Assessing the effect of treatment regimes on longitudinal outcome data: application to revamp study of depression

Depression studies frequently adopt two-stage designs to examine the efficacy of augmenting pharmacotherapy with psychotherapy. Initially subjects receive one of the several treatments; if  hey respond, they continue the same treatment; however, if they fail to respond, they move to the next stage  nd are randomized to other treatment options. Outcomes such as 24-item Hamilton Rating Scale of Depression (HRSD24) scores are then collected repeatedly to monitor the progress of the subject. The goal is to assess the effect of treatment regimes (consisting of initial treatment, initial response and the second stage treatment combinations) on HRSD24 profile. Statistical inference for assessing treatment regimes using a summary outcome measure such as mean response has been well-studied in the literature. Statistical methods for assessing the effect of treatment strategies on repeated measures data focused mainly on estimating equations. In this article, we propose two methods based on mixed models and multiple imputations to assess the effect of treatment regimes on the longitudinal HRSD24 scores. Methods are compared through simulation studies and through an application to a depression study. The simulation studies showed that the estimates from both methods are approximately unbiased, and provide good coverage rates for 95% confidence intervals.