Atrial fibrillation burden and risk of new development through artificial intelligence analysis of an electrocardiogram in hospitalized patients with heart failure and preserved ejection fraction. (25th November 2020)

• Record Type: Journal Article
• Authors: Verbrugge, F.H; Reddy, Y.N.V; Kapa, S; Borlaug, B.A
• Is Part Of: European heart journal. Volume 41:(2020)Supplement 2
• Publication Date: 2020-11-25
• Subjects: ECG and Arrhythmia Analysis; Cardiology -- Periodicals; Heart -- Diseases -- Periodicals; 616.12005
• DOI: 10.1093/ehjci/ehaa946.3443
• Languages: English
• ISSNs: 0195-668X

Abstract

Abstract: Background: Analysis of a 12-lead electrocardiogram (ECG) in sinus rhythm by artificial intelligence (AI) has demonstrated to identify the risk of underlying paroxysmal atrial fibrillation (AF) with reasonable accuracy in the overall population. Purpose: This study investigates whether AI based ECG analysis can predict new AF development in patients with heart failure and preserved ejection fraction (HFpEF) without previous AF history, as well as predict AF burden in patients with AF history. Methods: This retrospective cohort study includes 424 patients with HFpEF, consecutively admitted to receive treatment with intravenous diuretics for congestion. A previously validated AI algorithm that provides an AF risk score based on ECG analysis was applied in all subjects. Patients were stratified according to AF history: (1) no history; (2) paroxysmal AF; (3) persistent AF; or (4) permanent AF. In patients without AF history, the impact of AI-predicted AF risk on new AF development was assessed. In patients with previous AF episodes, the relationship between AI-predicted AF risk and AF burden as well as underlying echocardiography substrate was evaluated. Results: Eighty-three patients had paroxysmal AF (19.5%), 48 persistent AF (11%), and 121 permanent AF (28.5%). AF patients were older, with lower body mass index and higher heart rate. Lower systolic blood pressure, larger left atrial volume index (LAVI), worse diastolic function, more tricuspid valve regurgitation, and more need for pacing were observed with increasing AF severity. In 172 patients without AF history followed for 979±875 days, 61 developed AF (35%) after a median time of 2, 118 days. AI-predicted AF risk was associated with new-onset AF after adjustments for age, gender, systolic blood pressure, body mass index and LAVI [HR (95% CI) = 1.15 (1.04–1.28); P-value=0.007; figure]. Only 5/61 patients who developed AF during follow-up (8.2%) were anticoagulated at baseline. In patients with previous AF, the AI-predicted risk score increased significantly with AF burden (46±25% in paroxysmal AF, 59±22% in persistent AF, and 69±17% in permanent AF; P-value<0.001). The score was weakly but significantly correlated with LAVI (Pearson's r=0.27; P-value<0.001) and right ventricular systolic pressure (Pearson's r=0.30; P-value<0.001). When stratifying patients according to AI-predicted AF risk (0–25% versus 26–50% versus 51–75% versus 76–100%), there was a progressively longer time since onset of the first AF episode [29 months (4–77 months) versus 9 months (0–62 months) versus 38 months (2–88 months) versus 54 months (3–112 months), respectively; P-value=0.046]. Conclusions: An AI-predicted AF risk score based on ECG analysis was associated with AF burden in HFpEF patients and did independently predict new-onset AF in those without previous history. Funding Acknowledgement: Type of funding source: Foundation. Main funding source(s): Belgian American Educational Foundation (B.A.E.F.); Special Research Fund (BOF) of Hasselt University (Hasselt, Belgium)