(584b) Uncertainty Estimation in Drug Substance Process Development with Mechanistic Models

Understanding and defining the design space with statistical and mechanistic methods is the cornerstone for quality by design (QbD) in pharmaceutical process development. Using probabilistic (Bayesian) models to estimate the risk of failure within the design space has been an emerging approach for risk-based control of the design space. In this work, we implemented an adaptive mechanistic model in a full Bayesian framework for design space mapping. Compared to traditional statistical models, this approach allows the simultaneous estimation of multiple CQAs and their associated uncertainty estimations. This framework can be easily adapted to different chemical transformations through a customized interface with Reaction Lab^TM from Scale-Up Systems, lowering the barrier for wider use by other scientists. This model was applied to recent portfolio projects. The data supporting the model was collected from a time-resolved DOE, which allows comparison of the mechanistic Bayesian model to a statistical Bayesian model. This approach allows a risk-based evaluation and confirmation of the existing proven acceptable ranges (PARs) to assure smooth tech transfer of the process to a new site.