Health Services and Outcomes Research. Development of a Clinical Registry-Based 30-Day Readmission Measure for Coronary Artery Bypass Grafting Surgery

Transcription

1 Health Services and Outcomes Research Development of a Clinical Registry-Based 30-Day Readmission Measure for Coronary Artery Bypass Grafting Surgery David M. Shahian, MD; Xia He, MS; Sean M. O Brien, PhD; Frederick L. Grover, MD; Jeffrey P. Jacobs, MD; Fred H. Edwards, MD; Karl F. Welke, MD; Lisa G. Suter, MD; Elizabeth Drye, MD, SM; Cynthia M. Shewan, PhD; Lein Han, PhD; Eric Peterson, MD, MPH Background Reducing readmissions is a major healthcare reform goal, and reimbursement penalties are imposed for higherthan-expected readmission rates. Most readmission risk models and performance measures are based on administrative rather than clinical data. Methods and Results We examined rates and predictors of 30-day all-cause readmission following coronary artery bypass grafting surgery by using nationally representative clinical data ( ) from the Society of Thoracic Surgeons National Database linked to Medicare claims records. Among eligible Medicare records, (86%) were successfully linked to Society of Thoracic Surgeons records; (61%) isolated coronary artery bypass grafting admissions constituted the study cohort. Logistic regression was used to identify readmission risk factors; hierarchical regression models were then estimated. Risk-standardized readmission rates ranged from 12.6% to 23.6% (median, 16.8%) among 846 US hospitals with 30 eligible cases and 90% of eligible Centers for Medicare and Medicaid Services records linked to the Society of Thoracic Surgeons database. Readmission predictors (odds ratios [95% confidence interval]) included dialysis (2.02 [ ]), severe chronic lung disease (1.58 [ ]), creatinine (2.5 versus 1.0 or lower:1.49 [ ]; 2.0 versus 1.0 or lower: 1.37 [ ]), insulin-dependent diabetes mellitus (1.45 [ ]), obesity in women (body surface area 2.2 versus 1.8: 1.44 [ ]), female sex (1.38 [ ]), immunosuppression (1.38 [ ]), preoperative atrial fibrillation (1.36 [ ]), age per 10-year increase (1.36 [ ]), recent myocardial infarction (1.24 [ ]), and low body surface area in men (1.22 [ ]). C-statistic was Fifty-two hospitals (6.1%) had readmission rates statistically better or worse than expected. Conclusions A coronary artery bypass grafting surgery readmission measure suitable for public reporting was developed by using the national Society of Thoracic Surgeons clinical data linked to Medicare readmission claims. (Circulation. 2014;130: ) Key Words: coronary artery bypass patient readmission registries risk adjustment The reduction of hospital readmissions is a major goal of healthcare reform and has been the focus of Medicare Payment Advisory Commission reports. 1,2 The Patient Protection and Affordable Care Act incorporates a Hospital Readmissions Reduction Program, which includes public reporting of risk-standardized readmission rates and progressively increasing reimbursement penalties for hospitals with higher-than-expected unplanned readmission rates. Bundled payment initiatives may create additional incentives for improved discharge planning and postdischarge care coordination by making the index hospital financially responsible for subsequent readmissions. 3 Clinical Perspective on p 409 Adjustment for differences in case mix is necessary to fairly implement readmission reduction initiatives and avoid inappropriately penalizing providers or misleading consumers. 4 Previous readmission risk models developed for use by the Centers for Medicare and Medicaid Services (CMS) have been based on Medicare administrative claims data, 5 7 and a claims-based Received November 15, 2013; accepted May 27, From the Massachusetts General Hospital and Harvard Medical School, Boston, MA (D.M.S.); Duke Clinical Research Institute, Durham, NC (X.H., S.M.O., E.P.); University of Colorado School of Medicine-Anschutz Medical Campus, Aurora, CO, and Denver Department of Veterans Affairs Medical Center, Denver, CO (F.L.G.); All Children s Hospital, John Hopkins University, Saint Petersburg, FL (J.P.J.); University of Florida College of Medicine, Jacksonville, FL (F.H.E.); Children s Hospital of Illinois and the University of Illinois College of Medicine, Peoria, IL (K.F.W.); Yale-New Haven Health Services Corporation Center for Outcomes Research and Evaluation (CORE) and Yale School of Medicine, New Haven, CT (L.G.S., E.D.); Society of Thoracic Surgeons, Chicago, IL (C.M.S.); and Centers for Medicare and Medicaid Services, Baltimore, MD (L.H.). The online-only Data Supplement is available with this article at /-/DC1. Correspondence to David M. Shahian, MD, Department of Surgery and Center for Quality and Safety, Massachusetts General Hospital, 55 Fruit St, Boston, MA dshahian@partners.org 2014 American Heart Association, Inc. Circulation is available at DOI: /CIRCULATIONAHA

2 400 Circulation July 29, 2014 coronary artery bypass grafting surgery (CABG) readmission measure is under development. However, some previous CABG studies have raised concerns about the accuracy of profiling measures based on administrative data 8,9 because of their inadequate risk adjustment, case misclassification, and other issues. Although claims-based risk adjustment, with its inherent limitations, may be the only measure development option for many conditions, the vast majority of CABG procedures performed in the United States are entered into the Society of Thoracic Surgeons (STS) Adult Cardiac Surgery Database. This registry contains detailed clinical data needed for robust risk adjustment. With the use of previously described algorithms, 10 it has also been successfully linked to various claims data sources that can be used to determine readmissions, longterm outcomes, and resource use. Accordingly, STS and the Duke Clinical Research Institute, in collaboration with the Yale Center for Outcomes Research and Evaluation group and CMS, have developed a 30-day all-cause readmission measure based on STS Adult Cardiac Surgery Database clinical data linked to Medicare readmission data. Methods Data Sources The STS Adult Cardiac Surgery Database provides distinct advantages in comparison with administrative data. These include uniform cohort definitions, standardized risk factor specifications designed by surgical content experts, and rigorous quality control. Eight to ten percent of participating sites undergo extensive annual external audits, and data accuracy is high (97% overall). These advantages are shared by other robust clinical registries such as those maintained by the State of New York and used for public reporting. 