Analysis of selection bias in a case-control study of renal adenocarcinoma.

Crude estimation of the selection probability ratio (SPR), described previously, was extended to stratified and multivariate estimation and used to assess selection bias in a case-control study of renal adenocarcinoma. It was shown that the directly pooled estimate of the SPR, using the same weights as the directly pooled estimate of the exposure odds ratio (OR) from the case-control study (assuming the OR and SPR are common to all strata and data are abundant), can be multiplied with the OR to yield an adjusted OR that is free from selection bias. Medical records of 548 interviewed cases were compared with 640 noninterviewed cases, and interviews of 640 controls were compared with mailed questionnaires from 272 (60%) of the noninterviewed controls. Age-sex-adjusted point estimates of SPRs ranged from 0.65 to 1.4. Multivariate estimates from binomial regression ranged from 0.34 to 2.0. Higher socioeconomic status and history of renal stones were predictors of participation by both cases and controls. Obesity in women, hypertension, and nonsmoking were predictors in cases only. Heart disease was associated with control participation and case nonparticipation. This study cast doubt on the OR for obesity in women and hypertension in the case-control risk analysis.