Covariate Adjustment in Oncology Clinical Trials: Past, Present and Future

Baseline covariate adjustment in randomized controlled trials (RCTs) in oncology with primary endpoint time to event has a long history. This paper reviews the past, present and future of this practice. We distinguish between stratification by baseline covariates which is used to gain precision in estimates and use of log linear models, such as proportional hazards regression, with or without stratification. The latter is done either to increase precision in estimates or to find which covariates may be considered prognostic factors. Knowledge of prognostic factors will allow us to define characteristics of patients who are likely to benefit from treatment and are a vital part of the new initiative for precision or personalized medicine. We provide insight into the effect of stratification on estimation of standard errors and size and power of statistical hypothesis tests and the use of minimization randomization as another method to create covariate balance between treatment groups. We close with current challenges In covariate adjustment in RCTs that use restricted mean survival time, win statistics, new estimands, allow treatment crossover and tissue-agnostic clinical trials.