Methods for Developing a Process Design Space Using Retrospective Data

Abstract

Prospectively planned designs of experiments (DoEs) offer a valuable approach to preventing collinearity issues that can result in statistical confusion, leading to misinterpretation and reducing the predictability of statistical models. However, it is also possible to develop models using historical data, provided that certain guidelines are followed to enhance and ensure proper statistical modeling. This article presents a methodology for constructing a design space using process data, while avoiding the common pitfalls associated with retrospective data analysis. For this study, data from a real wet granulation process were collected to pragmatically illustrate all the concepts and methods developed in this article.