Machine Learning-Based Prediction of Tablet Weight Uniformity in High Drug Load Continuous Wet Granulation

Background/Objectives: Tablet weight uniformity is a critical quality attribute of high drug load tablets because weight relative standard deviation (RSD) is closely related to dose variation. This study applied machine learning to predict tablet weight variability from granule attributes after continuous wet granulation and milling. Methods: A high drug load acetaminophen formulation without filler, containing one of three glidant grades, was granulated using a ConsiGmaTM-1 continuous twin-screw wet granulation system. The dried granules were milled using two impeller geometries at three speeds, giving 18 formulation–milling conditions, and compressed under fixed conditions. Tablet weight RSD was modeled using adaptive elastic net regression with linear and quadratic terms and evaluated by nested leave-one-out cross-validation, permutation testing, and bootstrap analysis. Results: Glidant addition improved powder feeding, and milling speed was the main factor affecting granule properties. Tablet weight RSD ranged from 0.45% to 2.09% under identical compression conditions. The granule-attribute model predicted tablet weight RSD better than the process-variable model (Q2 = 0.764 vs. 0.480; permutation p = 0.002). Basic flowability energy (BFE) was the most consistently selected attribute and showed a U-shaped relationship with tablet weight RSD. Conclusions: Granule attributes measured before compression may serve as practical indicators of tablet weight variability in high drug load acetaminophen formulations without filler.