributed in terms of the proportion of patients receiving THA or TKA who were minority and had a low income. The hospitals performed an average combined total of 289 THA and TKA procedures in 2011, with an overall risk-adjusted readmission rate (without race/ethnicity and socioeconomic status included in the model) of 4.12% for both surgical procedures. When the analysis was limited to the Medicare population, it showed that 1,163 hospitals had performed an average of 151 THA and TKA procedures in 2011, with a mean risk-adjusted readmission rate of 4.85%.
Race/ethnicity and socioeconomic status were significant predictors of readmission rates when added to the CMS risk-adjustment model (not shown). The readmission rate was positively associated with black race (all payers, p < 0.0001; Medicare, p = 0.0003) and negatively associated with socioeconomic status in the analysis of all payers (p = 0.013).
When race/ethnicity and socioeconomic status were added to the risk-adjustment model, there were small changes in the hospitals’ risk-adjusted readmission rates, as shown in Figure 1. For all payers, the risk-adjusted hospital THA and TKA readmission rates without race/ethnicity and socioeconomic status included in the model ranged from 2.3 to 7.7 per 100 arthroplasties. When race/ethnicity and socioeconomic status were added to the algorithm, the adjusted readmission rates ranged from 2.4 to 7.4 (representing a rate change of −0.39 to +0.28), with an average percentage rate change of 0.8%. For Medicare patients, the risk-adjusted hospital THA and TKA readmission rates without race/ethnicity and socioeconomic status included in the model ranged from 3.0 to 8.9 per 100 arthroplasties. When race/ethnicity and socioeconomic status were added to the algorithm, the adjusted readmission rates ranged from 3.0 to 8.5 (representing a rate change of −0.42 to +0.05), with an average percentage rate change of 0.4%.
Inclusion of race/ethnicity and socioeconomic status in the risk-adjustment model changed the designation of the readmission rate (as better than, worse than, or no different from the overall population mean) at very few hospitals (Table II). Across all payers, the designation did not change for 99.5% of the hospitals. It changed for only 6 hospitals: to a more favorable classification for 3 and to a less favorable classification for the other 3. In the analysis of the Medicare patients, the designation did not change for 99.9% of the hospitals and changed to a less favorable classification for 1 hospital.
The addition of race/ethnicity and socioeconomic status to the risk-adjustment model resulted in a change in the excess readmission ratio at a relatively small number of hospitals (Table III). Across all payers, there was no change for 98.3% of the hospitals and a change for 20 hospitals: the change was to a more favorable ratio for 11 hospitals and to a less favorable ratio for 9. In the analysis of Medicare patients only, there was no change in the ratio for 97.6% of the hospitals and a change for 28 hospitals, with a change to a more favorable ratio for 15 hospitals and to a less favorable ratio for 13 hospitals.
The sensitivity analysis, which included only hospitals at which at least 25 arthroplasties had been performed, produced similar results (see Appendix).
In this study, we explored the impact of adding 2 patient factors, race/ethnicity and socioeconomic status, to the CMS risk-adjustment algorithm for calculating hospital readmission rates following THA and TKA. We found that these rates changed little (<1% on average) when race/ethnicity and socioeconomic status were included in the risk-adjustment algorithm. Furthermore, the rate changes affected the classifications of <1% of hospitals with regard to readmission rates in comparison with the population mean and <3% with regard to their excess readmission ratio. Classification changes that did occur were almost as likely to be unfavorable as favorable.
Our findings corroborate prior research documenting significant outcome disparities related to race/ethnicity and socioeconomic status in analyses of patient-level data10-13. However, we found that adding these measures to the CMS risk-adjustment algorithm had a relatively small effect on critical hospital impact measures. This may be explained in a number of ways. First, the method for calculating predicted values for hospital performance in the CMS model, which includes a hospital-specific intercept, may account indirectly for some differences (such as in race/ethnicity and socioeconomic status) in the populations served by hospitals in different communities. Including hospital-specific intercepts may leave little remaining variance to be explained by race/ethnicity and socioeconomic status. This is an empirical question that could be addressed in future research.
