Project Abstract Diagnosis of Alzheimer’s disease (AD) is crucial for individuals to pursue treatments and plan for the future. Unfortunately, AD is underdiagnosed in community settings compared to the estimated prevalence from longitudinal cohort studies. AD underdiagnosis is exacerbated in understudied populations, including Hispanic/Latino (HL) and non-Hispanic African American (NH-AfAm) groups. Mining patients’ electronic health records (EHR) using machine learning may help identify patients with undiagnosed AD. Prior studies have identified individual comorbidities associated with AD. However, patients accumulate disease conditions sequentially accumulate over time. These pathways of disease accumulation are known as disease trajectories. Few studies have investigated disease trajectories in AD and none have comprehensively evaluated them in understudied populations. Disease trajectories derived from the EHR have the potential to predict undiagnosed AD as the next diagnosis follows from the prior diagnoses. Current models that predict AD do not utilize disease trajectories or account for AD underdiagnosis. EHR data can be complemented by genetics to determine AD risk. Incorporating polygenic risk with EHR data to identify undiagnosed AD has not been evaluated. To address these challenges, our objectives are to identify undiagnosed AD in non-Hispanic white (NH- white), HL, and NH-AfAm groups utilizing disease trajectories and genetics followed by validation and replication. We leverage existing resources from millions of patients with EHR and hundreds of thousands with genetic data linked to the EHR. We propose the following aims. Aim 1 will identify disease trajectories in AD. Aim 2 will identify undiagnosed AD based on disease trajectory while mitigating the bias of underdiagnosis in HL and NH-AfAm groups. We will validate patients predicted to have undiagnosed AD via multiple outcome measures and replicate our results in another EHR cohort, All of Us. Aim 3 will integrate polygenic risk with disease trajectories to detect undiagnosed AD. We will validate these patients predicted to have undiagnosed AD via multiple outcome measures, and replicate our results in All of Us. This proposal aligns with our long-term goal to realize precision medicine in dementias for all racial and ethnic groups by leveraging EHR, genomics, and computational tools. Our contributions will enable the application of disease trajectories in precision medicine, the detection of undiagnosed AD at UCLA and other health systems, and future work to reduce the bias of underdiagnosis in understudied populations.