11 STS Adult Cardiac Surgery Database data elements include preoperative clinical characteristics, operative techniques, complications, and mortality. With the use of these data, risk-adjusted outcomes feedback reports are provided to >1050 cardiac surgery programs (90% 95% of all US programs). The Medicare Part A inpatient database contains fee-for-service claims for reimbursement of hospital inpatient facility costs. Medicare data elements used for this project include dates of admission and discharge, patient demographics, International Classification of Diseases, Ninth Revision diagnosis and procedure codes, and hospital provider number. The Medicare Enrollment Database contains patient demographic information, vital status, and monthly indicators of fee-for-service eligibility (to implement measure inclusion/exclusion criteria). Cohort Development, Linkage Methodology, and Exclusions Derivation of the isolated CABG study cohort and linkage of STS and Medicare data are summarized in Figure 1 and described in detail in online-only Data Supplement Appendix I. The STS algorithm for defining an isolated CABG in Database version 2.61 (used for these analyses) is available on the STS website. 12 Comparison of CMS and STS records that did or did not link is provided in Tables I and II in the online-only Data Supplement. From 2008 to 2010, a total of CMS admissions from 1172 hospitals met eligibility criteria to be considered for linkage to the STS database. Among index CABG admissions from 1012 STS-participating hospitals, admissions linked to an STS record and did not link. Nonlinked patients were more often female (36.2% versus 31.9%), black (6.8% versus 4.6%), and more likely to be readmitted (20.2% versus 18.6%). We also performed a complementary analysis of 2008 to 2010 CMS admissions that could be linked to STS records of patients 65 years of age who met the definition of isolated CABG. Among STS records at 1024 centers that submitted CMS claims during the study period, linked to a CMS admission and did not. Nonlinked records were more often male (70.6% versus 69.5%), black (6.6% versus 5.3%), and diabetic patients with smaller body surface areas. They less often had chronic lung disease, peripheral vascular disease, or cerebrovascular disease, and more often had creatinine levels of <1 mg/dl. They were more often undergoing urgent or emergent procedures. Outcome The measure outcome is 30-day all-cause readmission, defined as a subsequent admission to an acute-care facility on or before the 30th day after the date of discharge (day 0) from the index CABG. Admissions with a primary diagnosis of care involving use of rehabilitation procedures (v57.x) were assumed to be to rehabilitation units and were not counted as readmissions for this measure, nor were Figure 1. Definition of study sample. CABG indicates coronary artery bypass grafting; CMS, Centers for Medicare and Medicaid Services; ICD-9, International Classification of Diseases, Ninth Revision; and STS, Society of Thoracic Surgeons.

3 Shahian et al CABG Readmission Measure Based on Clinical Data 401 Table 1. Distribution of Patient Baseline Characteristics in Measure Cohort Readmission = Yes Readmission = No Variable Level n % n % Total records Demographics Age, y 65 and < and < Sex Male Female Race/ethnicity White Black Hispanic Asian Risk factors Body surface area (m 2 ) < Diabetes mellitus No diabetes mellitus Diabetes mellitus noninsulin Diabetes mellitus insulin Hypertension No Yes Renal function, mg/dl Creatinine < Creatinine Creatinine Creatinine Creatinine Dialysis Chronic lung disease None Mild Moderate Severe Peripheral vascular disease No Yes Cerebrovascular disease No Yes Cerebrovascular accident No CVA Remote CVA (> 2 wk) Recent CVA ( 2 wk) Immunosuppressive treatment No Yes Previous CV interventions Number of previous CV surgeries No previous CV surgery prior CV surgery prior CV surgeries Prior PCI No PCI PCI - within 6 h (Continued )

4 402 Circulation July 29, 2014 Table 1. Continued Readmission = Yes Readmission = No Variable Level n % n % Preoperative cardiac status Acuity status Elective Urgent Emergent Emergent salvage Myocardial infarction No prior MI MI > 21 days MI 8 21 days MI 1 7 days MI > 6 and < 24 h MI 6 h Angina No Yes Arrhythmia No arrhythmia AFib/flutter Heart block Sustained VT/VF Multiple types Preop IABP No Yes Congestive heart failure No CHF CHF NYHA-I CHF NYHA-II CHF NYHA-III CHF NYHA-IV CHF missing NYHA Number of diseased coronary None vessels One Two Three Left main disease > 50% No Yes Ejection fraction, % < Aortic insufficiency None Trivial Mild Moderate Severe N/A Mitral insufficiency None Trivial Mild Moderate Severe N/A (Continued )

