PROJECT SUMMARY/ABSTRACTExisting heart disease and stroke prediction models (e.g. Framingham) tend to overestimate risk forHispanics/Latinxs (H/L)s. This inaccuracy has significant economic and public health impacts associated withinaccurate surveillance intervention targeting and medical management. Model inaccuracies likely stem frompervasive underrepresentation of H/Ls in model development and validation efforts. Consequently traditionalrisk factors for cardiovascular disease (CVD) may be specific to the populations upon whom they were derivedand not generalizable to H/Ls. In addition there may be unique disease determinants for H/Ls that remainuntested or unincorporated leading to error in prediction. Importantly resilience factors such as culturally-moderated social capital may be critical to understanding risk in this population. Addressing these gaps willlead to better understanding of CVD risk with corresponding implications for targeted intervention strategies.This K99/R00 MOSAIC proposal will use secondary data to inform current 10-year CVD risk models usingtheory and data-driven methods to increase CVD prediction model accuracy in H/Ls. The proposed trainingplan establishes a solid foundation for a career investigating H/L CVD risk and resilience factors. The trainingplan leverages substantial resources at The University of Arizona and a mentoring team of senior contentexperts. The candidate will gain the following 1) expertise in H/L CVD disparities 2) advanced knowledge inCVD epidemiology risk and etiology and pathophysiology of atherosclerotic disease 3) applied machinelearning cross-validation and selection of risk prediction models and 4) cultural factors and social capitalinfluencing H/L CVD. The research proposal has three aims focused on evaluating and informing existing 10-year CVD prediction in H/Ls. Using secondary data from the Hispanic Community Health Study/Study ofLatinos (HCHS/SOL) the candidate will (Aim 1 K99) evaluate the prediction accuracy of current 10-yearCVD risk models using a large H/L sample with significant representation of diverse H/Ls (HCHS/SOL). (Aim 2 R00) the candidate will use available data to identify a group of target risk factors that improve risk predictionin H/Ls. (Aim 3 R00) the candidate will test whether adding a social resilience component to CVD riskmodels will improve their prediction accuracy for this group. Machine learning will be used to identify validpredictors of 10-year CVD in Latinos. The social resilience component will capture the multi-dimensionality ofsocial environments (e.g. spouse family neighborhood) using data reduction methods. The proposed researchproposal adopts a holistic view of cardiovascular health to elucidate both risk and resilience factors in thisgrowing ethnic group.