Prediction of tablet weight variability from bulk flow properties by sparse modeling

Abstract

Die filling is a critical step in the pharmaceutical tableting process as it determines the uniformity in tablet weight, which affects the drug content uniformity and appearance of tablets. With the progress in powder flow characterization, a wide variety of bulk flow properties can be obtained, and several studies have been conducted to predict tablet weight variability (TWV) from bulk flow properties by applying multivariate analysis; however, there is still room for improvement in the selection of properties for predictive model construction. In this study, least absolute shrinkage and selection operator regression, a type of sparse modeling, was applied to select the critical flow properties from various ones, and TWV was predicted. To obtain blends with a wide range of properties, 27 powder blends were prepared by changing the active pharmaceutical ingredients (APIs), API loading in the formulation, and the grade of the lubricant. Bulk flow properties were evaluated, and a good prediction model was obtained by selecting five out of 14 bulk flow properties. The constructed model also predicted TWV well in the validation dataset. The applied approach is useful for constructing a simple model to better understand the phenomenon (die filling in this case), and the constructed model can be expanded further by including blends manufactured by dry or wet granulation.