Applying big data and analytics to timely and accurately predict prevalent acute diseases such as stroke can significantly reduce health care spending and improve patient outcomes. This session explores insights from a study on whether and how we can improve our ability to distinguish between stroke and stroke-like conditions ('stroke mimics') at hospital admission. Machine learning (ML) models analyzed data from 2012-2014 patient discharge records and patient-level hospitalization data from all Florida hospitals, and social determinants of health (SDOH) information. Analysis shows that ML models can significantly improve the predictive accuracy of stroke diagnosis where the most important predictors are a patient's age, number of chronic conditions, and comorbidity index. While SDOH are associated with stroke incidence and severity, adding community-level SDOH factors contributes minimally to the diagnosis prediction. Explore how developing and integrating individual-level SDOH screening tools into EHRs to improve patient risk assessment and prevention.