Total hip replacement and total knee replacement were performed in >1.1 million patients in the United States in 20131, of which >50% were Medicare cases. Because the projected volume of these cases is large, the health-care costs associated with total joint replacement have made it a common focus of attention in the Acute Care Episode (ACE) Demonstration2 and the Bundled Payment for Care Improvement (BPCI)3 projects by Medicare. The Comprehensive Care for Joint Replacement (CJR) Payment Model was launched by Medicare on April 1, 2016, for selected regions of the United States and is likely to be expanded to all at the conclusion of this trial experience4. Payment for total joint replacement will be totally bundled into a single payment for all hospital, professional, and 90-day post-discharge services. The total payment will be indexed to the Medicare-Severity, Diagnosis-Related Group (MS-DRG) assignment of the patient at hospital discharge. This program makes understanding the excess costs of total patient care essential. Because most hospitals have already improved inpatient efficiency, reductions in complications and readmissions must become the focus of attention. Providers must know their results of care across the entire 90-day post-discharge period and then must adopt care redesign methodologies to reduce complications and readmissions.
In this study, we have developed national risk-adjustment models for the adverse outcomes of care among Medicare total joint replacement patients that can be objectively measured, including inpatient mortality, prolonged inpatient length of stay as a surrogate marker for severe inpatient complications of care, 90-day post-discharge deaths without readmission, and relevant 90-day readmissions. Observed hospital performances are compared with national prediction models to benchmark outcomes and identify opportunities for improvement.
Materials and Methods
Procedure codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) of 81.51 and 81.53 for total hip replacement and 81.54 and 81.55 for total knee replacement were used to identify study subjects from the Centers for Medicare and Medicare Services (CMS) Limited Data Set for 2010 to 2012. The study population was limited to cases with a primary diagnosis ICD-9-CM code of 715.00-715.99, 996.4, 996.66, and 996.77. Acute hip fractures were excluded because of disposition and length-of-stay issues. Because they were common principal diagnoses reflecting degenerative conditions of the knee, the diagnosis codes 717.00-717.99 were also included for total knee replacement. Rheumatoid arthritis was excluded because of case complexity and the common use of immunomodulatory treatments that are unable to be tracked in administrative data. Inclusion criteria required that the procedure was performed on hospital day 0, 1, or 2, because prolonged preoperative inpatient stay increases infectious morbidity5. Exclusion criteria included patient age of <65 years, missing critical data (for example, patient or hospital identifier), patient transfers to or from another acute care hospital before total joint replacement, or discharges against medical advice. Patients with simultaneous multiple total joint replacements or a second primary total joint replacement within 90 days of the index operation were also excluded.
Two separate but overlapping Medicare Limited Data Set databases were used. The developmental database for prediction models represented only the patients from hospitals that demonstrated accurate coding practices from a screening program that we have developed6. A minimum of 20 cases per hospital are required for our control chart methods described below for model development. The final prediction models were then applied to the study database, which included all patients from all hospitals in the Medicare database with ≥50 cases that met inclusion criteria. Thus, the developmental database was a subset of the all eligible cases that met coding qualifications, and the study database was all cases regardless of coding accuracy that had ≥50 eligible cases.
Separate total hip replacement and total knee replacement models were designed. Stepwise logistic regression was used for risk models. Candidate risk factors reflected comorbid conditions used in prior studies7-9. Revision operations were included as an additional risk factor. Hospital dummy variables were employed to remove hospital-specific effects on final coefficients. Only risk factors with significance of p < 0.001 were retained in the final models. The Schwarz criterion was used to avoid overfitting models10. C-statistics were used to evaluate discrimination of final models. SAS software (version 9.4; SAS Institute) was used for all analyses.
