Abstract 10142: Stratification of the Risk of Developing Heart Failure in Patients With Left Bundle Branch Block: Approach Using Artificial Intelligence

Abstract

Introduction: Left Bundle Branch Block (LBBB) is a common ECG abnormality associated with impaired ventricular function. However, large individual variation in heart failure (HF) risk in patients with LBBB has been observed. Hypothesis: Subtle phenotypic changes reflecting cardiac dysfunction are present on ECG and are detectable using artificial intelligence (AI). Methods: All 12-lead ECGs with LBBB recorded from 1/1/2015 to 12/31/2019 were identified in our institution, and analyzed as a matrix of the dimension of 12 (representing twelve leads) x 2500 (representing 250Hz x 10 seconds). A 2-dimensional convolutional neural network was trained to detect ECG features related to HF. The model was then applied to an external cohort of patients with LBBB but without history of known HF. Cumulative incidences of HF admission were compared by stratifying according to tertiles of the model output (low, intermediate and high AI score groups) using Kaplan-Meier curves, log-rank test, and Cox proportional hazard model. Results: The test cohort consisted of 2,664 individuals with LBBB and 490 (18.4%) were admitted for HF during median follow-up of 1,091 days (IQR 209-1,666). While 12.6% in the low score group experienced HF hospitalization, 18.0% in the intermediate score group and 24.5% in the high score group were admitted due to HF (high vs low AI score group: hazard ratio (HR) 2.11 [95%CI 1.68-2.65]; intermediate vs low AI score group: HR 1.44 [95%CI 1.13-1.83]; log rank p <0.0001; Figure ). The association between AI score and HF admission was significant after adjusting for patient characteristics and conventional ECG parameters (high vs low AI score group: adjusted HR 1.94 [95%CI 1.52-2.47], p<0.0001; intermediate vs low AI score group: HR 1.34 [95%CI 1.05-1.72], p=0.02), and in a subpopulation with at least 2-years of follow-up (log rank p=0.002). Conclusions: The AI model exhibited excellent discrimination for HF admission risk in patients with LBBB.