Planning a method for covariate adjustment in individually-randomised trials: a practical guide

Background: It has long been advised to account for baseline covariates in the analysis of confirmatory randomised trials, with the main statistical justifications being that this increases power and, when a randomisation scheme balanced covariates, permits a valid estimate of experimental error. There are various methods available to account for covariates. Methods: We consider how, at the point of writing a statistical analysis plan, to choose between three broad approaches: direct adjustment, standardisation and inverse-probability-of-treatment weighting (IPTW), which are in our view the most promising methods. Using the GetTested trial, a randomised trial designed to assess the effectiveness of an electonic STI (sexually transmitted infection) testing and results service, we illustrate how a method might be chosen in advance and show some of the anticipated issues in action. Results: The choice of approach is not straightforward, particularly with models for binary outcome measures, where we focus most of our attention. We compare the properties of the three broad approaches in terms of the quantity they target (estimand), how a method performs under model misspecification, convergence issues, handling designed balance, precision of estimators, estimation of standard errors, and finally clarify some issues around handling of missing data. Conclusions: We conclude that no single approach is always best and explain why the choice will depend on the trial context but encourage trialists to consider the three methods more routinely.