Risk-Guided Screening for Atrial Fibrillation Using Electronic Health Records

Introduction

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and is strongly associated with an increased risk of ischemic stroke. Early detection of AF is critical because anticoagulant therapy significantly reduces stroke risk. However, AF is often asymptomatic or paroxysmal, making routine detection challenging. This has driven interest in risk-guided screening methods that efficiently identify those at higher risk for AF to focus diagnostic resources effectively.

Background and Rationale

Screening for AF based solely on age or broad population criteria may result in limited efficiency and resource utilization. Leveraging electronic health records (EHRs) combined with machine learning algorithms offers the potential to stratify patients by risk more precisely, enhancing targeted screening approaches. By identifying patients with a higher likelihood of developing AF within a short timeframe, healthcare providers can prioritize them for intensified monitoring.

Development of the FIND-AF 2.0 Model

Researchers developed FIND-AF 2.0, an advanced random forest machine learning model, using large-scale EHR data from multiple countries, including the UK, Japan, Israel, Canada, and China. It utilizes simple clinical variables readily available in EHRs — age, sex, and ten comorbidities — to predict the risk of newly diagnosed AF within six months. This concise input set facilitates easy integration into diverse healthcare systems.

Validation Across Cohorts

FIND-AF 2.0 was externally validated in these international cohorts, demonstrating high prediction accuracy with area under the receiver operating characteristic curve (AUROC) values ranging from approximately 0.75 to 0.84 across regions. The model outperformed existing risk scores like CHA2DS2-VASc and C2HEST in all tested populations, showing robust applicability regardless of geographic or ethnic variations.

Prospective Study Design and Results

A prospective multicenter study involving 1,923 participants aged 30 and above, all without known AF but with an elevated stroke risk (men with CHA2DS2-VASc ≥2, women with ≥3), used FIND-AF 2.0 to stratify participants into high- and low-risk groups. Participants performed four daily handheld ECG recordings for three weeks to detect new AF.

AF diagnosis rates differed significantly between the risk groups: 4.5% in the high-risk group versus 0.6% in the low-risk group, with an odds ratio of 8.46 indicating a markedly increased likelihood of detection in those flagged by FIND-AF 2.0. Most newly diagnosed AF patients (96.1%) initiated oral anticoagulation promptly, highlighting the clinical utility of this screening method.

Stroke Risk Analysis in High-Risk AF Patients

Using data from the Finnish Anticoagulation in Atrial Fibrillation (FinACAF) registry, the study further estimated stroke risk in patients identified as high risk by FIND-AF 2.0 but not receiving anticoagulation. The ischemic stroke incidence was 6 events per 100 patient-years, reaffirming the high stroke risk associated with undiagnosed or untreated AF in this population.

Clinical Implications

The application of FIND-AF 2.0 in routine clinical practice could revolutionize AF screening by enabling scalable, cost-effective risk stratification directly from EHR data without requiring additional clinical assessments. This may facilitate early AF detection and targeted initiation of stroke prevention therapies.

The model’s simplicity and strong validation across diverse health systems underscore its generalizability. By focusing screening resources on those most likely to benefit, healthcare systems can improve outcomes and reduce the downstream burden of stroke-related morbidity and mortality.

Considerations and Future Directions

While promising, the model’s performance depends on quality and completeness of EHR data, which can vary by healthcare setting. Integrating such algorithms into clinical workflows requires appropriate infrastructure, clinician education, and patient engagement.

Future research should explore long-term outcomes of risk-guided screening implementation, cost-effectiveness analyses, and its integration with wearable or continuous monitoring technologies to capture paroxysmal AF episodes missed by intermittent ECG.

Summary

Using EHR-based machine learning, FIND-AF 2.0 identifies high-risk patients for atrial fibrillation effectively, enabling targeted ECG screening to diagnose AF early in those at increased stroke risk. Early detection facilitates timely anticoagulation, potentially preventing strokes and improving patient outcomes. The approach promises a scalable and adaptable strategy for global public health against AF-related stroke.