OBJECTIVES: Phase III trials typically exclude patients with certain baseline characteristics, such as older age or co-morbidities, and thereby hamper learning of new drugs’ effectiveness in real-life. A simulation study was conducted to support implementation of new inclusion criteria for Phase 3 trials in schizophrenia without increasing sample size nor compromising detection of the new drug effect. METHODS: A simulation study was performed examining the impact of re-introducing through stratifying by each of the following excluded patients population: age > 65 years, duration of illness < 3 years, patients with previous suicide attempts, patients with history of alcohol or substance abuse, and patients treated in private practices.  Patients with these characteristics were multiplied in a synthetic trial population until their real-life proportion in schizophrenia was reached. The simulation used data subsets from the 10,281-patient observational SOHO cohort study.  A “base case RCT” was created by applying typical Phase 3 exclusion criteria. A series of “synthetic RCTs” were defined by replacing patients with SOHO patients that were initially excluded. The real-life drug effect was predicted from each synthetic RCT through regression models and compared with the real-life effect in SOHO. RESULTS: Perhaps surprisingly, effects of all 3 investigated drugs were found to be larger in real-life than in the base case RCT. Synthetic RCTs were created by replacing patients of the base RCT with patients with a given baseline characteristic. Prediction of real-life effects improved with increasing replacement in terms of mean squared prediction errors and coverage of confidence interval. However, the impact of introducing these "real-life" populations was not equal among factors. For instance, introducing older patients minimally improved prediction of real-life effects, while allowing inclusion of just 5% of patients with past suicide attempts (who make up 25% of the real-life schizophrenia population) significantly improved effectiveness predictions.