Andrew Ryan, Doctoral Candidate

1. The Effects of the Hospital Quality Incentive Demonstration on Medicare Patient Mortality and Cost Andrew Ryan, Doctoral Candidate

2. Acknowledgements • Support • Agency for Healthcare Research and Quality (AHRQ) • Jewish Healthcare Foundation • Dissertation Committee • Stan Wallack • Chris Tompkins • Deborah Garnick • Kit Baum

3. The Hospital Quality Incentive Demonstration (HQID) • Collaboration between Premier Inc. and CMS • Implemented in 4th quarter of 2003, continues today • Pays a 2% bonus on Medicare reimbursement rates to hospitals performing in the top decile of a composite quality measure • Pays 1% bonus for hospitals performing in the second decile (80th -90th percentile) of composite measure • Incentivized conditions • Acute myocardial infarction (AMI) • Heart failure • Community-acquired pneumonia • Coronary-artery bypass graft (CABG) • Hip and knee replacement

4. Evidence of Effectiveness of HQID • Two of the three published evaluations conclude that the HQID improved process quality (Grossbart 2006; Lindauer et al. 2007) • One article concluded that the HQID did not significantly improve process or mortality quality (Glickman et al. 2007) • Effect of HQID on mortality examined only by Glickman (and only for AMI) • No research has studied effect of HQID on Medicare cost

5. Why might HQID impact Medicare cost? • Medicare has administered prices • Cost can change if: • Admissions change • If higher quality care results from HQID, readmissions may decrease • Outlier categorizations change • Quality improvement may increase resource use and costs • Hospitals may attempt to recoup costs from quality improvement through increased outlier categorizations

6. Data and Methods

7. Challenges estimating the effects of the HQID • Selection effect: • Because the HQID is voluntary, hospital participation in the HQID is a signal of a hospital’s interest in improving its clinical quality or reputation • Of the 421 hospitals asked to participate in the HQID, 266 (63%) chose to participate (Lindauer et al. 2007) • Hospitals’ eligibility to participate in the HQID was based on their subscription to Premier’s Perspective database, a database used for benchmarking and quality improvement activities • Confounding from other time-invariant of time-varying factors (e.g. hospital size, hospital technology) • Econometric approach employs 3 estimators of the effect of the HQID in the presence of unobserved selection and confounding from observables

8. Data • Panel of hospitals from 2000-2006 • Because the HQID began in the fourth quarter of 2003, the study period spans the six year period from the fourth quarter of 2000 through the third quarter of 2006 • Medicare fee-for-service inpatient claims -used to identify the primary diagnoses for which beneficiaries were admitted, secondary diagnoses and type of admission for risk adjustment, cost data, and discharge status to exclude transfer patients • Medicare beneficiary denominator files - used to add additional risk adjusters and to determine mortality • Medicare Provider of Service files - used to identify hospital structural characteristics • Only short-term, acute care hospitals are included in the analysis

9. Data continued • Dependent variables • Risk adjusted (RA) 30-day mortality • RA 60-day cost • RA Outlier categorization (day or cost) • Incentivized conditions examined in study: AMI, heart failure, pneumonia and CABG • Hip and knee replacement excluded because of low mortality • Risk adjusted outcomes: Hospital-level observed / expected • Expected outcome estimated from patient-level logit and regression models • Age, gender, race • Elixhauser comorbidities (Elixhauser et al. 1998) • Type of admission (emergency, urgent, elective) • Season of admission

10. Approach #1: Fixed Effects among all hospitals • RA Outcome jkt = b1Yeart + b2h j + b3Yeart * Z jt + δ HQID jt + e jkt • Where j indexes to hospitals, k indexes to clinical condition, and t indexes to year. • h jis a vector of hospital fixed effects • Z is a vector of hospital characteristics (number of beds, teaching status, coronary-care unit, inpatient surgery, intensive care unit, open-heart surgery facility, and condition-specific Herfindahl index (a measure of market concentration)) • Characteristics in Z have limited within-hospital variation so must be interacted with year in fixed effects model • Assumes effect of unobserved selection is constant over observation period • Effects of hospital characteristics are allowed to varyover observation period • Models estimated among all acute-care hospitals

11. Approach #2: Fixed Effects among hospitals eligible for HQID • Equation 1 is estimated in sample of hospitals eligible for HQID hospitals • HQID Eligibility is based on subscribing to Perspective database -likely indicative of an interest in improving quality • Approach may better account for time-varying confounds (including selection effects)

12. Approach #3 Difference-in-difference-in-differences (DDD) among all hospitals (2) std(RA Outcome HQID Condition jkt ) - std( RA Outcome Reference jkt )= b1Yeart + b2hj + b3Yeart * Z jt + δ HQID jt + e jkt • Where std is the z transformation • Dependent variable is the difference between the z-transformed RA mortality for a given condition incentivized under the HQID (e.g. AMI) and a condition not incentivized under the HQID • Three differences: • Between conditions incentivized in HQID and conditions not incentivized in HQID • Between hospitals in the HQID and those not in the HQID • Before and after the HQID began • Assumes that a hospital’s unobserved interest to improve quality applies to multiple domains of its clinical care while the effects of the HQID are limited to the incentivized conditions under this program

13. Approach #3 continued • Z transformation normalizes RA mortality across conditions by standardizing RA rates to mean 0 and standard deviation 1 allowing for an interpretation of differences in mortality rates that is not obscured by baseline differences • Difference in log mortality also evaluated • Reference condition candidates are chosen from among the AHRQ inpatient mortality indicators • Criteria for selection of reference conditions • Criterion #1, reference condition will have a reasonably large number of hospitals that treat patients with the condition in a sufficient volume • Criterion #2, reference condition will have a positive time-varying correlation with the HQID condition. • Criterion #3, reference condition will not be subject to spillover effects of the HQID

14. Reference conditions for DDD

15. Standard error specification • Multiple observations for hospitals over time give rise to group-level heteroskedasticity • Cluster-robust standard errors are estimated • Hospital-level mortality rates vary in their precision as a result of the number of cases treated • Analytic weights (Gould, 1994), based on number of cases treated by a hospital in a given year, are employed

16. Results

17. Effect of HQID on 30-day mortality Fixed effects all hospitals Fixed effects HQID eligible DDD - Gastrointestinal hemorrhage DDD - stroke DDD - AAA DDD - Craniotomy = 95% confidence intervals of point estimates

18. Effect of HQID on 60-day cost Fixed effects all hospitals Fixed effects HQID eligible DDD - Gastrointestinal hemorrhage DDD - stroke DDD - AAA DDD - Craniotomy = 95% confidence intervals of point estimates

19. Effect of HQID on outlier categorization Fixed effects all hospitals Fixed effects HQID eligible DDD - Gastrointestinal hemorrhage DDD - stroke DDD - AAA DDD - Craniotomy = 95% confidence intervals of point estimates

20. Unique number of hospitals in models Note: included are acute care hospitals with atleast two years of data from 2000-2006 Note: if two hospitals merged in the observation period, the analysis treats them as three unique hospitals

21. Conclusion • Using 3 different estimators of effect of HQID: • No evidence of effect of HQID on mortality • No evidence of effect of HQID on cost • Limited evidence of a positive relationship between HQID and outlier categorization • Study is consistent with finding of Glickman et al. that HQID has not reduced mortality for AMI • Value-based purchasing modeled on the HQID may not increase value for Medicare