Leveraging complementary functions of hydroxypropyl methylcellulose and croscarmellose sodium to develop gastroretentive tablets via predictive regression-based modeling.

We report gastroretentive (GR) tablets optimized via a data-driven regression-based predictive modeling strategy to determine the optimal hydroxypropyl methylcellulose (HPMC)-to-croscarmellose sodium (CCS) ratio. A botanical extract (BE, Layla® soft extract), composed of 12 medicinal herbs, was selected as the model drug and was first solidified into granules using Neusilin via fluid bed granulation. To optimize the formulation of GR tablets, a four-parameter sigmoid-based regression model was constructed in MATLAB and compared with a quadratic response surface model (RSM) and an artificial neural network (ANN, 2-8-1 structure). Although the ANN exhibited the best fit to the training data (highest R2), the regression model provided more accurate predictions under extrapolated conditions, outperforming both ANN and RSM. Cross-validation, residual diagnostics, and bootstrap-based prediction interval analysis collectively confirmed the statistical robustness of the regression model. The optimized composition of BE, HPMC, and CCS was determined as 600:160:140 (w/w/w), followed by compression into tablet form. The GR tablet successfully incorporated BE without evidence of physicochemical interactions among BE, HPMC, and CCS. Upon contact with simulated gastric acid, the GR tablet floated within 10 s, remained buoyant for over 18 h in an expanded state, and sustained the release of BE. In vivo X-ray imaging in beagle dogs confirmed that the GR tablets remained in the stomach for up to 6 h. Altogether, this study presents a formulation strategy that systematically optimizes the complementary functions of HPMC and CCS through predictive modeling, leading to prolonged gastric retention and sustained drug release.