Abstract 14409: Deep Learning Approach for Analyzing Chest X-rays to Predict Cardiac Events in Heart Failure

Abstract

Introduction: A deep learning (DL) model based on a chest X-ray (CXR) was reported to predict elevated pulmonary artery wedge pressure (PAWP) as congestive heart failure (HF). Hypothesis: We hypothesize that a previously developed application of a CXR-based DL algorithm could also be used to predict re-hospitalized HF in patients with HF. This study aimed to 1) investigate the role of probability of elevated PAWP to predict clinical outcomes in association with other parameters, and 2) to evaluate whether probability of elevated PAWP based on DL added prognostic information to other conventional clinical prognostic factors in HF. Methods: We evaluated 192 patients hospitalized with HF. We used a previously developed AI model to predict HF and calculated probability of elevated PAWP in the study cohort. The primary endpoint was readmission following HF and cardiac death. Results: Probability of elevated PAWP was associated with echocardiographic parameters including left atrial size, filling pressure, and pulmonary artery pressure. During a median follow-up period of 58 months, 57 patients were either readmitted or died. Probability of elevated PAWP appeared to be associated with worse clinical outcomes. After adjustment for readmission score and laboratory data in a Cox proportional-hazards model, probability of elevated PAWP at pre-discharge (per 1SD) was associated with event free survival, independent of elevated left atrial pressure (LAP) by echocardiography (hazard ratio:1.39, 95%CI:1.17-1.65, p<0.001). In sequential Cox models, a model based on clinical data (χ2, 16.8) was improved by elevated LAP (χ2,24.4; p =0.005), and increased further by probability of elevated PAWP (χ2,41.1; p <0.001). Conclusions: This study suggests that probability of elevated PAWP provides important incremental prognostic value for predicting readmission and cardiac mortality risk assessment in patients with HF.