PROJECT SUMMARY / ABSTRACT The purpose of this award is to provide Dr. Brian Rice, Assistant Professor of Emergency Medicine at Stanford University, the support necessary for his transition from a junior investigator into an independent clinician- scientist using applied biomedical informatics to address health disparities. Dr. Rice is an emergency medicine physician with an advanced degree in epidemiology and global health, and a background in computer programming and artificial intelligence. His long-term goal is to utilize his interdisciplinary training to develop and implement machine learning tools to empower precise, high-value clinical decision-making surrounding emergency care and transport in historically disadvantaged populations. His training activities focus on advancing his ability to apply biomedical informatics to address health disparities via these training objectives: 1) expanding his skills in data management and computational statistics 2) learning methods for community- engaged and participatory approaches to health disparities research, and 3) acquiring new skills machine learning and classification model building. The candidate has convened a mentorship team that includes Dr. Tina Hernandez-Boussard, a biomedical artificial intelligence expert with a focus on improving transparency and minimizing bias in machine learning models to make them more equitable and generalizable, and Dr. Stacy Rasmus, a leading Alaska Native behavioral scientist with extensive experience successfully conducting community-engaged qualitative research in rural Alaska. The research proposal builds off the candidate’s prior work with air medical evacuation (medevacs) in rural Alaska which established the central hypothesis that medevacs can be classified as appropriate or inappropriate by machine learning models built on outcome data and enriched by qualitative methods. This central hypothesis will be tested by the following specific aims: 1) define the burden and outcomes of medevacs in rural Alaska; 2) identify key context-specific contributors to medevac utilization in rural Alaska; and 3) develop machine learning models to classify appropriateness of medevac utilization in rural Alaska. The research proposed in this application is innovative because it employs accepted methods of machine learning classification modelling and applies them to novel fields of medevac and Alaska Native health disparities. The significance of the proposed training grant is it will provide the data and the skills required for Dr. Rice to subsequently study the implementation of these models as a decision tool in a future R01-level application. Ultimately, this continuum of research has the potential to decrease expenses and improve safety by redirecting medevac resources towards patients whose time-sensitive conditions benefit from medevacs and away from patients that incur risk and cost without benefit, both in Alaska Native communities in rural Alaska and for all Americans living in rural regions nationwide.

PROJECT NARRATIVE Alaska Native people predominate in rural Alaska where the lack of roads creates a dependency on air medical evacuation (medevacs) for access to emergency care which in turn creates the highest medevac rate, the highest airplane crash rate, and the poorest access to emergency care of any region in the US. Work proposed in this grant aims to generate a novel medevac outcomes database from existing electronic health records, combine those data with community-engaged qualitative methodology to ultimately create context-specific machine learning models able to classify the appropriateness of medevac utilization. This continuum of research has the potential to directly address the disparities in access to and outcomes of emergency care experienced by both Alaska Native people in rural Alaska and by rural Americans more generally, by improving the safety, timeliness and appropriateness of medevacs.