Mahrhoztal Rosier
Pronouns: He / Him
Research Mentor(s): Matthew Hodgman
Research Mentor School/College/Department: Department of Computational Medicine and Bioinformatics / Medicine
Program: UROPF
Session: Session 3 (11:00am – 11:50am)
Authors: Mahrhoztal Rosier, Matthew Hodgman, Emily Wittrup, Kayvan Najarian
Abstract
Heart failure is one of the leading causes of death worldwide. While many studies attempt to predict long-term mortality in heart failure patients, their models lack interpretability and incorporation of time series information. Using data from heart failure patients in the MIMIC-III database, we develop and test methods to encode time-series electronic health record (EHR) data to maximize interpretability and signal for predicting one-year mortality or readmission. We assess the efficacy of the encoding methods with interpretable models including logistic regression, random forest, and a tropical geometry fuzzy neural network. We anticipate that by leveraging the temporal relationships in EHR data, predictive models will be more accurate and interpretable for clinical implementation.