5 Shahian et al CABG Readmission Measure Based on Clinical Data 403 Table 1. Continued Readmission = Yes Readmission = No Variable Level n % n % Tricuspid insufficiency None Trivial Mild Moderate Severe N/A Although the differences in readmission rates were small between groups, because of the large sample sizes, all P values were < (χ 2 tests of null hypotheses that no association existed between risk factors and readmission) except for the number of diseased coronary vessels (P=0.03) and left main disease >50% (P=0.007). AFib indicates atrial fibrillation; CHF, congestive heart failure; CV, cardiovascular; CVA, cerebrovascular accident; IABP, intra-aortic balloon pump; MI, myocardial infarction; N/A, not available; NYHA, New York Heart Association; PCI, percutaneous coronary intervention; and VT/VF, ventricular tachycardia/ ventricular fibrillation. transfers from the index CABG hospital to another acute-care facility. Follow-up of transfer patients commenced on the day of discharge from the last hospital in the transfer chain. Readmissions were always attributed to the hospital that initially performed the CABG. Risk Adjustment Candidate risk adjustment covariates were selected based on prior literature including STS 2008 CABG mortality and morbidity models 13 and previous CMS readmission measures. 5 7 Race and ethnicity were candidate covariates in previous STS models but were excluded from the current measure, in accordance with National Quality Forum measure criteria and CMS practice, to avoid potentially masking care disparities. Candidate and final model covariates, and their mathematical representation in the model, are summarized in Table III and Appendix II in the online-only Data Supplement. Variable Selection The final list of covariates selected by the surgeon panel included all covariates that were either (1) selected at the 0.05 level in the original full sample for at least 1 calendar year; or (2) were selected in at least 50% of bootstrap replicates at the 0.05 level in at least 1 calendar year. Details of the variable selection process are provided in onlineonly Data Supplement Appendix III. Assessment of Model Calibration Before estimating hospital performance, we assessed the adequacy of the proposed case mix adjustment model. Coefficients from the final marginal model were reestimated by using only 2008 data, and then tested in a separate sample of 2009 data. To assess calibration, we compared observed versus expected all-cause 30-day readmission rates within patient subgroups based on deciles of predicted risk in 2009 data. We also used 2009 data to estimate the C-index of the model estimated from 2008 data. Although a low C-index does not imply that the model is misspecified or that hospital comparisons will be biased, it may serve as a benchmark for comparing alternative models for the same end point in the same target population. Missing Data Missing data for model covariates were rare, <1% of records for most variables. Candidate covariates with the most missing data were ejection fraction (3.6%), aortic stenosis (0.7%), and aortic insufficiency (0.6%). Missing data were imputed to the most common category of categorical variables and to the median or subgroup-specific median of continuous variables. Multiple imputation was not used because it had little impact in previous STS risk model analyses. 13 Hospital-Specific Risk-Standardized Readmission Rates To estimate hospital-specific performance, the selected covariates were entered into a hierarchical logistic regression model with hospital-specific random intercept parameters. This approach explicitly models hospital-level differences while adjusting for case mix. These hierarchical models were used to estimate hospital-specific risk-standardized readmission rates (RSRRs) using methodologies identical to the existing CMS readmission measures. 5 7 RSRRs were calculated for each hospital as the ratio of their predicted (analogous to observed, but incorporating both case mix and an estimate of hospital-specific effect) to their expected number of readmission events (calculated by using the national average effect in place of the hospital-specific effect), multiplied by the national unadjusted readmission rate (see online-only Data Supplement Appendix IV). This hierarchical RSRR is analogous to the commonly used ratio of observed-toexpected outcomes. It reflects the performance of a specific hospital with its unique mix of patients in comparison with the performance of an average hospital having the same patient mix. Hospitals were classified as better than expected if the 95% interval of their RSRR fell entirely below the aggregate readmission rate, as worse than expected if the 95% interval of the RSRR fell entirely above the aggregate readmission rate, and same as expected if their 95% interval overlapped the overall aggregate readmission rate. Measure Reliability An important criterion for evaluating hospital performance measures is the extent to which between-hospital variation in the measure is explained by true differences versus chance variation (ie, signal versus noise). 14 Measures with a high proportion of signal variance are more reliable and useful because of their higher power to discriminate high and low performers. Measures dominated by chance variation may lead to unfair conclusions and misinform consumers. To assess the reliability of the CABG readmission measure, we estimated the percentage of overall variation explained by true signal rather than random noise, using a Bayesian version of the hierarchical logistic regression model and WinBUGS software (see online-only Data Supplement Appendix V). Results Among eligible Medicare patients, (86%) were successfully linked to an STS record; (61%) index CABG admissions from 1012 CMS providers met the project definition of isolated CABG (Figure 1) and were included in the final cohort (Table 1). These data are therefore estimated to have been derived from 86% of all CMS CABG

