National Cancer Institute Home at the National Institutes of Health |
Please wait while this form is being loaded....
The Applied Research Program Web site is no longer maintained. ARP's former staff have moved to the new Healthcare Delivery Research Program, the Behavioral Research Program, or the Epidemiology & Genetics Research Program, and the content from this Web site is being moved to one of those sites as appropriate. Please update your links and bookmarks!

Publication Abstract

Authors: Panageas KS, Schrag D, Riedel E, Bach PB, Begg CB

Title: The effect of clustering of outcomes on the association of procedure volume and surgical outcomes.

Journal: Ann Intern Med 139(8):658-65

Date: 2003 Oct 21

Abstract: BACKGROUND: A large body of literature documents associations between the volume of cases a hospital or surgeon treats and clinical outcomes. Most of these studies have used conventional statistical methods that do not recognize the fact that hospitals or surgeons with similar volumes may have very different outcomes because of systematic differences in processes of care, a phenomenon that exaggerates the true statistical significance of the effect of volume on outcome. OBJECTIVE: To describe methods to assess the degree of this "clustering" of outcomes and to explore the impact of available statistical techniques that correct for clustering. DESIGN: Reanalysis of 3 previously published volume-outcome studies. SETTING: Medicare beneficiaries 65 years of age or older undergoing surgery for colon, prostate, or rectal cancer in the population defined by the Surveillance, Epidemiology, and End Results cancer registries during 1992 to 1996. PATIENTS: 3 data sets were analyzed to assess the impact of surgeon volume on outcomes: 1) 24 166 colectomies performed by 2682 surgeons, 2) 10 737 prostatectomies performed by 999 surgeons, and 3) 2603 rectal resections performed by 1141 surgeons. MEASUREMENTS: Volume-outcome trends were analyzed by a conventional method (logistic regression) and corrected for clustering. Two widely used statistical methods for analyzing clustered data, a random-effects model and generalized estimating equations, were used and compared, and the degree of clustering was presented graphically. RESULTS: Substantial clustering was observed in the analyses involving morbidity end points. The 2 statistical techniques produced noticeably different results in some analyses. CONCLUSIONS: The presence of clustering represents variations in outcomes among providers with similar volumes. Thus, in volume-outcome studies, the degree of clustering of outcomes should be characterized because it may provide insight into variations in quality of care.

Last Modified: 03 Sep 2013