The Least Bad Option: A Simulation Approach to Minimizing Bias When Accounting for Treatment Switching in RCTs

• Activity Number: 233 - Statistical Considerations for Adjusting Overall Survival in Randomized Trials with Treatment Switching
• Type: Topic Contributed
• Date/Time: Tuesday, August 9, 2022 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Daniel Leibovitz* and Alessandro Previtali and Revathi Ananthakrishnan and James Lymp
• Companies: Bristol Myers Squibb and Bristol Myers Squibb and Bristol Myers Squibb and Bristol Myers Squibb
• Keywords: treatment switching; treatment effect; IPCW; RPSFTM; 2-Stage Estimator; simulator

Abstract

Treatment switching (TS) refers to the phenomenon in which participants in a randomized controlled trial switch from their randomly assigned treatment to an alternative. Typically, TS only occurs from control to experimental arm and does not occur randomly. In this case, the standard causal intention-to-treat analysis gives conservatively biased estimates of treatment effects (TEs). Several methods have been developed for the estimation of TEs under conditions of TS (IPCW, RPSFTM and 2-Stage Estimator). When the assumptions of these methods are met, they each provide unbiased estimates of TEs in the presence of TS. However, the assumptions of each of these methods are untestable and when not met, their estimates are again biased. Users of these methods must make informed decisions about which methods’ assumptions are most likely to be violated, by what magnitude, and how much bias each violation is likely to add to the TE estimate. In the current research, we present (i) a new theoretical schema to guide statisticians in choosing an appropriate adjustment method, and (ii) a simulator which quantifies method-specific bias on survival data generated from given parameters.