Design and analysis considerations for first treatment escalation in clinical trials

Protocol-mandated, outcome-driven treatment adaptations are common in many types of clinical trials, and a prominent feature of so-called treat-to-target trials. Successive treatment escalation (dosage increase or addition of a drug) if disease activity targets are not reached is a typical design element in these studies. Focusing on the first treatment escalation step, here we address ways of estimating the effect of this outcome-based intervention as well some issues pertaining to the design of such treatment changes in randomized trials. Estimating escalation effects requires to disentangle them from concurrent effects, including persisting effects of the randomized treatment and regression to the mean. A regression-based method and a likelihood ratio test were adapted and assessed in simulations and in data from a recent treat-to-target study in rheumatoid arthritis. In simulations, the procedures were satisfactory in terms of bias and sensitivity with some advantage for the likelihood ratio test. They were able to identify evidence for the escalation effect in the examined study. In summary, both analysis methods are useful, but are sensitive to key assumptions and rely on compliance to the protocol as well as frequent and complete assessments. Furthermore, we examined different treatment escalation designs, including escalation at multiple time points (early escalation). If a longitudinal model for the disease activity is available, we describe how early escalation strategies can decrease the overall disease burden. We provide recommendations for the design of treatment escalation procedures in typical settings.