Modelling and simulation of the Positive and Negative Syndrome Scale (PANSS) time course and dropout hazard in placebo arms of schizophrenia clinical trials.

BACKGROUND AND OBJECTIVES The likelihood of detecting a therapeutic signal of an effective drug for schizophrenia is impeded by a high placebo effect and by high dropout of patients. Several unsuccessful trials of schizophrenia, at least partly due to highly variable placebo effects, have indicated the necessity for a robust methodology to evaluate such a placebo effect and reasons for dropout. Hence, the objectives of this analysis were to (i) develop a longitudinal placebo model that accounts for dropouts and predictors of the placebo effect, using the Positive and Negative Syndrome Scale (PANSS) score; (ii) compare the performance of empirical and semi-mechanistic placebo models; and (iii) compare different time-to-event (TTE) dropout modelling approaches used to account for dropouts. METHODS The PANSS scores from 1436 individual patients were used to develop and validate a placebo model. This pooled dataset included 16 trials (conducted between 1989 and 2009), with different study durations, in both acute and chronic schizophrenic patients. A nonlinear mixed-effects modelling approach was employed, using NONMEM VII software. RESULTS Among the different tested placebo models, the Weibull model and the indirect response model adequately described the PANSS data. Covariate analysis showed that the disease condition, study duration, study year, geographic region where the trial was conducted, and route of administration were important predictors for the placebo effect. All three parametric TTE dropout models, namely the exponential, Weibull and Gompertz models, described the probability of patients dropping out from a clinical trial equally well. The study duration and trial phase were found to be predictors for high dropout rates. Results of joint modelling of the placebo effect and dropouts indicated that the probability of patients dropping out is associated with an observed high PANSS score. The indirect response model was found to be a slightly better model than the Weibull placebo model to describe the time course of the PANSS score. CONCLUSIONS Our modelling approach was shown to adequately simulate the longitudinal PANSS data and the dropout trends after placebo treatment. Data analyses suggest that the Weibull and indirect response models are more robust than other placebo models to describe the nonlinear trends in the PANSS score. The developed placebo models, accounting for dropouts and predictors of the placebo effect, could be a useful tool in the evaluation of new trial designs and for better quantification of antipsychotic drug effects.