Covariate Adjustment in Analyzing Randomized Clinical Trials: Approaches, Software, and Application

In randomized clinical trials (RCTs), randomization ensures a balance of observed and unobserved baseline risk factors between treatment groups, thus limiting the potential impact of confounders for treatment effect evaluation. In general, unadjusted analysis without taking covariates into account allows valid testing and inference on treatment effects in RCTs. However, covariate adjustment in RCTs can help address bias due to the random balance of baseline covariates of RCTs with small sample sizes and, more importantly, improve the power of statistical tests and the precision of estimation on treatment effect. In this book chapter, we review four commonly used covariate adjustment methods in RCTs, including regression, G-computation, inverse probability treatment weighting (IPTW), and augmented inverse probability treatment weighting (AIPTW). The implementation of these methods for binary and continuous outcomes is discussed, and guiding steps are included for reference. For binary outcomes, outcome models and estimating equations for three different estimands, including odds ratios (OR), risk ratios (RR), and risk differences (RD) are provided. The issue of non-collapsibility associated with nonlinear estimands, such as OR, is also discussed. We summarize the relative performance of these four methods and recommend guidance for choosing the appropriate covariate adjustment methods for individual cases. We further review applicable R packages for covariate adjustment and propose a three-stage workflow to help with the use and reporting of covariate adjustment analysis in practice. In the end, we illustrate the use of these four methods through a motivating example of studying the effects of adjuvant chemotherapy treatments in a randomized phase III trial for early-stage non-small-cell lung cancer.