B1904
Title: Accurate and interpretable clinical predictive modeling using high-dimensional electronic health records
Authors: Stathis Gennatas - University of California, San Francisco (United States) [presenting]
Abstract: Clinical predictive modeling requires the training, updating, and monitoring of multiple models on real-time, high-dimensional, electronic health record data. Such models need to be not only accurate but also interpretable/explainable to clinicians and patients alike. A large, longitudinal, electronic health record dataset is used to develop a clinical predictive model for the risk of unplanned hospital readmission using a custom rule-fit implementation. The model allows for the discovery of interactions among clinical, demographic, and social factors that advance the understanding of the risk of hospital readmissions and help set a plan to reduce them.
