Abstract
Mathematical modeling is a powerful quantitative tool to investigate the pathogenesis of type 2 diabetes (T2D). Most modeling work on the progression of T2D has been formulated by modifying the pioneering model of Topp et al, which established the paradigm of glucotoxicity as the main driver of pathogenesis. However, certain parameter values in the Topp model deviate from physiological data, leading to predictions that deviate from clinical scenarios. Moreover, the simple structure of the model limits its explanatory capacity for clinical data. Leveraging a four-dimensional longitudinal dataset from Southwest Native Americans who progressed from normal glucose tolerance to T2D, we developed a series of models, starting with a minimally modified version of the Topp model and iteratively incorporating additional model elements to account for new biological mechanisms until optimal data fit was achieved. The notable variability of the individual trajectories was overcome by the non-linear mixed-effect modeling approach. Despite the absence of a discernible common trend among the individual trajectories of each variable, the model effectively captured the diverse glucose-insulin dynamics of individuals progressing to T2D. The reliability of the model was reinforced by its successful cross-validation against a subset of individuals progressing only to prediabetes. The systematic model selection process aided in navigating the trade-off between model complexity and practicability, culminating in a robust framework to address controversial questions in the diabetes field in future research.