Computational Predictability of Microsponge Properties Using Different Multivariate Models

The capabilities of principal component regression (PCR) and multiple linear regression (MLR) were evaluated to decipher and predict the impact of formulation and process parameters on the modeled metronidazole benzoate (MB)-ethyl cellulose (EC) microsponge (MBECM) properties. MBECM were prepared by a quasi-emulsion solvent diffusion method. A minimum experimentation was designed using Box-Behnken approach with one center point after initial screening experiments. Data was modeled by principal component analysis (PCA), PCR, and MLR. Two distinct groupings of developed MBECM was observed in initial qualitative PCA as a function of their respective formulation and processing parameters. Group A formulations with low dichloromethane, high PVA, and low stirring speed exhibited larger particle size, lower entrapment efficiency (EE), and lower actual drug content (ADC) than Group B formulations. Optimized quantitative PCR and MLR models demonstrated a linear dependence of particle size and quadratic dependence of EE and ADC on the studied formulation and process parameters. Interestingly, MLR models showed relatively better predictability of the selected MBECM formulation properties when compared with PCR. MBECM were amorphous in nature and spherical shaped. Carbopol® 940 NF based hydrogel of selected MBECM formulation exhibited a prolonged MB release than the commercial MB gel (Metrogyl®), showing no signs of necrosis in the goat mucosa. Thus, a properly designed minimum experimentation coupled with multivariate modeling generated a knowledge-rich target space, which enabled to understand and predict the performance of developed MBECM within a prescribed design space.