LAUSR

Ji et al. propose methodology to correct the bias inherent in the estimates of a treatment/response relationship conditional on an ancillary biomarker BN (adopting authors’ notation), and in the estimates of the ancillary biomarker/response relationship conditional on treatment assignment, in a randomized clinical trial in which outcome response randomization (OAR) is stratified on a different biomarker B0. In this OAR setting with a binary outcome, the authors propose an estimator for the log odds ratio of response for treatment S versus treatment E conditional on biomarker BN status and an estimator for the log odds of response for biomarker BN positive versus BN negative conditional on treatment assignment. Such analyses may be of interest in retrospective analyses of a clinical trial investigating a new biomarker, perhaps with the goal of replacing B0 with BN .