Inpatient adverse outcomes were evaluated by models of inpatient deaths and prolonged length of stay. Prolonged length of stay is used as a composite measure of severe inpatient complications because individually coded complications are impacted by coding accuracy, inadequate surveillance, and a large number of potential events that cannot practically be risk-adjusted individually. Patients who are prolonged length-of-stay outliers were identified by first designing a linear prediction model of length of stay for patients with no coded complications (that is, routine cases). A moving-range control chart was then used to identify cases where the observed length of stay exceeds prediction by 3 standard deviations11,12. Outlier cases were removed, the 3 standard deviations were recalculated, and cases were iteratively removed until all remaining cases had lengths of stay <3 standard deviations from the predicted value. All cases that exceeded the 3 standard deviation prediction limit for excess length of stay were considered prolonged length-of-stay outliers. Once prolonged length-of-stay cases were identified, a logistic model was applied to predict that event. Prolonged length-of-stay outliers have greater costs than cases with coded complications that are not outliers13 and are strongly associated with post-discharge deaths and readmissions14-16.
Post-discharge adverse outcomes were defined for 90 days following discharge. Post-discharge deaths without readmission were identified in the Medicare data set. Readmissions were categorized by MS-DRG. Ninety-day readmissions models were designed by using cases after exclusions of patients with Major Diagnostic Categories (MDCs) 02 (diseases & disorders of the eye), 17 (myeloproliferative diseases & disorders, poorly differentiated neoplasms), 22 (burns), 24 (multiple significant trauma), and all MS-DRGs related to the management of trauma or cancer regardless of MDC. For 90-day post-discharge deaths without readmission and 90-day readmission models, prolonged length-of-stay events of the index hospitalization were also used as a variable to identify the relationship of inpatient complications with post-discharge adverse outcomes.
Final risk models were used to predict each of the 4 adverse outcomes for all hospitals in the study database. The predicted inpatient death cases were calculated, live discharges were identified and were used to calculate predicted prolonged length of stay, predicted live discharges without prolonged length of stay were then used to calculate predicted 90-day post-discharge deaths without readmission, and all live discharges that were not prolonged length of stay or 90-day post-discharge deaths without readmission were used to predict 90-day readmission cases. Once total predicted adverse outcomes were determined, a threshold was established in which all hospitals included in the comparative analysis had ≥4.5 predicted adverse outcome events. This resulted in a cut-point requirement of 50 total cases per facility. Observed adverse outcomes were totaled for cases with at least 1 event. Total predicted adverse outcomes were set equal to total observed events by multiplying predicted values by the constant of [total observed case] ÷ [total predicted cases]. The standard deviation (SD) of predicted adverse outcome values (p) for each hospital was calculated: (SD =
). For each hospital, [observed − predicted adverse outcomes] ÷ [hospital-specific standard deviation] was calculated to give a z-score. Negative z-scores represented observed adverse outcomes that were better than predicted, and positive z-scores were poorer than predicted. Risk-adjusted outcomes for each hospital were calculated by the equation (overall observed adverse outcomes rates) × (hospital observed adverse outcomes/hospital predicted adverse outcomes) and were stratified by decile. Hospital performance was then stratified into deciles by volumes of cases to identify relationships between hospital case volume and risk-adjusted outcomes.
Total Hip Replacement
The 4 prediction models derived from the developmental database for total hip replacement are summarized in Table I. Hospital influences on the c-statistics were least with the 90-day readmissions model. The risk factors for total hip replacement models are detailed in the Appendix. Revision procedures were significant variables (p < 0.001) for both prolonged length of stay (odds ratio [OR], 1.58) and 90-day readmissions (OR, 2.40). Prolonged length of stay was a risk factor for post-discharge deaths without readmission (OR, 5.05) and 90-day readmissions (OR, 1.99).
The MS-DRGs of total hip replacement readmissions in the developmental database are identified in Table II. The most frequent MS-DRGs of readmissions were technical complications, cardiopulmonary events, and infections. Among all readmissions, 45% occurred in the interval of time from 31 to 90 days.
There were 253,978 eligible total hip replacement cases from 1,483 hospitals with ≥50 cases. There were a mean of 171 cases and a median of 122 cases per hospital. There were 233 inpatient deaths (0.09%) and 11,676 patients (4.6%) with prolonged length of stay. A total of 666 patients (0.26%) died within 90 days after discharge without readmission and 22,337 patients (8.8%) were readmitted within 90 days; of these patients, 383 (0.15%) died following readmission. A total of 1,282 deaths (0.50%) occurred across the full study period. There were 30,416 total cases (12.0%) with ≥1 adverse outcomes.
