Mid-life and older adults have the highest rates of suicide of all age groups in the US. Suicide is currently the 11th leading cause of overall mortality; suicide attempts, behaviors, injuries and related complications comprise some of the greatest causes of morbidity. Although considerable progress has been made in mapping the risk architecture of later-life suicide outcomes, major gaps exist in our understanding of the impacts of risk and protective factors, including novel contributors, on racial/ethnic differences in later-life suicide outcomes. Resolving these gaps requires multi-level analyses using longitudinal designs in large, diverse and well-characterized samples. New tools are needed to improve suicide risk prediction in sub-groups, such as by race/ethnicity or gender. To address these needs, this project will create a new resource of an innovative linked dataset of longitudinal patient-reported survey variables and Medicare/Medicaid administrative claims data among over 80,000 adults in two diverse US cohorts "“ the VITAL-DEP (VITamin D and OmegA-3 TriaL-Depression Endpoint Prevention) study and SCCS (Southern Community Cohort Study) "“ with broad racial/ethnic and geographic representation. Follow-up in these cohorts exceeds 2,000,000 person-years, and over 3,000 suicide attempts, injuries or deaths have been identified to date. We propose two Specific Aims. For Aim 1: We will use a multi-level framework, employing individual-, provider- and regional/contextual-level information, to evaluate the role of disparities in risk of late-life suicide events. For Aim 2: We will develop machine learning-based risk prediction models for suicide events that are applicable to racially/ethnically diverse groups. This project features several innovative and impactful elements. First, we will create a new resource for suicide outcomes research in diverse populations by generating the novel linked datasets of survey and CMS (Centers for Medicare & Medicaid Services) variables from two racially/ethnically diverse national cohorts. Second, the new dataset generated by this study will be employed for the innovative tasks of development and preliminary validation of machine learning-driven suicide risk prediction models informed by multi-level contributions of risk and protective factors; we will also apply deep learning methods. Third, the new dataset and suicide risk prediction tools will become unique resources for future studies by suicide researchers to use in addressing health disparities in suicide outcomes. Fourth, in future research, the machine learning-derived models from this study could be tested and refined in other administrative and/or EHR (electronic health records) systems "“ with a view toward improving the quality of suicide predictor data collection in EHRs among racially/ethnically diverse populations. Fifth, future work could also include adaptations to the model to incorporate interval-specific and novel temporal or historical factors (e.g., COVID-19 pandemic, major economic or political events) into the prediction model. Ultimately, innovations in this study will contribute to the critical objective of refining targets for prevention, so that reduction in the total risk of suicide outcomes can be optimized at all levels, for all people. Therefore, this project has direct relevance to the mission of the American Foundation for Suicide Prevention, and we anticipate high impact and scientific value of our proposed study.