Artificial intelligence based electrocardiogram analysis for prediction of cardiovascular prognosis in dystrophinopathy patients

Abstract Introduction Dystrophinopathy is a spectrum of muscle disease caused by mutations in the dystrophin gene located on the X chromosome resulting in progressive muscle weakness. Ventricular dysfunction due to weakening cardiac muscle is the leading cause of morbidity and mortality in these patients. However, predicting cardiovascular prognosis in dystrophinopathy patients is difficult because degree and progression of cardiac muscle involvement usually does not correlate with the extent of skeletal and/or respiratory muscle weaknesses. Recent studies have shown that artificial intelligence (AI) can be applied to detect cardiac diseases such as atrial fibrillation and ventricular dysfunction using electrocardiogram (ECG) data. Purpose The purpose of this study was to investigate if AI based 12-lead ECG analysis may be used to detect early ventricular dysfunction and predict cardiovascular prognosis in dystrophinopathy patients. Methods We developed an AI based ECG model using a convolutional neural network to detect ventricular dysfunction and predict cardiovascular outcomes from 12-lead ECG data. The medical records of patients genetically diagnosed with dystrophinopathy between January 2005 and May 2023 at a single tertiary center were reviewed. ECG data, age at ECG, and ventricular function on echocardiography were retrieved for AI analysis. Outcomes were cardiomyopathy, defined as ejection fraction (EF) <55% on echocardiography, heart failure symptoms requiring hospitalization, and death. We calculated the area under the receiver operating characteristic curve (AUROC) to evaluate model performance on the development data set. Results Total 610 patients were included in the study. Mean age at last follow up was 17.9 ± 6.9 years with mean follow up duration of 9.6 ± 5.1 years. Fifty-eight percent (354/610) developed cardiomyopathy at mean age of 14.9 ± 4.4 years, 5.9% (36/610) were hospitalized due to heart failure symptoms at mean age of 19.2 ± 4.9 years, and 11.5% (70/610) died during follow up at mean age of 19.7 ±4.1 years. AI model was able to detect cardiomyopathy with an AUROC of 0.914 when ECG and age at ECG were used as inputs. AUROC for prediction of heart failure requiring hospitalization and death within 10 years were 0.919 and 0.916, respectively when ECG was used as the input. Conclusion AI based ECG models might be a useful and accurate method to detect cardiomyopathy and predict cardiovascular outcomes in dystrophinopathy patients allowing for development of individualized follow up schedule and timely intervention in these patients.