6 404 Circulation July 29, 2014 Table 2. Estimated Odds Ratios (95% Confidence Intervals) From Final Marginal Model (Estimated From Data) Effect OR (95% CI) P Value Ejection fraction per 10 U decrease 1.07 ( ) < Preoperative atrial fibrillation 1.36 ( ) < Unstable angina (no MI 7 days) 1.05 ( ) Congestive heart failure 1.17 ( ) < Age per 10-y increase 1.36 ( ) < Dialysis and creatinine* Dialysis vs no dialysis and creatinine = 1.0 or lower 2.02 ( ) < Creatinine 1.5 vs ( ) < Creatinine 2.0 vs ( ) < Creatinine 2.5 vs ( ) < Status (vs elective) Urgent 1.09 ( ) < Emergent/emergent salvage 1.17 ( ) Female (at BSA=1.8) vs male (at BSA=2.0) 1.38 ( ) < Reoperation (vs no previous operation) 1 previous operation 1.14 ( ) < Chronic lung disease (vs none) Mild 1.21 ( ) < Moderate 1.35 ( ) < Severe 1.58 ( ) < Diabetes mellitus (vs no diabetes mellitus) Noninsulin diabetes mellitus 1.11 ( ) < Insulin diabetes mellitus 1.45 ( ) < Preop IABP or inotrope 1.08 ( ) Immunosuppressive treatment 1.38 ( ) < Peripheral vascular disease 1.20 ( ) < MI (vs MI > 21 days or no MI) 1 21 days 1.14 ( ) < > 6 and < 24 h 1.11 ( ) h 1.24 ( ) BSA 1.6 vs 2.0 in male 1.22 ( ) < vs 2.0 in male 1.04 ( ) vs 2.0 in Male 1.07 ( ) < vs 1.8 in female 0.99 ( ) vs 1.8 in female 1.13 ( ) < vs 1.8 in female 1.44 ( ) < Surgery date per half-year increase 0.99 ( ) Cerebrovascular disease 1.14 ( ) < Hypertension 1.07 ( ) PCI 6 h 1.10 ( ) Left main disease 0.97 ( ) AFib indicates atrial fibrillation; BSA, body surface area; CHF, congestive heart failure; CI, confidence interval; IABP, intra-aortic balloon pump; MI, myocardial infarction; OR, odds ratio; PCI, percutaneous coronary intervention; and VT/VF, ventricular tachycardia/ventricular fibrillation. *The final model assumed that creatinine in range 0 1 had the same effect on readmission, and that different linear effects (on log-odds scale) existed between range and 1.5 +; the odds ratios were reported for a few chosen points on the continuous scale to illustrate the overall nonlinear relationship. The final model included the quadratic terms of BSA, and interactions between BSA and sex; the odds ratios were reported for chosen points on the continuous scale for each sex group to illustrate the overall nonlinear relationship and the differences between groups.

7 Shahian et al CABG Readmission Measure Based on Clinical Data 405 providers (1210 original hospitals minus 38 hospitals with <10 CABG (isolated or combined) over 3 years=1172 likely true CMS CABG hospitals; 1012/1172=86%). The median number of 2008 to 2010 eligible isolated CABG admissions per hospital, after all exclusions, was (range, ; interquartile range, ). Among isolated CABG admissions, (16.8%) were readmitted within 30 days. Comparing risk factor prevalences between patients who were or were not readmitted, all differences were statistically significant (P<0.05) owing to the large sample sizes. Risk Adjustment Covariates Odds ratios (95% CI) from the marginal logistic model based on 2008 to 2010 data are summarized in Table 2. Important predictors included preoperative atrial fibrillation, age, elevated creatinine or dialysis, female sex, severe chronic lung disease, insulin-dependent diabetes mellitus, immunosuppressive therapy, myocardial infarction within 6 hours preoperatively, low body surface area in men, and obesity in women. Validation When the 2008 model was applied to 2009 data, patient-level predicted risk estimates ranged from 8.9% in the lowest decile to 31.9% in the highest decile of predicted risk. There was excellent agreement between predicted and observed readmission rates across deciles of predicted risk (Figure 2). C-index was Hierarchical Model With the use of 2008 to 2010 data, index admissions from 1012 hospitals were included in the estimation of the final hierarchical model. Odds ratio estimates from the hierarchical version of the model were nearly identical to those of the marginal model (Table IV in the online-only Data Supplement). When risk estimates were calculated by using each hospital s own estimated intercept parameter, the C-statistic of this model was Risk-Standardized Readmission Rates Figure 3 displays the distribution of hospital-specific RSRRs for the subset of 846 hospitals with at least 30 eligible cases and at least 90% of eligible CMS records linked to the STS database. Estimated RSRRs ranged from 12.6% to 23.6% (median, 16.8%; interquartile range, 15.6% 18.1%). Fiftytwo hospitals (6.1%) were identified as having readmission rates statistically different from expected (25 better than expected, 27 worse than expected). Measure Reliability In hospitals with at least 30 cases over 3 years (a more stringent requirement to ensure adequate sample size), the proportion of variation in RSRRs attributable to true signal variation (versus random noise) was 47%, comparable to or higher than that of other CMS readmission measures. In hospitals with larger case volumes, the reliability was higher (49%, 54%, 64%, and 71% for hospitals with at least 50, 100, 200, and 300 cases, respectively). Discussion Major goals of healthcare reform include enhanced care coordination across providers 15 and reduction of unplanned readmissions through more effective peridischarge processes and interventions. 16,17 Strategies to achieve these goals include the systematic assessment of patients readiness for discharge (by all caregivers, ideally in collaboration with the patients and their families); accurate and complete discharge summaries; explicit postdischarge planning with community caregivers; postdischarge telephonic or in-person nurse contact; and early follow-up appointments. For cardiac surgical procedures, discharge planning is also critical. However, readmissions after these procedures are often associated with specific surgical complications, which provide additional mitigation opportunities. There are a number of reasons why CABG may be an appropriate procedure for which to monitor risk-adjusted readmission rates. Frequency and Financial Impact of CABG Readmissions Among 7 conditions accounting for 30% of 2005 Medicare spending on readmissions within 15 days of discharge, CABG had the highest readmission rate (13.5%) 1 and second highest Medicare payment per readmission ($8136). In the 2008 MEDPAC Report to Congress, 2 the 30-day potentially preventable readmission rate for CABG in 2005 was 18.1%, accounting for $215 million of a total $12 billion dollars (1.8%) of Medicare