The total hip replacement outcomes by 1,483 qualifying hospitals demonstrated a range of z-scores from −5.09 to +5.62. There were 98 hospitals that were 2 standard deviations (z-score < −1.96) better than expected, and 142 hospitals were 2 standard deviations poorer than expected. In Figure 1, the risk-adjusted median adverse outcome rates by hospital decile are illustrated. Error bars reflect the interquartile range. The best-performing first decile had a risk-adjusted adverse outcome rate of 6.6%, and the poorest-performing tenth decile was at 19.8%.
In Figure 2, the adverse outcome performance is illustrated by decile of hospital case volume. The first decile (lowest volume) had a range between 50 and 59 cases per hospital and a mean adverse outcome rate of 13.3%. The tenth decile (highest volume) hospitals had >337 cases per hospital and a risk-adjusted adverse outcome rate of 11.2%. The correlation coefficient of improved risk-adjusted adverse outcome rates by increasing hospital volume of cases was 0.094 (p < 0.001).
Total Knee Replacement
The details of the 4 prediction models for total knee replacement in the developmental database are shown in Table I. As with the total hip replacement models, the hospital influence on the c-statistics was least with the 90-day readmission model for total knee replacement. Risk factors of significance (p < 0.001) are presented in the Appendix. Total knee replacement revision procedures were a significant risk factor for 90-day readmissions (OR, 1.64), but not for prolonged length of stay. Prolonged length of stay was a significant risk factor (p < 0.05) for 90-day post-discharge death without readmission (OR, 5.18) and for 90-day readmissions (OR, 1.96).
The MS-DRGs of total knee replacement readmissions are identified in Table III. Compared with total hip replacement, the most frequent MS-DRGs of readmission for total knee replacement were cardiopulmonary events and infections. A total of 55.4% of total knee replacement readmissions occurred in the first 30 days following discharge.
For total knee replacement, there were 672,515 qualifying cases from 2,349 hospitals with ≥50 cases in the study database. There were a mean of 286 cases and a median of 195 cases per hospital. There were 615 inpatient deaths (0.09%) and a total of 30,951 cases (4.6%) of prolonged length of stay. A total of 1,322 patients (0.2%) died within 90 days after discharge without readmission. There were 50,772 cases (7.6%) of readmissions by 90 days following discharge; of these patients, 806 (1.6% of readmissions) died following readmission. A total of 2,743 patients (0.4%) died across the inpatient and 90-day post-discharge period. There were a total of 77,823 patients (11.6%) with ≥1 adverse outcomes.
The performance by 2,349 hospitals for total knee replacement identified the best z-scores of −5.85 to the poorest of +11.75. There were 223 hospitals that were 2 standard deviations better than predicted, and 319 hospitals were 2 standard deviations poorer than predicted. In Figure 1, the median risk-adjusted adverse outcome rates for total knee replacement are illustrated by hospital decile. The error bars demonstrate the interquartile range within each decile. The best-performing first decile had a risk-adjusted adverse outcome rate of 6.4%, and the poorest-performing tenth decile was at 19.3%.
In Figure 2, the mean hospital adverse outcome performance of total knee replacement by decile of case volume is presented. The first decile (lowest volume) had a range of 50 to 67 cases per hospital and a mean adverse outcome rate of 12.4%. The tenth decile (highest volume) hospitals had >611 cases and had a risk-adjusted mean adverse outcome rate of 11.0%. The correlation coefficient of improved risk-adjusted adverse outcome rates by increasing hospital volume of total knee replacement cases was 0.13 (p < 0.001).
The comparisons in Figures 1 and 2 between total hip replacement and total knee replacement illustrate that outcomes by decile of hospital performance and by decile of hospital case volume are quite similar. Differences in causes for readmission were that technical complications were most common for total hip replacement and cardiopulmonary events were most common for total knee replacement17.
