Abstract 10932: Prediction of 30-Day Hospital Readmission in High-Risk Atherosclerotic Cardiovascular Disease Patients Using Machine Learning Methods on Electronic Health Record Data from Medical Information Mart for Intensive Care-3 Database

Abstract

Introduction: Despite significant therapeutic advancements, Atherosclerotic Cardiovascular Disease (ASCVD) patients require frequent hospitalization. Machine learning (ML) algorithms present an opportunity to develop improved and more generalizable prediction models for 30-day hospital readmission due to ASCVD. Objectives: The current study aims to leverage electronic health record (EHR) data to predict 30-day hospital readmission in high-risk ASCVD patients. Methods: This study utilized the MIMIC III database (a publicly available large, single-centre critical care database of 46,520 patients) comprising of deidentified ~60,000 ICU admissions. Adult patients with the first admission due to ASCVD event and whose length of stay (LOS) in the ICU was >48 hours were included. To develop a prediction model, features representing groups of diagnosis data along with demographics and length of stay in the hospital were used. The performance of ML models was evaluated using the area under the receiver operating characteristic curves (AUCs). Results: Our cohort consisted of 22,666 admissions (mean age 70.1 years) due to an ASCVD event. Out of the 9022 readmissions within 180 days, 60.5% were observed within 30 days of discharge. Among the 17 prospective predictors age, LOS and first hospitalization due to coronary atherosclerosis, heart failure, and haemorrhage issues were the most important factors to predict readmission within 30 days (Fig A). Random Forest was the best performing model with an AUC of 0.66 (Fig B). Conclusions: This model can be helpful in predicting the readmission of high-risk patients after the first ASCVD event which might address the huge unmet need of aiding healthcare resource planning, better patient care, and prevention of rehospitalization and death by intensifying treatment interventions in those patients and taking appropriate decisions to discharge patients based on their age, LOS and ASCVD events resulting in first hopitalization.