USING ENSEMBLE CLUSTERING TO IDENTIFY PHENOTYPES OF DIABETES PATIENTS FOR EVALUATING DISEASE PROGRESSION

Abstract

Diabetes Mellitus (DM) is a chronic health condition that affects multiple organs and is associated with significant morbidity and mortality. The management of diabetes requires periodic pathology investigations and physician examinations to manage the disease’s progression. The ability to predict the temporal progression of the disease for a patient can significantly impact therapeutic choices and improve outcomes. This thesis presents investigations in stratifying diabetes patients based on their pathology test results in terms of temporally salient patient clusters. We investigated machine learning-based clustering methods, especially ensemble clustering and applied them to time-series data of pathology tests and their results to generate patient clusters. Using the three identified clusters, we generated patient phenotypes comprising clinical characteristics that are then used to develop classification-based prediction models to predict the disease’s temporal progression and suggest the potential pathology tests. The ability to predict and understand disease progression will lead to novel personalized medicine for managing patients with diabetes.