Use of the Reverse Propensity Score to Detect and Correct for Selection Bias in Clinical Trials

• Activity Number: 196
• Type: Topic Contributed
• Date/Time: Tuesday, August 13, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section*
• Author(s): Vance Berger*+ and Costas Cristophi
• Affiliation(s): National Cancer Institute and George Washington University
• Keywords: Covariate Adjustment ; Reverse Propensity Score ; Valid Between-Group Inference ; Allocation Concealment ; Randomized Clinical Trials ; Selection Bias

Abstract

While selection bias is generally held to exist in the purview of non-randomized studies, even randomized clinical trials (RCTs) are susceptible. Specifically, restricted randomization allows prediction of future allocations based on past ones. This lack of allocation concealment, in conjunction with investigator enrollment discretion, can lead to differential recruitment across treatment groups, with systematically better responders recruited to the preferred group. We define the reverse propensity score (RPS) as the probability, conditional on all previous allocations and the allocation procedure (restrictions on the randomization), that a given patient is to receive a given treatment. If the RPS is common for all patients, because of either unrestricted randomization or perfect masking of previous allocations, then there is no basis for enrolling patients differentially across treatment groups, and selection bias cannot occur. However, restricted randomization is generally used and masking can rarely, if ever, be assured. In fact, selection bias has been documented to occur in RCTs. We demonstrate how the RPS allows for both detection of and correction for selection bias.