readmission spending. Many CABG Readmissions Go to Other Hospitals Among published studies, including the current report, half of all CABG readmissions are to hospitals other than where surgery was performed. This is substantially >20% to 40% average for all Medicare readmissions 16 and emphasizes the need to link clinical data with claims sources to ensure complete case ascertainment. Without accurate, all-inclusive feedback reports, many programs will be unaware of their true CABG readmission rates and performance relative to other institutions, thus limiting improvement efforts. Patients living further away will more likely present to their local physician or emergency department. They may be readmitted for problems that could have been treated as an outpatient had the surgeon been aware. This presents an opportunity to educate referring doctors not only about common post-cabg problems, but also the need to contact the surgeon about any early postoperative concerns before admitting these patients locally. Lack of Improvement in CABG Readmission Rates Although CABG mortality rates have steadily declined, CABG readmission rates have not. Among New York CABG patients from 1999, the overall readmission rate was 15.3%; 12.9% of patients were readmitted for reasons that appeared directly related to surgery. 11 In a follow-up study of patients from 2005 to 2007, 21 the comparable rates were even higher (16.5% and 14.3%, respectively). Clinical Impact of CABG Readmissions In a study of New York CABG readmissions, patients who were readmitted had a 4-fold increase in 30-day mortality (2.79% versus 0.76%) in comparison with those who were not. 21 Although this may reflect underlying patient severity, it is also

8 406 Circulation July 29, 2014 Figure 2. Observed vs model-predicted readmission rates. The coefficients fitted with 2008 data (training sample) were applied to 2009 data (validation sample) to calculate the predicted rates. Vertical bars represent 95% exact binomial confidence intervals of observed readmission rates in 10 deciles. likely that the conditions requiring readmission, often surgical complications, were major contributors to increased mortality. Significant Interhospital Variation in CABG Readmission Rates Similar to the national variability in overall Medicare readmission rates, 16 substantial variation has also been observed in reports of cardiac surgery readmissions, ranging from 0% to 26.9%. 21,22 The existence of wide variability in previous studies, the substantial variation observed in the current report, and the very low readmission rates achieved by some programs, 23,24 all suggest opportunities for improvement. Improvement Opportunities Unlike medical readmissions, CABG readmissions are more often the result of delayed occurrence or recognition of procedural complications. With shorter hospital stays, complications will often be first noted after discharge. Figure 3. Distribution of risk-standardized readmission rates. Numerous studies 11,18 22,24 26 have examined the causes and predictors of readmission after cardiac surgery, including CABG. Common reasons for readmission include wound infections; heart failure; arrhythmias; indeterminate chest pain, myocardial infarction, or angina; pneumonia or pleural effusions; stroke; sepsis; gastrointestinal complications; and renal failure. Long length of stay during the initial hospitalization is a consistent predictor of CABG readmission, probably reflecting a more severely ill patient preoperatively or the development of postoperative complications. Conversely, short index hospitalizations are often associated with a lower risk of readmission, because such patients have had an optimal recovery. Specific readmission risk factors include advanced age; female sex; nonwhite race; obesity; anemia; heart, kidney, or liver failure; chronic lung disease; and diabetes mellitus. Although not appropriate for inclusion in profiling risk models because they reflect complications of care, certain postoperative events have been associated with readmission, including sternal infections, reoperation for bleeding or tamponade, inotrope use, administration of blood products, atrial fibrillation, and renal failure This information may be helpful in identifying patients who may benefit from enhanced postdischarge follow-up. Finally, although not traditionally considered appropriate for inclusion in profiling risk models, several studies 11,20 22,26 have identified sociodemographic and hospital factors that are associated with readmission risk including low household income, discharge to a venue other than home, distance from hospital to home, low surgeon volume, and high risk-adjusted hospital mortality rates. These findings suggest that the patient s preoperative clinical status, the quality of their in-hospital operative and perioperative care, certain hospital characteristics, and a variety of socioeconomic and demographic factors may all be associated with CABG readmission risk. Some of these factors may be modifiable, whereas others may serve as triggers for enhanced peridischarge planning and follow-up. Better inpatient care, more complete predischarge resolution of certain problems, more effective care transitions to the community environment, and closer postoperative follow-up may be useful in reducing readmissions. 26,27 Specific clinical interventions might include optimization of diabetes mellitus control; reduction of wound infections (antibiotic timing and selection, meticulous surgical technique); prevention of early graft closure (enhanced use of arterial grafts, consistent use of postoperative aspirin); minimizing pulmonary complications (early ambulation and pulmonary rehabilitation; predischarge thoracentesis or early follow-up chest x-ray and return visit for pleural effusions); more aggressive prophylaxis and treatment of perioperative atrial fibrillation; guidelinedirected use of medications such as β-blockade, statins, aspirin, and angiotensin-converting enzyme inhibitors or angiotensin receptor blockers; and daily postdischarge monitoring of patient weights to assess fluid retention and the need for diuresis. Readmission and Mortality Rates Are Complementary Performance Measures Isolated CABG case volumes (and thus sample size) for most providers have declined substantially over the past decade,