The risk-adjusted total joint replacement outcomes in this study identified wide variation. The range of risk-adjusted outcomes for both total hip replacement and total knee replacement exceeded 10 standard deviations. Nearly 10% of hospitals performing total hip replacement and >13% of hospitals performing total knee replacement were 2 standard deviations poorer than expected. Improvement of suboptimally performing hospitals to the mean level would have a dramatic impact on patient morbidity and costs.
Traditional measurement of surgical outcomes in total joint replacement has included inpatient events and 30-day post-procedural deaths. Death rates in total hip replacement and total knee replacement are very low even when 90-day post-discharge care is included. Death rates following elective surgical procedures, especially total joint replacement, are not sensitive indicators of quality outcomes18,19. The trend to shorter lengths of stay and early transfer of frail patients to skilled nursing and rehabilitation facilities means that many complications of care are not identified until after discharge20. Only patients with the most severe inpatient complications had and will likely have prolonged inpatient stays. Earlier discharge has increased attention to readmissions as a measure of outcomes.
The debate over readmission as an outcome metric has centered on the length of post-discharge time for measurement. We have chosen 90-day readmission. Most studies of readmissions have used 30 postoperative days21-24, although some have extended the readmission threshold to 90 days25,26. The 90-day readmission rate is the most common adverse outcome identified in this study and will be an important outcome in CJR. Because individual patients are not readmitted multiple times within the initial 30 days following discharge, the 30-day columns in Tables II and III are useful for comparing our readmission results with the results of other reports of 30-day readmissions.
The extension of the postoperative period for outcomes will result in unrelated readmissions being counted. Although a small number of additional readmission events still need to be excluded from future analyses, the current exclusions have defined important causes of readmissions. The readmission causes identified in Tables II and III are linked to complications of care and destabilization of pre-existing chronic medical conditions, even for events at 31 to 90 days.
The causes of readmissions provide a template for management and prevention in the redesign of postoperative care. With better definition of causes, focused improvement initiatives by providers will change outcomes. Critical review of technical failures in prosthetic placement, standardization of antibiotic and infection control practices, expanded use of non-narcotic pain management strategies to avoid gastrointestinal-associated readmissions, and careful monitoring of the male urinary tract for retention and infection all become focused methods to avoid readmissions. Better systems for following patients after discharge with more frequent follow-up visits, telephone calls, and use of social media to identify potential problems before readmission are necessary. In this study, some hospitals were doing better than others in this regard.
Conflicting information has been reported about hospital case volume and outcomes of care. Some have found that volume makes a difference in surgical mortality rates27, and others have not28. Our study demonstrates a small but valid difference between the top and bottom-volume decile groups. The major differences in top and bottom-performing hospitals in total joint replacement were not explained by hospital volume alone.
This study had some limitations. First, it used administrative data and not clinical information. Although the accuracy of administrative data is always questioned, we did use best-coding hospitals for model development. It will be administrative and not clinical data that will inform CJR. The evolution of the electronic medical record should permit the inclusion of clinical information (for example, admission laboratory data)29,30 with administrative information to refine assessment of adverse outcomes in the future. Administrative data have the advantage of defining the post-discharge adverse events that are missing or unknown in clinical databases. Up to 40% of surgical readmissions through 90 post-discharge days occur at facilities other than the index hospital and may not be appreciated by providers31.
A limitation of our exclusions for readmissions was that they are comparable but not identical to those that will be used in CJR. An important difference is that we have included revision joint replacements, and CJR does not include them. Consistent with the Medicare CJR program, we have included all medical readmissions, which are problematic when orthopaedic providers will be expected to assume economic accountability for these readmissions. The exclusion of trauma and cancer readmissions is similar but not exactly the same as CJR. Our study excluded acute fractures and these will be included in CJR with a separate payment structure from non-fracture total joint replacements. Finally, our study included only patients ≥65 years of age, but CJR will retain all Medicare patients including disability patients.
