Correct Statistical Inference from Simulated Clinical Trials: Meta-modeling and Statistical Assurance for Virtual Bioequivalence Assessments Using PBPK Models

We show that correct statistical inference from simulated virtual clinical trials requires a Bayesian framework. Simulating frequentist test procedures degrades the prior information contained in the model. We demonstrate the application of a Bayesian workflow to PBPK-simulated virtual bioequivalence trials. The workflow extends and clarifies the notion of statistical assurance proposed by O'Hagan. The heavy calculations involved are made possible by the development and use of a meta-model able to replace accurately the complex PBPK model which would othewise be required.