Presentation is loading. Please wait.

Presentation is loading. Please wait.

Surgery volume and operative mortality: A re-examination using fixed-effects regression Amresh Hanchate, PhD Section of General Internal Medicine Boston.

Similar presentations


Presentation on theme: "Surgery volume and operative mortality: A re-examination using fixed-effects regression Amresh Hanchate, PhD Section of General Internal Medicine Boston."— Presentation transcript:

1 Surgery volume and operative mortality: A re-examination using fixed-effects regression Amresh Hanchate, PhD Section of General Internal Medicine Boston University School of Medicine Funding: AHRQ AcademyHealth Annual Research Meetings 2006 June 26, 2006

2 2 Acknowledgements Arlene Ash, Ph.D (Boston University) John Birkmeyer, MD (University of Michigan) Therese Stukel, PhD (Institute for Clinical Evaluative Sciences, Toronto) AHRQ Grant

3 3 Provider Volume & Operative Mortality: Background Most studies indicate that hospitals and surgeons with higher volumes have significantly lower operative mortality True for a variety of complex, high-risk surgeries Volume – proxy for volume process effect

4 4 Patient profiles across hospitals and surgeons Studies are based on observational data -- administrative or clinical chart databases We may not observe significant –process of care details –patient characteristics (severity) Patient profiles may vary systematically across hospitals and surgeons (Dranove, Kessler, et al, JPE, 2003) Which patient goes to which provider? –physician referral (technology, expertise) –patient choice (experience, recommendations, report cards) –provider choice (report cards) To estimate volume process effect, need to adjust for patient profiles at provider level –Fixed effects: allows cluster effects to be correlated with covariates –Random effects: assumes cluster effects to be uncorrelated with covariates

5 5 Objective To compare the estimates of the association of hospital and surgeon volumes with operative mortality for CABG using (a) fixed effects (FE) regression (b) random effects (RE) regression.

6 6 Data: Birkmeyer, Stukel et al (NEJM, 2003) N Mean volume / year Patients220,592 Hospitals Surgeons2,77285 All surgeries at one hospital (%)56%77 All surgeries at 2+ hospitals (%)44%97 Source: All CABGs during 1998 & 1999 from Medicare Fee for Service Inpatient Files Clustering: Patients clustered among surgeons and hospitals Surgeons not nested in hospitals

7 7 Variables Outcome: Operative mortality (1/0) within 30-day or before discharge Patient characteristics –Age, gender and race –Charlson score –Elective / Non-elective admission –Area (zip code) income indicator Surgeon –Volume per year (at all hospitals) Hospital –Volume per year –Teaching status –Ownership (not-for-profit, government, for-profit) –Urban / Non-urban

8 8 Regression Methodology Linear probability model Fixed Effects Regression –within cluster –Outcome and covariates are transformed by differencing out the cluster mean –To estimate surgeon volume effect Hospital as fixed cluster – surgeons within same hospital compared Within Hospital Cohort – Patients in hospitals with at least two surgeons (1% patients excluded) –To estimate hospital volume effect Surgeon as fixed cluster Within Surgeon Cohort – Patients whose surgeons operated at two or more hospitals (49% patients, 44% surgeons and 21% hospitals excluded) Random Effects Regression – 3-tiered hierarchical

9 9 Comparison of the two cohorts Operative mortality & patient characteristics alike Relatively fewer high volume hospital patients in Within Surgeon Cohort

10 10 Estimates of Volume Effects # Excess Operative Deaths per 1,000 CABG surgeries Surgeon volume effects: FE and RE are similar Hospital volume effect: FE > RE

11 11 Adjusted Operative Mortality (Operative Deaths Per 1,000 CABG surgeries)

12 12 Volume Effect Decomposition (Operative Deaths Per 1,000 CABG surgeries) Unobserved – Operative mortality effect of unobserved factors at hospital level Low volume hospitals have protective factors not being captured in our data High volume hospitals may have sicker patient profile than low volume hospitals

13 13 Conclusions The FE approach decomposes volume effect into Observed and Unobserved component effects. The RE estimate may be viewed as the net of Observed and Unobserved effects from FE regression. The sizable Unobserved effect indicates that for CABG patients in high volume hospitals are different from those in low volume hospitals – they may be sicker. Question: To what extent can the Observed component (FE) be seen as the volume process effect? Limitations –Poor measure of illness severity (Charlson Scores) –Other unobserved phenomena


Download ppt "Surgery volume and operative mortality: A re-examination using fixed-effects regression Amresh Hanchate, PhD Section of General Internal Medicine Boston."

Similar presentations


Ads by Google