An additional limitation was early discharge to skilled nursing facilities for major complications of inpatient care, which potentially escapes our prolonged length-of-stay outlier method. The quality of care provided by skilled nursing becomes an additional variable in the outcome equation and requires that providers in CJR will need to evaluate outcomes and costs at post-acute care facilities32,33. Finally, emergency department visits without readmission remain another element to be considered in post-discharge adverse outcomes.
CJR is only the first of the bundled payments that will affect orthopaedic and other surgical care. Private payers are approaching providers to participate in similar bundled care arrangements. These payment models will make it necessary for surgeons and hospitals to know their outcomes data across the full continuum of care. With readmissions and emergency department visits occurring at hospitals other than where the procedure was performed34, better systems are needed to track patient outcomes. The success of providers in this new environment will require knowledge about the variation of risk-adjusted hospital performance and the adoption of measures that will improve outcomes.
Tables showing the prediction models for inpatient mortality, prolonged length of stay, 90-day post-discharge deaths without readmission, and 90-day readmissions in the developmental database for total hip and knee replacement in Medicare patients are available with the online version of this article as a data supplement at jbjs.org.
Investigation performed at MPA Healthcare Solutions, Chicago, Illinois
Disclosure: There was no source of external funding for this study. On the Disclosure of Potential Conflicts of Interest forms, which are provided with the online version of the article, one or more of the authors checked “yes” to indicate that the author had other relationships or activities that could be perceived to influence, or have the potential to influence, what was written in this work.
5. Vogel TR, Dombrovskiy VY, Lowry SF. In-hospital delay of elective surgery for high volume procedures: the impact on infectious complications. J Am Coll Surg. 2010 ;211(6):784–90. Epub 2010 Oct 25.
6. Pine M, Fry DE, Jones B, Meimban R. Screening algorithms to assess the accuracy of present-on-admission coding. Perspect Health Inf Manag. 2009;6:2. Epub 2009 Feb 11.
7. Fry DE, Pine M, Jones BL, Meimban RJ. Surgical warranties to improve quality and efficiency in elective colon surgery. Arch Surg. 2010 ;145(7):647–52.
8. Pine M, Fry DE, Jones BL, Meimban RJ, Pine GJ. Controlling costs without compromising quality: paying hospitals for total knee replacement. Med Care. 2010 ;48(10):862–8.
9. Fry DE, Pine M, Locke D, Pine G. Prediction models of Medicare 90-day postdischarge deaths, readmissions, and costs in bowel operations. Am J Surg. 2015 ;209(3):509–14. Epub 2014 Dec 20.
10. Schwarz GE. Estimating the dimension of a model. Ann Stat. 1978;6(2):461–4.
11. Fry DE, Pine M, Jones BL, Meimban RJ. Control charts to identify adverse outcomes in elective colon resection. Am J Surg. 2012 ;203(3):392–6; discussion 396. Epub 2011 Dec 28.
12. Koetsier A, van der Veer SN, Jager KJ, et al.. Control charts in healthcare quality improvement: a systematic review on adherence to methodologic criteria. Methods Inf Med. 2012;51(3):189–98.
13. Fry DE, Pine M, Jones BL, Meimban RJ. Adverse outcomes in surgery: redefinition of postoperative complications. Am J Surg. 2009 ;197(4):479–84. Epub 2009 Feb 26.
14. Fry DE, Pine M, Pine G. Medicare post-discharge deaths and readmissions following elective surgery. Am J Surg. 2014 ;207(3):326–30, discussion: 330. Epub 2013 Dec 8.
15. Fry DE, Pine M, Locke D, Pine G. Composite measurement of outcomes in Medicare inpatient laparoscopic cholecystectomy. J Am Coll Surg. 2015 ;221(1):102–9. Epub 2015 Mar 4.
16. Fry DE, Pine M, Locke D, Reband A, Torres Z, Pine G. Medicare inpatient and 90-day postdischarge adverse outcomes in carotid artery surgery. Surgery. 2015 ;158(4):1056–62, discussion: 1062-4. Epub 2015 Jul 7.
17. 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.
18. Lilford R, Pronovost P. Using hospital mortality rates to judge hospital performance: a bad idea that just won’t go away. BMJ. 2010 ;340:c2016.
