Exploratory Discriminant Analysis in Phase 2 Clinical Trials to investigate Treatment Effect Heterogeneity
*Lev Sverdlov, Merck Pharmaceutical Inc. 

Keywords: Discriminant Analysis, Treatment Effect Heterogeneity, Minimum Projected Therapeutic Concentration, Proof of Concept Study, Major Depression

Phase 2 Proof of Concept (PoC) studies are imperative to drug development. Using standard endpoint parameters PoC studies can result in two possible scenarios: - a false positive signal, may lead to failure a research compound in Phase 3; or - a false negative signal, may lead to burying a research compound with undiscovered potential; In order to decrease false positive or negative signals from PoC studies, certain exploratory multivariate statistical analysis are used. For example, Discriminant Analysis (DA) can be used to predict dichotomous group memberships based on linear combination of interval variables. This presentation examines the methodology and results of the exploratory DA in the PoC study of a new antidepressant. The trial data was collected from a randomized, placebo-controlled, double-blind pilot clinical trial involving 51 evaluable subjects diagnosed with major depression who underwent either one or two 5-day treatment cycles. Due to the relatively small sample size, all subjects available for analysis were selected for the training (calibration) data set. Key variables in the DA were the Psychometric scores with respect to eleven major time points. The performance of the DA was evaluated by estimating error rates (probability of misclassifications). Based on the Minimum Projected Therapeutic Concentration (retrospective PK analysis), the DA was successful in separating treatment subgroups: 82.4% to 86.7% of subjects had correct classification in any two of three subgroups and 80% of subjects had correct classification in all three subgroups. The results of the exploratory DA played an important role in investigating treatment effect heterogeneity and in aligning the business decisions.