Modelling placebo response in depression trials using a longitudinal model with informative dropout

Dropouts are common events in longitudinal studies in depression. Ignoring missing information may lead to biased and inconsistent assessment of study results.A non-linear model was recently developed to describe the time-course of HAMD-17 clinical score in the placebo arms of antidepressant clinical trials. In this paper we complemented this model by introducing an informative dropout component to jointly estimate HAMD-17 time-course and dropout mechanism.The aims of this work were to: (a) characterise typical placebo response in depression trials in presence of dropouts, (b) explore which dropout mechanism better describe the time-varying probability of a subject to dropout from the trial, and (c) define a framework for the development of clinical trial simulation in depression.A meta-analytic approach was used on placebo data collected in 6 clinical trials including 695 subjects suffering from Major Depressive Disorders.Alternative hypotheses for “missingness” were evaluated using different hazard models. The “Missing Not At Random” performed statistically (p < 0.01) better than “Missing At Random”, that in turn performed better (p < 0.01) than “Missing Completely At Random” model. This finding provided new insights on the validity of the analyses currently used in many longitudinal clinical trials to support the registration of a new medicinal product.