Integrating Clinical Chart and Laboratory Data for Predicting Heart Failure Recurrence

Heart failure (HF) has the highest 30-day rehospitalization rate among all medical and surgical conditions.This study proposes a method for predicting HF recurrence using longitudinal chart data and laboratory measurements from patients in the critical care unit (CCU).We analyzed data from 8,720 HF patients in the MIMIC-III database, extracting time series of five physiological measurements and deriving 11 statistical features from each, along with demographic and laboratory data, totaling 67 features.We employed a Random Forest (RF) classifier to predict HF recurrence over four time intervals (week, month, half a year, and a year) and conducted survival analysis using HF-recurrence probabilities from the trained model.For the test set, we achieved a mean F1 score of 0.570.09,sensitivity of 0.770.03,and specificity of 0.770.04.Our findings suggest that machine learning algorithms applied to chart data and laboratory measurements could potentionally help guide patient discharge decisions and reduce HF recurrence.