Development of Polyvinylpyrrolidone-Based Spray-Dried Solid Dispersions Using Response Surface Model and Ensemble Artificial Neural Network

Abstract

A model for spray drying processes was developed using polyvinylpyrrolidone (PVP)-K29/32 as a placebo formulation to predict quality attributes (process yield, outlet temperature, and particle size) for binary solid dispersions (SDs). The experiments were designed to achieve a better understanding of the spray drying process. The obtained powders were analyzed by modulated differential scanning calorimetry, thermogravimetric analysis, X-ray diffraction, polarized light microscopy, and particle size analysis. On the basis of the experimental data, a response surface model and an ensemble artificial neural network were developed. Both models showed significant correlation between experimental and predicted data for all quality attributes. In addition, a Pearson correlation analysis, response surface curves, Kohonen's self-organizing maps, and contribution plots were used to evaluate the effect of individual process parameters on quality attributes. The predictive abilities of both models were compared using separate validation datasets. These datasets contained binary SDs of four model drugs with PVP based on root mean square error and mean absolute error for each quality attribute. The results indicate that both models show reliable predictivity for all quality attributes. The present methodology provides a useful tool for designing a spray drying process, which will help formulation scientists save time, drug usage, and resources in the development of spray-dried SDs.