Second, the absolute difference in THA and TKA readmission rates among racial/ethnic and socioeconomic groups is relatively small. It is important to note that our findings are specific to non-urgent THAs and TKAs, which are associated with low readmission rates overall compared with other conditions20. Moreover, the CMS approach used for comparing readmission rates with population means creates relatively wide confidence intervals, making it more difficult to detect differences. The impact of race/ethnicity and socioeconomic status on readmission rates for other conditions could be different. In 1 recent study using Missouri hospital data, the addition of socioeconomic status (but not race/ethnicity) to the CMS risk-adjustment algorithm had a significant impact on readmission rates associated with other CMS HRRP conditions (acute myocardial infarction, heart failure, and pneumonia)21. Continued study of the effect of race/ethnicity and socioeconomic status on other readmission-rate performance measures is warranted.
Although we characterized the impact of race/ethnicity and socioeconomic status on readmission rates as being relatively small, the relative performance of up to 2.4% of the hospitals changed after those factors were added to the risk-adjustment model. Some stakeholders, especially the hospitals for which the performance rating changed, might not consider the results to be trivial. The extent to which our result should be considered meaningful from a policy perspective is unclear. For example, the hospitals for which the classification of the readmission rate changed may serve a higher proportion of minority patients or act as safety-net hospitals. We attempted to investigate this issue, but the number of hospitals for which the categories changed was too small for meaningful comparisons across groups. Policymakers still may deem it appropriate to use race/ethnicity and socioeconomic status to adjust readmission rates in order to address such criticisms, although our research suggests that the number of hospitals that would receive better and the number that would receive worse rankings related to readmissions following THA and TKA would be approximately equal.
Our results should be considered in light of the following limitations. First, our measure of socioeconomic status was limited by the fact that it was based on ZIP-code-level rather than patient-level data and on household income rather than other measures of socioeconomic status (for example, education). However, previous studies have used ZIP-code-level socioeconomic status to adjust hospital readmission rates21 and physician cost profiles22. Moreover, ZIP-code-level data are more likely than patient-level data to be available to most payers; thus, this approach is more consistent with the way that socioeconomic status-based risk adjustment probably would be implemented. The results of our study were similar when we applied a 6-item socioeconomic status index (described by Bird et al.23) based on aggregated 2009-2013 data from the American Community Survey (for example, median household income and percentage male unemployment).
A second limitation of our study is that race was either self-reported or assigned by hospital personnel, which is prone to error. HCUP partner organizations have worked over the last 10 years to evaluate and improve hospital-level reporting of patient race and ethnicity24.
Third, we did not have discharge data from all 50 states, so our sample cannot be considered nationally representative. However, our data set included 16 geographically dispersed states, many of which are among the largest and most diverse states in the country (for example, California and Florida).
In conclusion, hospital readmission rates following THA and TKA are increasingly included among measures used to hold hospitals accountable for their performance. Our study showed that including race/ethnicity and socioeconomic status in the CMS risk-adjustment algorithm had a minimal effect on measurement of hospitals’ readmission rates and on their performance relative to other hospitals.
Tables showing the CMS criteria for calculating risk-standardized readmission rates and the modifications used in the present study, definitions of hospital characteristics, and the effects on designations of readmission rates and hospital excess readmission ratios following THA and TKA after adjustment for race/ethnicity and socioeconomic status at hospitals with at least 25 TKAs and/or THAs are available with the online version of this article as a data supplement at jbjs.org.
NOTE: The authors gratefully acknowledge Bob Houchens, PhD (Truven Health Analytics), for providing statistical expertise; Minya Sheng, MS (Truven Health Analytics), and Timothy Kenney, MA (Kenney I.S. Consulting, Inc.), for assistance in programming and data management; and Linda Lee, PhD (Truven Health Analytics), for providing editorial review of the manuscript. They also acknowledge the 16 HCUP partner organizations that contributed to the 2011 HCUP State Inpatient Databases (SID) used in this study: Alaska State Hospital and Nursing Home Association, Arkansas Department of Health, California Office of Statewide Health Planning and Development, Florida Agency for Health Care Administration, Georgia Hospital Association, Hawaii Health Information Corporation, Massachusetts Center for Health Information and Analysis, Mississippi Department of Health, Missouri Hospital Industry Data Institute, New Mexico Department of Health, New York State Department of Health, South Carolina Revenue and Fiscal Affairs Office, Tennessee Hospital Association, Utah Department of Health, Virginia Health Information, and Washington State Department of Health.
