Misclassification of Hospital Volume With Surveillance,Epidemiology,and End Results—Medicare Data

Abstract

Surveillance, Epidemiology, and End Results (SEER)— Medicare data are frequently used for studying relationships between volume and outcomes after cancer surgery; however, because patients often cross SEER boundaries for treatment, SEER-Medicare data may misclassify hospital volume. Thus, we measured the agreement of hospital volume as determined by SEER-Medicare and 100% national Medicare data and determined the extent to which misclassification alters the apparent relationship between volume and operative mortality. This is a retrospective cohort study of SEER-Medicare patients undergoing a major cancer surgery for colon, lung, bladder, and esophageal cancers between 1994 and 1999. Hospital procedure volumes were assessed with both SEER-Medicare and 100% national Medicare data and sorted into terciles. Logistic regression models were fit using generalized estimating equations to assess associations between mortality and volume, as determined from each data source. Compared with 100% Medicare data, SEER-Medicare data misclassified 13% (colectomy) to 36% (esophagectomy) of patients; however, fewer than 3% of patients were misclassified by more than 1 volume stratum. For cystectomy, the apparent association between volume and mortality was relatively weak and not statistically significant based on SEER-Medicare data (adjusted odds ratio, low vs high volume 1.41, 95% confidence interval, 0.89-2.23), but stronger and significant when volume was obtained from 100% Medicare data (odds ratio, 1.82; 95% confidence interval, 1.17 to 2.84). For the other 3 procedures, apparent volume/outcome relationships were similar when volume was assessed from the 2 data sources. Hospital volumes are frequently misclassified with SEER-Medicare data. Such misclassification generally biases volume/outcome associations toward the null, but this effect seems to be small for many procedures. Investigators should be cognizant of this bias and exercise caution when interpreting these relationships when using SEER-Medicare data alone.