Accounting for Covariates Observed Post Randomization for Discrete and Continuous Repeated Measures Data

Adjusting for covariates observed post randomization is a difficult issue in randomized clinical trials. One approach is to enter these covariates as time-dependent covariates in a general linear model. However, this approach fails to account for interactions between the primary end point and confounding variable as they evolve over time. In this paper, we adopt the view that, since the confounder is observed following randomization, it should be treated as an outcome and analysed accordingly. We consider the repeated measures design where both the primary end point and the confounding measure are observed repeatedly over patient follow-up. A generalized estimating equation model is applied to allow each set of repeated measures to be modelled in terms of important explanatory variables. Seemingly unrelated regression combines the two models into an overall framework for analysis. Inference is then performed by imposing restrictions on the confounder regression parameters to reflect the behaviour profile of interest. Estimation and identification procedures are described and the methodology is illustrated with an example.