19. Hogan H, Zipfel R, Neuburger J, Hutchings A, Darzi A, Black N. Avoidability of hospital deaths and association with hospital-wide mortality ratios: retrospective case record review and regression analysis. BMJ. 2015 ;351:h3239.
20. Cram P, Lu X, Kates SL, Singh JA, Li Y, Wolf BR. Total knee arthroplasty volume, utilization, and outcomes among Medicare beneficiaries, 1991-2010. JAMA. 2012 ;308(12):1227–36.
21. Vorhies JS, Wang Y, Herndon J, Maloney WJ, Huddleston JI. Readmission and length of stay after total hip arthroplasty in a national Medicare sample. J Arthroplasty. 2011 ;26(6)(Suppl):119–23. Epub 2011 Jul 1.
22. Mesko NW, Bachmann KR, Kovacevic D, LoGrasso ME, O’Rourke C, Froimson MI. Thirty-day readmission following total hip and knee arthroplasty - a preliminary single institution predictive model. J Arthroplasty. 2014 ;29(8):1532–8. Epub 2014 Mar 4.
23. Avram V, Petruccelli D, Winemaker M, de Beer J. Total joint arthroplasty readmission rates and reasons for 30-day hospital readmission. J Arthroplasty. 2014 ;29(3):465–8. Epub 2013 Aug 28.
24. Pugely AJ, Callaghan JJ, Martin CT, Cram P, Gao Y. Incidence of and risk factors for 30-day readmission following elective primary total joint arthroplasty: analysis from the ACS-NSQIP. J Arthroplasty. 2013 ;28(9):1499–504. Epub 2013 Jul 26.
25. Zmistowski B, Restrepo C, Hess J, Adibi D, Cangoz S, Parvizi J. Unplanned readmission after total joint arthroplasty: rates, reasons, and risk factors. J Bone Joint Surg Am. 2013 ;95(20):1869–76.
26. Schairer WW, Sing DC, Vail TP, Bozic KJ. Causes and frequency of unplanned hospital readmission after total hip arthroplasty. Clin Orthop Relat Res. 2014 ;472(2):464–70.
27. Reames BN, Ghaferi AA, Birkmeyer JD, Dimick JB. Hospital volume and operative mortality in the modern era. Ann Surg. 2014 ;260(2):244–51.
28. Khuri SF, Henderson WG. The case against volume as a measure of quality of surgical care. World J Surg. 2005 ;29(10):1222–9.
29. Pine M, Jordan HS, Elixhauser A, Fry DE, Hoaglin DC, Jones B, Meimban R, Warner D, Gonzales J. Enhancement of claims data to improve risk adjustment of hospital mortality. JAMA. 2007 ;297(1):71–6.
30. Fry DE, Pine M, Jordan HS, Elixhauser A, Hoaglin DC, Jones B, Warner D, Meimban R. Combining administrative and clinical data to stratify surgical risk. Ann Surg. 2007 ;246(5):875–85.
31. Kim Y, Gani F, Lucas DJ, Ejaz A, Spolverato G, Canner JK, Schneider EB, Pawlik TM. Early versus late readmission after surgery among patients with employer-provided health insurance. Ann Surg. 2015 ;262(3):502–11; discussion 509-11.
32. Ottenbacher KJ, Karmarkar A, Graham JE, Kuo YF, Deutsch A, Reistetter TA, Al Snih S, Granger CV. Thirty-day hospital readmission following discharge from postacute rehabilitation in fee-for-service Medicare patients. JAMA. 2014 ;311(6):604–14.
33. Ramos NL, Karia RJ, Hutzler LH, Brandt AM, Slover JD, Bosco JA. The effect of discharge disposition on 30-day readmission rates after total joint arthroplasty. J Arthroplasty. 2014 ;29(4):674–7. Epub 2013 Oct 30.
34. Cloyd JM, Chen J, Ma Y, Rhoads KF. Association between weekend discharge and hospital readmission rates following major surgery. JAMA Surg. 2015 ;150(9):849–56.