Investigation performed at Truven Health Analytics, Inc., Santa Barbara, California
Disclosure: This work was supported by the Agency for Healthcare Research and Quality (AHRQ) Center for Delivery, Organization, and Markets, Healthcare Cost and Utilization Project (HCUP) under contract HHSA-290-2013-00002-C. The Disclosure of Potential Conflicts of Interest forms are provided with the online version of the article.
Disclaimer: The views expressed in this article are those of the authors and do not necessarily reflect those of the AHRQ or the U.S. Department of Health and Human Services.
1. Kurtz S, Ong K, Lau E, Mowat F, Halpern M. Projections of primary and revision hip and knee arthroplasty in the United States from 2005 to 2030. J Bone Joint Surg Am. 2007 ;89(4):780–5.
2. Kozma CM, Slaton T, Paris A, Edgell ET. Cost and utilization of healthcare services for hip and knee replacement. J Med Econ. 2013 ;16(7):888–96. Epub 2013 May 23.
3. Soohoo NF, Zingmond DS, Lieberman JR, Ko CY. Primary total knee arthroplasty in California 1991 to 2001: does hospital volume affect outcomes? J Arthroplasty. 2006 ;21(2):199–205.
4. SooHoo NF, Lieberman JR, Ko CY, Zingmond DS. Factors predicting complication rates following total knee replacement. J Bone Joint Surg Am. 2006 ;88(3):480–5.
5. Mahomed NN, Barrett JA, Katz JN, Phillips CB, Losina E, Lew RA, Guadagnoli E, Harris WH, Poss R, Baron JA. Rates and outcomes of primary and revision total hip replacement in the United States Medicare population. J Bone Joint Surg Am. 2003 ;85(1):27–32.
6. Mehrotra A, Sloss EM, Hussey PS, Adams JL, Lovejoy S, Soohoo NF. Evaluation of Centers of Excellence program for knee and hip replacement. Med Care. 2013 ;51(1):28–36.
10. Dailey EA, Cizik A, Kasten J, Chapman JR, Lee MJ. Risk factors for readmission of orthopaedic surgical patients. J Bone Joint Surg Am. 2013 ;95(11):1012–9.
11. Girotti ME, Shih T, Revels S, Dimick JB. Racial disparities in readmissions and site of care for major surgery. J Am Coll Surg. 2014 ;218(3):423–30. Epub 2013 Dec 19.
12. Kheir MM, Clement RC, Derman PB, Flynn DN, Speck RM, Levin LS, Fleisher LA. Are there identifiable risk factors and causes associated with unplanned readmissions following total knee arthroplasty? J Arthroplasty. 2014 ;29(11):2192–6. Epub 2014 Jul 2.
13. Singh JA, Lu X, Rosenthal GE, Ibrahim S, Cram P. Racial disparities in knee and hip total joint arthroplasty: an 18-year analysis of national Medicare data. Ann Rheum Dis. 2014 ;73(12):2107–15. Epub 2013 Sep 18.
14. Jha AK, Zaslavsky AM. Quality reporting that addresses disparities in health care. JAMA. 2014 ;312(3):225–6.
17. Improving Medicare Post-Acute Care Transformation Act of 2014, Pub L No. 113-185 (2014).
21. Nagasako EM, Reidhead M, Waterman B, Dunagan WC. Adding socioeconomic data to hospital readmissions calculations may produce more useful results. Health Aff (Millwood). 2014 ;33(5):786–91.
22. Timbie JW, Hussey PS, Adams JL, Ruder TW, Mehrotra A. Impact of socioeconomic adjustment on physicians’ relative cost of care. Med Care. 2013 ;51(5):454–60.
23. Bird CE, Seeman T, Escarce JJ, Basurto-Dávila R, Finch BK, Dubowitz T, Heron M, Hale L, Merkin SS, Weden M, Lurie N. Neighbourhood socioeconomic status and biological ‘wear and tear’ in a nationally representative sample of US adults. J Epidemiol Community Health. 2010 ;64(10):860–5. Epub 2009 Sep 16.