Statistical approaches supporting QbD milestones via bioprocess digital twins

Abstract

Quality by Design is a life-cycle paradigm used by regulators to encourage pharmaceutical companies to consider product quality from the earliest development stages. The objective is to identify and document relationships between critical process parameters and productquality attributes via risk assessments and experimental results before validating for commercial manufacturing. This methodology has steadily gained traction over the last decade and applied statistical tools are increasingly leveraged to reach these goals quantitatively.Nonetheless, significant gaps remain between the methodological intent and the current state-of-the-art practices. For example, during the identification of critical process elements,latent variables have largely been overlooked due to over-reliance on process knowledge and the absence of relevant extraction and multivariate methods. Subsequent risk assessments and data-driven models are siloed in the individual process steps (unit operations) and are not linked to the patient-relevant outcome: drug substance specifications. Lastly, there is an absence of data feedback loops between the above procedures and the manufacturing data in the commercial life-cycle.This thesis addresses the above gaps via improvements and applications of an integratedprocess model; a framework centered on concatenating unit operation models and propagatingerror via Monte Carlo simulations. To realize this potential, novel procedures were first designed to uncover latent bioprocess variables via extraction and multivariate analysis.Once in place, an innovative Monte Carlo-based application was developed that establishes intermediate acceptance criteria for quality attributes via parameter sensitivity analysis. A further simulation procedure was created which, when combined with linearization techniques,enables the determination of parameter proven acceptable ranges and links these quantitatively to risk assessment severity rankings. Lastly, the integrated process model was substantially improved and inserted architecturally into manufacturing data feedback loops, enabling the model to react in real time to process conditions. The totality of these innovations depicts a major industry objective: a bioprocess digital twin.Leveraging the developments in this thesis, the proposed integrated process model now quantitatively links process parameters and quality attributes to patient-relevant outcomes.Moreover, it does so with a technology that can iteratively adapt to new manufacturing data,ensuring that it accompanies the process throughout its life-cycle, and thereby establishes an engine for a digital twin. Thus, with holistic process quality as a central goal, the industry will be able to better fulfill both the intention and the potential of the Quality by Designparadigm.