Detection of Titin-Associated Electrocardiography Features in Dilated Cardiomyopathy Using Conventional and Deep...

Titin truncating variants (TTNtvs) are the leading genetic cause of dilated cardiomyopathy (DCM). Although recommended, routine genetic testing is frequently not performed owing to resource constraints.

This study sought to identify electrocardiography (ECG) parameters predictive of an underlying TTNtv in DCM patients, comparing conventional ECG analysis with an ECG-based deep neural network (DNN) to identify patients that would benefit most from targeted genetic testing.

This retrospective multinational study compared baseline ECGs from 99 DCM patients with (likely) pathogenic TTNtv with 318 gene-elusive DCM patients. Conventional ECG parameters (eg, QRS duration) were extracted. The DNN was trained to compress ECGs into 21 explainable factors, summarizing relevant ECG features. Discriminative performances of both created models, built using LASSO regularization for variable selection to fit logistic regression model, were compared (eg, C-statistics).

TTNtv patients were younger (50.5 vs 56.9 years; P < 0.001), predominantly male (69.7 vs 54.7%; P = 0.008), and had lower left ventricular ejection fraction (28.0% vs 35.0%; P < 0.001) compared with gene-elusive patients. Conventional ECG analysis identified shorter QRS duration (P < 0.001), prolonged PR interval (P < 0.001), and a trend toward reduced QRS voltage (P = 0.098) as TTNtv characteristics. In the DNN model, factors F1 (inferolateral T wave inversion) and F9 (anterior T-wave inversion), among others, were associated with TTNtv. The conventional and DNN models showed good predictive performance for TTNtv (C-statistics: conventional 0.83, DNN 0.86; P = 0.197).

Conventional ECG and DNN analyses demonstrated similar good predictive performance in distinguishing TTNtv from gene-elusive DCM patients, emphasizing their potential as clinical tools to guide targeted genetic testing.