Data-Driven Quality Assessment of Polyoxyl 40 Hydrogenated Castor Oil: Utilizing CART® Modeling to Define Critical Thresholds for Chemical Stability

Abstract

Polyoxyl 40 Hydrogenated Castor Oil (PHCO) is a widely utilized non-ionic surfactant and solubilizer in the pharmaceutical industry. Ensuring the batch-to-batch consistency of its chemical quality attributes, particularly the Acid Value (AcidV), is critical for the stability and shelf-life of finished dosage forms. This study investigates the application of Classification and Regression Tree (CART^®) modeling as a predictive tool for quality assessment in an industrial manufacturing setting. Using a dataset of 48 industrial batches sourced from an Asian manufacturing facility and analyzed in strict accordance with United States Pharmacopeia (USP) standards, seven quality predictors were evaluated: temperature of turbidity, congealing temperature, hydroxyl value, iodine value (IodV), saponification value, water content (WC), and residue on ignition. The statistical analysis, conducted via Minitab with 10-fold cross-validation, produced an optimal 2-node regression tree to minimize the risk of overfitting. Results identified IodV as the most significant predictor of acidity, achieving 100% relative importance, followed by WC (71.9%) and hydroxyl value (52.6%). The model revealed a definitive non-linear threshold at IodV ≤ 0.45; batches meeting this criterion demonstrated a mean AcidV of 0.256, while those exceeding this limit showed a significantly higher mean AcidV of 0.968. The model attained a training R-squared of 42.68% and a test R-squared of 34.85%. These findings suggest that saturation levels and moisture content are the primary drivers of acidity, likely due to the hydrolysis of ester bonds within the polyoxyl chain. This study demonstrates that CART modeling provides robust, actionable decision rules that enhance Quality by Design (QbD) frameworks.

Bangladesh J. Sci. Ind. Res. 61(2), 55-62, 2026