In order to determine the efficacy of pharmacological interventions in clinical trials, the placebo effect must be taken into consideration. It is important to distinguish placebo effects from the effects of the actual treatment being tested. Most studies of the placebo effect have looked at responders who have been treated with a placebo and have tried to identify factors that can predict such response. This is obviously a very limited approach because data from subjects treated with the active drug is ignored. The research proposed here is to develop statistical methodology for identifying and differentiating placebo responses from true drug responses in the treatment of mental illnesses such as depression where placebo response rates tend to be high. The initial research will focus on clinical trial data studying treatments for depression. The proposed statistical methodology will combine functional data analysis with cluster analysis and mixture models. Outcome profiles for individual subjects will be estimated using longitudinal data from clinical trials. Appropriate basis functions will be determined to estimate the profile trajectories. The profiles will then be described by a small number of estimated basis function coefficients. Depending on the distribution of estimated coefficients, representative profiles will be estimated using principal point/cluster analysis or a finite mixture analysis. Data from the placebo arm of the studies will also be used to estimate and validate the representative profiles. The representative profiles will then be used to classify future subjects as placebo responders, true drug responders or a combination of a drug-placebo responder. Further work will refine the methodology by incorporating random effects models and addressing problems such as missing data. The derived models will be cross-classified with clinician determined responder/non-responder status for validation. In addition, data from discontinuation studies are available and will be used to further validate the models for placebo response. The second phase of the study will apply the methodology to characterize classes of drugs (ssri, tricyclics and maoi) with respect to their placebo response profiles. The final phase of the research will apply the methodology of determining the placebo effect from the true drug effect in other mental illnesses such as anxiety, obsessive compulsive disorders, panic and post traumatic stress syndrome.
