An approach to analyzing case-control screening data with an application to digital rectal exam and metastatic prostate cancer

Prostate cancer (PC) is a prominent killer; men over 40 should receive annual digital rectal exam (DRE) to prevent metastatic prostate cancer (MPC). For rare outcomes like MPC, case-control studies are common. Previous studies counted DREs as dichotomous and compared cases/controls using conditional logistic regression and the case-control odds ratio. Observation time for DRE ends at PC since afterwards DREs are diagnostic. Cases and controls are matched on time; but when cases are screen-detected there is bias counting more case screenings. Excluding tests shortly before diagnosis only reverses the bias. We begin instead with a life table for first DRE, stratified by case/control and propensity score (representing confounders), from which the Mantel-Haenszel odds ratio is calculated for case/control versus DRE on tables cross-classifying these, stratified by interval and propensity score. We find a mildly protective effect (OR=0.861, 95% CI=0.611,1.215), and find the approach superior to those used in previous studies.