A latent variable approach in simultaneous modeling of longitudinal and dropout data in schizophrenia trials.

Dropouts impact clinical trial outcome analyses. Ignoring missing data is not an acceptable option when planning, conducting or interpreting the analysis of a clinical trial. Treatment related efficacy and safety data observed in the trial may not always be sufficient in explaining the dropouts' mechanism. Nevertheless, these dropout data may carry important treatment-related information and present as an outcome by itself. Traditional analyses involve the use of the time-to-event approach assuming that the dropouts' hazard is solely related to the efficacy or safety profiles observed in a study. A latent variable approach was developed to generalize this approach and to implement a more flexible dropout hazard function in a schizophrenia trial. This unobserved latent variable was used to jointly model the longitudinal efficacy data and dropout profiles across treatments. The analysis provides a framework to model informative dropouts simultaneously with primary efficacy outcomes and make intelligent decisions in drug development.