Hydrophilic matrix tablets with prolonged release based on hypromellose (HPMC) enable the oral delivery of the active pharmaceutical ingredient (API) with a lower dosing frequency and better control of API plasma concentrations compared to immediate-release tablets. The main functionality-related characteristics (FRC) of HPMC include particle size (along with other physical properties of particles), the degree of hydroxypropoxy (HP) and methoxy (MeO) substitution, and molecular weight (expressed as the viscosity of aqueous HPMC solutions at a specified concentration). The FRC of HPMC influence the release of the API, even within the same type of HPMC. One potentially useful approach to compensate for the impact of batch-to-batch variation in FRC is to adjust the proportions of incorporated fillers based on the properties of the specific HPMC batch used. Within the framework of a doctoral dissertation, we investigated the influence of fillers on the release of carvedilol (a model API) from hydrophilic matrix tablets based on HPMC K15M, prepared by direct compression. Among water-soluble fillers, the smallest release variability and negligible 'burst effect' (uncontrolled rapid release of the API at the beginning) were achieved with polyethylene glycol (PEG) 8000 and polyethylene oxide (PEO), while among water-insoluble fillers, this was achieved with microcrystalline cellulose (MCC) and ethylcellulose (EC). Analysis of mathematical release models did not reveal a universally applicable model for application to all different experimental release profiles of carvedilol obtained using various fillers. In contrast, the LOESS local regression method proved to be more universally applicable for assessing the proportion of released API at different times, the 'burst effect,' and lag time (delay in onset of API release). Using a design of experiments (DoE) and multivariate analysis (MVA) with multiple linear regression (MLR), we confirmed the influence of HPMC FRC and the proportions of selected fillers in the tablet (lactose and MCC) on carvedilol release, as well as their interactions, which proved to be highly significant both in terms of their impact on the average release of carvedilol and the variability of release. We compared MLR with various other predictive modeling approaches (partial least squares regression (PLSR), support vector regression (SVR), ridge and lasso regression, neural networks, decision trees, etc.), where MLR emerged as the most robust.
|