Prediction of Paroxysmal Atrial Fibrillation With Incorporating Genomic Information Into AI-Based ECG Analysis

BACKGROUND: It remains unclear whether incorporating genetic information and blood biomarkers can improve artificial intelligence (AI) models predicting paroxysmal atrial fibrillation (PAF) from sinus rhythm electrocardiograms (ECGs). OBJECTIVES: This study aimed to explore the potential value of polygenic risk scores (PRS) and blood biomarkers in AI-based ECG (AI-ECG) analysis for predicting PAF. METHODS: Patients from 7 institutions in Japan were recruited between March 1, 2020 and December 31, 2021. AI-ECG scores and PRS were calculated using previously reported AI model and PRS algorithm. High-sensitivity C-reactive protein, N-terminal pro-B-type natriuretic peptide, and cell-free DNA (cfDNA) levels were also evaluated. RESULTS: A total of 2,128 cases were included (PAF: 1,055 of 2,128 [49.6%]). AI-ECG scores were significantly higher in PAF group than in non-AF group, with an area under the curve (AUC) of 0.882 (95% CI: 0.866-0.897) for the receiver-operating characteristic (ROC) curve in predicting PAF. PRS was also significantly higher in PAF group, with AUC-ROC of 0.655 (95% CI: 0.630-0.680). Although combining AI-ECG scores and PRS did not improve AUC-ROC, a significant improvement was observed in the net reclassification improvement of 0.467 (95% CI: 0.374-0.558) and integrated discrimination improvement of 0.028 (95% CI: 0.020-0.035). Among blood biomarkers, only cell-free DNA, along with PRS and AI-ECG, was associated with PAF in multivariable analysis; however, its contribution to PAF prediction was limited. CONCLUSIONS: Genetic information may provide complementary insights and improve risk stratification for the prediction of PAF using AI-ECG.