9 Shahian et al CABG Readmission Measure Based on Clinical Data 407 as have average mortality rates (now 1% 2%). These factors make it increasingly difficult to statistically distinguish levels of performance among providers. Furthermore, mortality is only 1 aspect of quality, which led the STS to implement and publicly report a composite CABG performance measure consisting of 11 individual measures of performance in 4 domains Readmission may be another important candidate metric for multidimensional performance assessment. For CABG, as with several medical conditions, 32 risk-adjusted mortality and readmission rates measure overlapping but different aspects of quality. Hannan and colleagues 11 found that some risk factors (eg, preoperative hemodynamic instability) were important CABG mortality predictors, but operative survivors with these characteristics did not have increased readmission risk. Other factors (eg, age, female sex, diabetes mellitus) were predictors of both mortality and readmission. The Pearson correlation coefficients between risk-adjusted CABG mortality and readmission rates were 0.09 and 0.32 in 2 studies by Hannan and colleagues, 11,21 and in the study of Li and colleagues. 22 When we used 2010 STS data among 827 CMS CABG providers to examine the correlation between STS CABG composite performance (risk-adjusted mortality, risk-adjusted major morbidity, use of at least 1 internal mammary artery graft, and use of National Quality Forum endorsed perioperative medications) and risk-adjusted readmission rates from our current model, the Spearman rank correlation was (ie, higher composite performance correlated very weakly with lower readmission rates). Programs with exceptionally low mortality and morbidity rates may not necessarily have concomitantly low readmission rates. Mortality (and the STS CABG composite) measure somewhat different aspects of care than readmission rates, and the measurement of both provides a more comprehensive CABG performance assessment. Design Features of a CABG Readmission Model Because CABG is often done in conjunction with other cardiac procedures, it is important to identify a homogeneous population of isolated CABG patients rather than aggregating these with patients who have undergone more complex combined procedures. We used all-cause CABG readmissions because attempting to distinguish those that are unrelated to the index hospitalization is problematic and could lead to gaming. Furthermore, empirical data suggest that the vast majority are related. 11,21 Unlike some previous readmission models, we regarded all CABG readmissions as unplanned. Thirty-day CABG readmissions may occasionally be planned, but these are very uncommon and would be difficult to audit; we did not attempt to identify and exclude these from the outcome. In a departure from previous CMS models, we attributed all readmissions to the hospital where the procedure was performed, regardless of subsequent interhospital transfers. For medical readmission models, readmissions are typically attributed to the last hospital in a transfer chain, presuming that the quality of that hospital s discharge planning most significantly impacts readmission risk. For CABG, postsurgical transfers are almost always necessitated by complications of the original procedure (eg, perioperative myocardial infarction leading to heart failure and the need for ventricular assist), and subsequent readmissions are ultimately the end result of earlier events at the index hospital. In comparison with outcomes such as mortality, regression models for readmission have lower discrimination. C-indices between 0.75 and 0.85 are not uncommon for mortality models. 13 However, C-indices in this range are rarely if ever found in readmission models, including those published by Hospital Compare and the State of New York, which often have C-indices of 0.60 to ,11,21,33 (with a C-index of 0.50 being equivalent to a coin toss). One plausible explanation is that, by convention, risk models used for provider profiling do not contain race, ethnicity, or socioeconomic status variables, because their inclusion might obscure disparities in care for vulnerable populations. However, such variables, though generally considered inappropriate for profiling models, are likely more important risk factors for readmission than mortality. They are surrogates for the patient s home and community environment and the support mechanisms available postdischarge, all of which are important determinants of readmission risk. Support for this hypothesis comes from models designed not for profiling but rather to proactively identify patients at risk for readmission so that targeted preventative interventions may be initiated. When racial, ethnic, and socioeconomic variables are included in such models, they significantly impact estimated readmission risk and improve model discrimination It must be acknowledged, however, that some hospitals serving disadvantaged populations have achieved readmission rates that overlap those in higher socioeconomic status areas, 39 suggesting that vulnerable populations may still experience reasonable readmission rates with appropriate peridischarge support. Despite the modest discrimination of most readmission risk models including ours, risk adjustment of readmission rates used for provider profiling is still important. Although Hannan and colleagues 21 found that unadjusted and risk-adjusted CABG readmission rates were similar, they still recommend using risk adjustment, because some hospitals (eg, tertiary centers, or those serving vulnerable populations) may have substantially more patients at high risk for readmission. Risk adjustment is essential for face validity, particularly within the provider community whose results are being assessed, and granular clinical data are the best source for these risk factors. Limitations Not all Medicare CABG admissions could be linked to an STS record, and some Medicare admissions were, by design, excluded from analysis (eg, unknown discharge date). Similarly, some Medicare hospitals were excluded because of extremely low CABG volumes, possibly indicating they were not actually CABG providers. All observational studies have inherent limitations, including bias, which cannot be fully eliminated. Conclusion A 30-day all-cause CABG readmission measure suitable for public reporting has been developed by using robust clinical data from the STS Database linked to Medicare readmission data. The C-index, calibration, and reliability of this measure

10 408 Circulation July 29, 2014 meet or exceed the performance of most previous readmission models including those adopted by CMS. 5 7,11,21,33 This measure complements mortality as a measure of CABG performance, and it has the potential to improve care and reduce costs. Sources of Funding This project was funded by DHHS/CMS contract HHSM STS01C. Disclosures Drs Peterson, O Brien, and He are employees of Duke Clinical Research Institute, the STS data warehouse and analytic center. Drs Drye and Suter work under contract with the Centers for Medicare and Medicaid Services (CMS) to develop and maintain performance measures. Dr Shewan is employed by STS. Dr Han is employed by CMS. The other authors report no conflicts. References 1. Report to the Congress: Promoting Greater Efficiency in Medicare Accessed December 31, A path to bundled payment around a hospitalization. In Report to the Congress: Reforming the Delivery System. Washington, DC: Medicare Payment Advisory Commission; Accessed August 6, Epstein AM. Revisiting readmissions changing the incentives for shared accountability. N Engl J Med. 2009;360: Krumholz HM, Brindis RG, Brush JE, Cohen DJ, Epstein AJ, Furie K, Howard G, Peterson ED, Rathore SS, Smith SC, Jr., Spertus JA, Wang Y, Normand SL. Standards for statistical models used for public reporting of health outcomes: an American Heart Association Scientific Statement from the Quality of Care and Outcomes Research Interdisciplinary Writing Group: cosponsored by the Council on Epidemiology and Prevention and the Stroke Council. Endorsed by the American College of Cardiology Foundation. Circulation.2006;113: Keenan PS, Normand SL, Lin Z, Drye EE, Bhat KR, Ross JS, Schuur JD, Stauffer BD, Bernheim SM, Epstein AJ, Wang Y, Herrin J, Chen J, Federer JJ, Mattera JA, Wang Y, Krumholz HM. An administrative claims measure suitable for profiling hospital performance on the basis of 30-day all-cause readmission rates among patients with heart failure. Circ Cardiovasc Qual Outcomes. 2008;1: Krumholz HM, Lin Z, Drye EE, Desai MM, Han LF, Rapp MT, Mattera JA, Normand SL. An administrative claims measure suitable for profiling hospital performance based on 30-day all-cause readmission rates among patients with acute myocardial infarction. Circ Cardiovasc Qual Outcomes. 2011;4: Lindenauer PK, Normand SL, Drye EE, Lin Z, Goodrich K, Desai MM, Bratzler DW, O Donnell WJ, Metersky ML, Krumholz HM. Development, validation, and results of a measure of 30-day readmission following hospitalization for pneumonia. J Hosp Med. 2011;6: Shahian DM, Silverstein T, Lovett AF, Wolf RE, Normand SL. Comparison of clinical and administrative data sources for hospital coronary artery bypass graft surgery report cards. Circulation. 2007;115: Mack MJ, Herbert M, Prince S, Dewey TM, Magee MJ, Edgerton JR. Does reporting of coronary artery bypass grafting from administrative databases accurately reflect actual clinical outcomes? J Thorac Cardiovasc Surg. 2005;129: Jacobs JP, Edwards FH, Shahian DM, Haan CK, Puskas JD, Morales DL, Gammie JS, Sanchez JA, Brennan JM, O Brien SM, Dokholyan RS, Hammill BG, Curtis LH, Peterson ED, Badhwar V, George KM, Mayer JE Jr, Chitwood WR Jr, Murray GF, Grover FL. Successful linking of the Society of Thoracic Surgeons adult cardiac surgery database to Centers for Medicare and Medicaid Services Medicare data. Ann Thorac Surg. 2010;90: ; discussion Hannan EL, Racz MJ, Walford G, Ryan TJ, Isom OW, Bennett E, Jones RH. Predictors of readmission for complications of coronary artery bypass graft surgery. JAMA. 2003;290: STS version 2.61 procedure ID table. STS sites/default/files/documents/pdf/trainingmanuals/adult2.61/vi-a_- Procedure_Identification_Table.pdf. Accessed February 28, Shahian DM, O Brien SM, Filardo G, Ferraris VA, Haan CK, Rich JB, Normand SL, Delong ER, Shewan CM, Dokholyan RS, Peterson ED, Edwards FH, Anderson RP. The Society of Thoracic Surgeons 2008 cardiac surgery risk models, I: coronary artery bypass grafting surgery. Ann Thorac Surg 2009;88:S Adams JL. The Reliability of Provider Profiling: A Tutorial TR653.pdf. Accessed July 31, Bodenheimer T. Coordinating care a perilous journey through the health care system. N Engl J Med. 2008;358: Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med. 2009;360: Kripalani S, LeFevre F, Phillips CO, Williams MV, Basaviah P, Baker DW. Deficits in communication and information transfer between hospital-based and primary care physicians: implications for patient safety and continuity of care. JAMA. 2007;297: Stewart RD, Campos CT, Jennings B, Lollis SS, Levitsky S, Lahey SJ. Predictors of 30-day hospital readmission after coronary artery bypass. Ann Thorac Surg. 2000;70: Lahey SJ, Campos CT, Jennings B, Pawlow P, Stokes T, Levitsky S. Hospital readmission after cardiac surgery. Does fast track cardiac surgery result in cost saving or cost shifting? Circulation. 