Abstract P2005: Applying a comprehensive cardiometabolic risk prediction model to real-world primary care patient data

Background: Comprehensive risk prediction models utilized at the point of care have potential to improve screening rates, facilitate prevention referrals, and reduce disparities. However, feasibility of deploying comprehensive models in the clinical workflow are largely unknown. Objective: To assess: (1) feasibility of applying a validated risk prediction model – the cardiometabolic disease staging system (CMDS) - that incorporates metabolic, vascular, and social determinants of health (SDoH) factors to electronic health record (EHR) data from primary care patients and (2) cardiometabolic disease risk for 10-year onset of type 2 diabetes (T2D) and major adverse cardiovascular events (MACE). Method: Data to calculate CMDS [clinical: BMI, glucose, blood pressure, HDL, triglycerides, smoking status; SDoH: income, education, insurance status, stress and neighborhood level social vulnerability] among patients without documented T2D or MACE diagnosis [myocardial infarction, stroke, cardiovascular death] were extracted. CMDS was applied using 3 algorithms: (1) clinical and (2) clinical + SDoH for predicting T2D; and (3) clinical for MACE. Feasibility of model application was determined by calculating the percent of patients with a score out of all eligible primary care patients. We estimated mean scores overall by outcome; quartiled scores to determine risk strata; and assessed mean MACE by diabetes risk quartile. Results: A total 10,398 patients [mean age: 63.7 years (SD 10.6), 63% female, 26% non-Hispanic Black] were identified. Full data were available to apply the clinical model to 73% of patients and the CMDS + SDoH to <1% for T2D outcome; 67% for clinical MACE model. Among patients that had a clinical CMDS calculated there was a mean 17.4% risk of developing T2D; and 10.6% of MACE. Quartiles of risk had natural breaks at <6%, 6-10%, 11-20% and >21% for T2D and <5%, 5-8%, 9-14%, >14% for MACE. Missing SDoH data [99.4% income; 95.4% education; 0% insurance; 94.7% stress; 33.5% neighborhood level social vulnerability; 0.6% marital status] was the limiting factor for CMDS + SDoH model application. Mean MACE risk was 11.0% among those at highest risk for T2D. Conclusion: While the clinical CMDS algorithms could be applied to EHR data, findings highlight the lack of available data to apply risk prediction models with SDoH measures. Future work should examine how to incorporate models into primary care workflow, including efforts to systematically screen social risks.