Using Geospatially Explainable Machine Learning to Examine Risk Factors of Out-Of-Hospital Cardiac Arrest Patient Survival Status

CGA "Brown Bag" Presentation by Dr. Jielu (Jenny) Zhang 

Abstract: Out-of-Hospital Cardiac Arrest (OHCA) impacts over 350,000 individuals annually in the U.S., with a survival rate below 6%. To address this, the study proposes a comprehensive geographically explainable AI (GeoXAI) framework that integrates spatial variation and causal inference to improve survival outcomes. The framework includes three components: (1) Overlayed Spatio-Temporal Optimization (OSTO) to optimize AED placement based on real-world mobility data, improving coverage under practical constraints; (2) Spatial Counterfactual Explainable Deep Learning (SpaCE) to identify spatially varying risk factors influencing survival and enhance predictive accuracy; and (3) Spatially-Aware Causal Inference (SpatialCausal) to estimate both individual and regional treatment effects, uncovering key variables like AED use before EMS arrival and the presence of witnesses. Together, these components offer a robust, adaptable solution for data-driven health interventions and policy planning in OHCA and other spatially influenced health challenges.

Speaker bio: Jielu (Jenny) Zhang is a postdoctoral researcher at the Center for Geographic Analysis at Harvard University. She holds a Ph.D. in Geography and an M.S. in Computer Science from the University of Georgia. With a strong interdisciplinary background, her research focuses on developing geospatial artificial intelligence (GeoAI) methods for public health applications. Dr. Zhang specializes in explainable AI, spatial causal inference, and spatio-temporal optimization of healthcare services. She has led NIH-funded projects, including the development of SpaCE, a model estimating spatial risk factors for out-of-hospital cardiac arrest (OHCA), and SpatialCausal, a deep learning framework for identifying causal effects across geographic areas. Her work has been published in journals such as International Journal of Geographical Information Science, Journal of the American Heart Association, and Circulation. She also developed an optimization model for AED placement and is currently exploring the use of large language models (LLMs) for predicting and explaining health outcomes using in-context learning.