1998;98(19 suppl):ii35 II D Agostino RS, Jacobson J, Clarkson M, Svensson LG, Williamson C, Shahian DM. Readmission after cardiac operations: prevalence, patterns, and predisposing factors. J Thorac Cardiovasc Surg. 1999;118: Hannan EL, Zhong Y, Lahey SJ, Culliford AT, Gold JP, Smith CR, Higgins RS, Jordan D, Wechsler A. 30-day readmissions after coronary artery bypass graft surgery in New York State. JACC Cardiovasc Interv. 2011;4: Li Z, Amstrong EJ, Parker JP, Danielsen B, Romano PS. Hospital variation in readmission after coronary artery bypass surgery in California. Circ Cardiovasc Qual Outcomes 2012;5: Berry SA, Doll MC, McKinley KE, Casale AS, Bothe A Jr. ProvenCare: quality improvement model for designing highly reliable care in cardiac surgery. Qual Saf Health Care. 2009;18: Sun X, Zhang L, Lowery R, Petro KR, Hill PC, Haile E, Garcia JM, Bafi AS, Boyce SW, Corso PJ. Early readmission of low-risk patients after coronary surgery. Heart Surg Forum. 2008;11:E327 E Beggs VL, Birkemeyer NJ, Nugent WC, Dacey LJ, O Connor GT. Factors related to rehospitalization within thirty days of discharge after coronary artery bypass grafting. Best Pract Benchmarking Healthc. 1996;1: Ferraris VA, Ferraris SP, Harmon RC, Evans BD. Risk factors for early hospital readmission after cardiac operations. J Thorac Cardiovasc Surg. 2001;122: Rumsfeld JS, Allen LA. Reducing readmission rates: does coronary artery bypass graft surgery provide clarity? JACC Cardiovasc Interv. 2011;4: Shahian DM, Edwards FH, Ferraris VA, Haan CK, Rich JB, Normand SL, Delong ER, O Brien SM, Shewan CM, Dokholyan RS, Peterson ED. Quality measurement in adult cardiac surgery, I: conceptual framework and measure selection. Ann Thorac Surg 2007;83:S3 S O Brien SM, Shahian DM, DeLong ER, Normand SL, Edwards FH, Ferraris VA, Haan CK, Rich JB, Shewan CM, Dokholyan RS, Anderson RP, Peterson ED. Quality measurement in adult cardiac surgery, II: statistical considerations in composite measure scoring and provider rating. Ann Thorac Surg. 2007;83(4 suppl):s13 S Shahian DM, Edwards FH, Jacobs JP, Prager RL, Normand SL, Shewan CM, O Brien SM, Peterson ED, Grover FL. Public reporting of cardiac surgery performance, I: history, rationale, consequences. Ann Thorac Surg. 2011;92(3 suppl):s Shahian DM, Edwards FH, Jacobs JP, Prager RL, Normand SL, Shewan CM, O Brien SM, Peterson ED, Grover FL. Public reporting of cardiac surgery performance, II, implementation. Ann Thorac Surg. 2011;92(3 suppl):s12 S Krumholz HM, Lin Z, Keenan PS, Chen J, Ross JS, Drye EE, Bernheim SM, Wang Y, Bradley EH, Han LF, Normand SL. Relationship between hospital readmission and mortality rates for patients hospitalized with acute myocardial infarction, heart failure, or pneumonia. JAMA. 2013;309: Kansagara D, Englander H, Salanitro A, Kagen D, Theobald C, Freeman M, Kripalani S. Risk prediction models for hospital readmission. JAMA 2011;306: Amarasingham R, Moore BJ, Tabak YP, Drazner MH, Clark CA, Zhang S, Reed WG, Swanson TS, Ma Y, Halm EA. An automated model to identify

11 Shahian et al CABG Readmission Measure Based on Clinical Data 409 heart failure patients at risk for 30-day readmission or death using electronic medical record data. Med Care. 2010;48: Rathore SS, Masoudi FA, Wang Y, Curtis JP, Foody JM, Havranek EP, Krumholz HM. Socioeconomic status, treatment, and outcomes among elderly patients hospitalized with heart failure: findings from the National Heart Failure Project. Am Heart J. 2006;152: Coleman EA, Min SJ, Chomiak A, Kramer AM. Posthospital care transitions: patterns, complications, and risk identification. Health Serv Res. 2004;39: Weissman JS, Stern RS, Epstein AM. The impact of patient socioeconomic status and other social factors on readmission: a prospective study in four Massachusetts hospitals. Inquiry. 1994;31: Joynt KE, Orav EJ, Jha AK. Thirty-day readmission rates for Medicare beneficiaries by race and site of care. JAMA. 2011;305: Medicare Hospital Quality Chartbook Quality-Initiatives-Patient-Assessment-Instruments/HospitalQualityInits/ Downloads/MedicareHospitalQualityChartbook2012.pdf. Accessed July 30, Clinical Perspective Reducing readmissions is a major goal of healthcare reform, and reimbursement penalties are imposed for higher-thanexpected readmission rates. We developed a prediction model for 30-day all-cause readmission following coronary artery bypass grafting by using nationally representative clinical data ( ) from the Society of Thoracic Surgeons National Database linked to Medicare claims records. Risk-standardized readmission rates ranged from 12.6% to 23.6% (median, 16.8%) among 846 US hospitals with 30 eligible cases and 90% of eligible Centers for Medicare and Medicaid Services records linked to the Society of Thoracic Surgeons database. Fifty-two hospitals (6.1%) had readmission rates statistically better or worse than expected. Important readmission predictors present at the time of admission included dialysis, severe chronic lung disease, elevated creatinine, insulin-dependent diabetes mellitus, obesity in women, female sex, immunosuppression, preoperative atrial fibrillation, older age, recent myocardial infarction, and low body surface area in men. This readmission risk model provides several useful functionalities. First, it allows cardiac surgery programs to proactively identify, at the time of admission, patients who may be at higher risk of readmission postdischarge. These patients may be targeted to receive enhanced assessment of their suitability for discharge, more rigorous screening by hospital case management and care coordinators of their home environment and support systems, earlier postdischarge outpatient follow-up, and, in some instances, telephonic or at-home visits by members of the cardiac team. Second, these models provide a rigorously developed and validated tool for determining whether programs have average, above average, or below average risk-adjusted readmission performance. This information can be used, as needed, for public reporting or reimbursement.