OBJECTIVES: The purpose of this study is to develop a diabetes outcomes model using a multi-ethnic, real-world-data cohort of newly diagnosed Type II diabetics in the US that can predict outcomes for a U.S. multi-ethnic type 2 diabetes population.

METHODS: We identified over 150,000 newly diagnosed diabetes patients between 2005-2016 with up to 13-year follow-up using merged EMR and claims from the Kaiser Permanente Northern California (KPNC) Diabetes Registry. The model integrates separate, but interdependent risk equations to predict events for each of the micro and macro-vascular events, hypoglycemia, dementia, depression, and death, predictive models for eight biomarker levels and fifteen-possible glucose –lowering treatment combinations. We used a provider preference as an instrument for the glucose lowering treatment models. Model accounted for static demographic factors (e.g., race), neighborhood deprivation, and dynamic factors, such as age, duration of diabetes, fifteen-possible glucose –lowering treatment combinations, biomarker levels, and history of diabetes-related events. Moreover, the models explicitly allow for a legacy effect (average A1c in the first year after diagnosis) for all outcomes.

RESULTS: Data were randomly split into 50%, 25%, and 25% to perform estimation, out-of-sample calibration, and validation respectively. Model predictions in the validation sample closely aligned with the observed longitudinal trajectory of biomarkers and outcomes. Moreover, we examine the model performance within by age, race/ethnicity, and sex and found excellent predictive performance within subgroups.

CONCLUSIONS: The DOMUS Model is able to simulate event histories and biomarker trajectories that closely match observed outcomes in a real-world, multi-ethnic population of newly diagnosed diabetics. DOMUS has the potential to carry out many complex analyses that examine the effectiveness and cost-effectiveness of new medications, value of diabetes prevention, and serve as decision support tools in health care settings.