Artificial Intelligence-Enabled ECGs for Atrial Fibrillation Identification and Enhanced Oral Anticoagulant Adoption: A Pragmatic Randomized Clinical Trial

Abstract

Background: Atrial fibrillation (AF) is often underdiagnosed and undertreated by noncardiologists. This study evaluated whether artificial intelligence-enabled ECG (AI-ECG) alerts could improve AF diagnosis and non-vitamin K antagonist oral anticoagulant prescriptions by noncardiologists.

Methods: In this open-label, cluster randomized controlled trial (NCT05127460) at 2 hospitals in Taiwan, noncardiologists were randomized to an intervention group (AI-ECG alerts) or control group (usual care). Alerts were sent to physicians when AI-ECG identified AF in emergency or hospitalized patients at risk of stroke (CHA₂DS₂-VASc ≥1 for men, ≥2 for women), excluding those with prior AF or oral anticoagulant use. Primary end points included a non-vitamin K antagonist oral anticoagulant prescription within 90 days after discharge, new AF diagnosis, echocardiogram arrangements, and cardiologist visits. Secondary end points were ischemic stroke, cardiovascular death, and all-cause death.

Results: A total of 8857 and 8960 patients were treated by 120 and 113 noncardiologists in the intervention and control groups, respectively; 275 and 245 patients had AI-detected AF. The non-vitamin K antagonist oral anticoagulant prescription rate was significantly higher in the intervention group (23.3% versus 12.0%; hazard ratio [HR], 1.85 [95% CI, 1.11-3.07]). The intervention group also had a higher rate of AF diagnosis (HR, 1.40 [95% CI, 1.03-1.90]). No significant differences were observed in echocardiogram arrangements, cardiologist visits, or the rates of ischemic stroke, cardiovascular death, and all-cause death.

Conclusions: An AI-ECG alert for AF identification promoted non-vitamin K antagonist oral anticoagulant prescriptions among noncardiologists, thus reducing the disparity in AF care quality between cardiologists and noncardiologists.

Registration: URL: https://clinicaltrials.gov/; Unique identifier: NCT05127460.

Keywords: ECG; artificial intelligence; atrial fibrillation; deep learning; ischemic stroke; non–vitamin K antagonist oral anticoagulants; randomized clinical trial.

Figure 1. Consolidated standards of reporting trials–AI flow diagram.

The analysis of primary and secondary outcomes in each group was conducted using an intention‐to‐treat approach. AI‐ECG indicates artificial intelligence–enabled ECG; eGFR, estimated glomerular filtration rate; and NOAC, non–vitamin K antagonist oral anticoagulant.

Figure 2. Comparison of primary end points between the intervention and control groups.

Kaplan–Meier curve for the primary end point. The table shows the at‐risk population and cumulative risk for the given time intervals in each risk stratification. The corresponding log‐rank test P values were as follows: diagnosis of new‐onset AF (P=0.030), NOAC usage (P=0.016), echocardiogram (P=0.250), and visits to cardiologists (P=0.063). AF indicates atrial fibrillation; HR, hazard ratio; and NOAC, non–vitamin K antagonist oral anticoagulant.

Figure 3. Comparison of secondary end points and post hoc analysis between the intervention and control groups.

Forest plot for secondary end points (A) and post hoc analysis (B). Patients with a history of heart failure were excluded from the analysis of new‐onset heart failure. HR indicates hazard ratio.

Figure 4. Summary of AI‐ECG results and follow‐up events.

Bars represent the proportion of event by each condition, which are simultaneously presented digitally. The proportion of <5% are hidden. The reasons for NOAC nonprescription included patient expiration, active bleeding (traumatic or nontraumatic), and cases deemed unsuitable for NOAC use. Not suitable for NOAC: moderate to severe mitral stenosis, mechanical valves, newly implanted bioprosthetic valves, and left ventricular thrombosis. AF indicates atrial fibrillation; AI, artificial intelligence; AI‐ECG indicates artificial intelligence–enabled ECG; and NOAC, non–vitamin K antagonist oral anticoagulant.