Electrocardiogram-based artificial intelligence to predict incident heart failure risk

Abstract Background Heart failure (HF) is a major public health problem associated with substantial morbidity. ECG-based artificial intelligence (AI) may enable efficient identification of individuals at elevated risk for future HF and enable targeted risk factor modification and preventive efforts. Existing models are proprietary with modest or inconsistent performance. Purpose To develop and validate a generalizable and publicly available convolutional neural network to predict incident HF using the 12-lead ECG waveform (Electrocardiogram-to-Heart Failure, "ECG2HF"). Methods We developed ECG2HF in 100,117 patients receiving longitudinal ambulatory care in our hospital, and validated the model among individuals without HF in one internal test set and 2 external test sets. HF events at 10 years were identified using a validated natural language processing model which has been shown to ascertain HF events using electronic health records with clinical adjudication committee-level accuracy. Discrimination was quantified by calculating the area under the receiver operating characteristic curve (AUROC) as well as hazard ratios per quintile of ECG2HF risk. We also compared discrimination and net reclassification (at