Applications of practical process modeling based on statistics, multivariate analysis, and simulation in pharmaceutical development
To assure the drug products’ quality, process understanding and control based on a process model is a common and important process development activity in the pharmaceutical industry. The most common of many approaches to building a process model is the Design of Experiments (DoE) and the multiple linear regression analysis. However, this conventional process modeling approach has challenges with respect to 1) the huge resources required to conduct the DoE and 2) the low prediction accuracy of the process model when multicollinearity is involved. The reason for the first challenge is because most industrial processes show high scalability and hence typically DoE runs for design space setting need to be performed at commercial scale. The second challenge is also common in the industrial process where multicollinearity exists, which cannot be described properly using multiple linear regression. In case studies 1 and 2 a computer-aided process simulation utilizing Discrete Element Method and statistics was demonstrated to address the first challenge in a blending process to reduce the process development cost by substituting actual experiments with in-silico experiments. Quantitative prediction of the blend uniformity was successfully demonstrated, opening up the possibility to reduce the number of experiments in process development. To address the second challenge alternative statistical process modeling was applied for a process involving multicollinearity in case studies 3 and 4, the tableting and coating processes. Process models were built based on partial least squares regression, which can cope with interdependent parameters by using latent variables. The response surfaces were in good agreement with the verification study results, suggesting a high prediction accuracy for a process involving multicollinearity. The process modeling approaches demonstrated in this thesis will facilitate understanding and control of the process performance with reasonable cost.