A quantitative size stability metrics for long-acting suspensions and its prediction with machine learning

Defining suspension stability can be extremely complex, but beyond critical during the formulation development of nano- and microsuspensions intended for long-acting injectables. As of now, the current practice is based on the trial-and-error approach to determine suitable stabilizers and corresponding stabilization concentrations to allow size reduction due to proper wetting while minimizing particle agglomeration or crystal growth by Ostwald ripening during storage, making the definition of suspension stability subjective in the existing literature. The present study proposed a quantitative measurement of the suspension stability (i.e., suspension stability index (SSI)) defined by two coefficients of variation from the suspension population median, i.e., D₅₀CV, and distribution, i.e., SpanCV, over 28 days of storage under accelerated conditions. Various predictive models for SSI were constructed with a defined threshold for the D₅₀CV and SpanCV where the support vector classification (SVC) model was found to hold the best performance based on a 5-fold cross-validated accuracy of 0.82 and F1 score of 0.82 while also having a good transferability. Critical factors that impacted the SSI included the stabilizer type and its corresponding concentration while the drug compound was found to interact with multiple parameters with the exception of the bead size. Overall, the presented quantitative metric of suspension stability based on the SSI could potentially contribute to a more comparable investigation of suspension stability during formulation development by using the classification model from the presented dataset as a screening tool.