The currently recommended strategies for managing diabetes rely primarily on broad, population-level data and average treatment effects observed in clinical trials. There is a critical need for an approach to personalizing type 2 diabetes (T2D) intervention, thereby refining treatment efficacy. In this thesis, we utilize unsupervised clustering techniques to investigate the various factors contributing to the prevalence and progression of T2D in individuals with different lifestyles and characteristics. We demonstrated our approach on two datasets from two projects.

In the first project, we developed a data-driven framework to find physical activity-related phenotypes in an underserved T2D population. Moreover, we examine the association between physical activity measures and participants’ diabetes progression in the exclusive subgroups. In the second case, we classified pediatric patients into five distinct diabetes subtypes using K-Prototypes cluster analysis. Additionally, our findings provide new insights on how the treatment strategies and risk stratifications can differ even at the time of diabetes diagnosis based on a more precise characterization of pediatric T2D. In conclusion, the proposed data-driven approaches could bring us one step closer to precision therapies and individualized recommendations to become a routine part of diabetes management.