The EHR-based machine learning model, FIND-AF 2.0, identifies a high-risk subpopulation for AF diagnosis among patients at elevated risk of stroke and could enable scalable, EHR-driven, risk-guided AF screening.
Abstract
Background
Screening for atrial fibrillation (AF) on the basis of AF risk may be more effective. We aimed to develop, externally validate, and prospectively test a machine learning prediction model using electronic health records (EHRs) to guide AF screening.
Methods
We developed and validated a random forest prediction model for new AF within 6 months, using age, sex, and 10 comorbidities (Future Innovations in Novel Detection of Atrial Fibrillation [FIND-AF] 2.0) in EHRs in the United Kingdom (n=2 081 139), Japan (n=7 795 244), Israel (n=2 166 795), Canada (n=627 919), and China (n=149 145). We conducted a prospective study where participants ≥30 years old without AF and with a CHA2DS2-VASc score ≥2 in men and ≥3 in women, stratified by FIND-AF 2.0 into high and low risk, undertook 4 ECG recordings per day for 3 weeks using a handheld ECG recorder, with a primary outcome of newly diagnosed AF. We estimated stroke risk associated with nonanticoagulated AF in patients with high FIND-AF 2.0 risk in the FinACAF (Finnish Anticoagulation in Atrial Fibrillation) registry of patients with AF (n=229 565).
Results
FIND-AF 2.0 was applicable to all EHRs and showed good to excellent prediction performance (United Kingdom: area under the receiver operating characteristic curve [AUROC], 0.819 [95% CI, 0.809–0.829]; Israel: AUROC, 0.835 [95% CI, 0.828–0.842]; Japan: AUROC, 0.751 [95% CI, 0.745–0.757]; Canada: AUROC, 0.747 [95% CI, 0.741–0.753]; China: AUROC, 0.753 [95% CI, 0.725–0.771]), with AUROC>0.7 in men and women in all cohorts, and improved performance compared with CHA2DS2-VASc and C2HEST (coronary artery disease or chronic obstructive pulmonary disease [1 point each]; hypertension [1 point]; elderly [age ≥75 years, 2 points]; systolic HF [2 points]; thyroid disease [hyperthyroidism, 1 point]). Of 1923 participants from 15 sites in the prospective study (mean age, 70.2 [SD 9.4] years), with a mean of 74.8 (SD, 19.4) ECG recordings, AF was diagnosed in 5 of 902 (0.6%) with low FIND-AF 2.0 risk and 46 of 1021 (4.5%) with high FIND-AF 2.0 risk (odds ratio, 8.46 [95% CI, 3.35–21.40], P<0.001). Median AF burden among high FIND-AF 2.0 risk–detected cases was 33.4% (interquartile range, 5.1%–91.6%), and 96.1% initiated oral anticoagulants. In the FinACAF registry, the rate of ischemic stroke for patients with high FIND-AF 2.0 risk, AF, and no anticoagulants was 6.0 events per 100 patient-years.
Conclusions
The EHR-based machine learning model, FIND-AF 2.0, identifies a high-risk subpopulation for AF diagnosis among patients at elevated risk of stroke and could enable scalable, EHR-driven, risk-guided AF screening.
A supervised machine learning algorithm for AF in a Western Pacific population was derived and demonstrated that higher risk was associated with hospitalisation for other cardio-renal diseases and death.
J. Hsu, C. Hayward, Tobin Joseph et al.· Heart, Lung and Circulation· 0 citations
BACKGROUND AND AIMS
Accurate stroke-risk stratification is central to anticoagulation decision-making in patients with atrial fibrillation (AF), but conventional scores may not fully capture risk heterogeneity. We aimed to develop and externally validate an interpretable weighted score using a time-to-event framework.
METHODS
GLORIA-AF Phase II/III data were used to evaluate 17 baseline predictors using LASSO-penalized Cox regression with stability selection; coefficients were converted into integer weights. Performance was assessed using discrimination, calibration, integrated discrimination improvement (IDI), continuous net reclassification improvement (NRI), and decision-curve analysis. External validation was performed in EORP-AF and APHRS-AF registries.
RESULTS
Among 20,517 patients included in the derivation cohort (mean [SD] age, 69.9 [10.3] years; 9,196 women [44.8%]), 487 (2.4%) had stroke, and 17,397 (84.8%) were receiving anticoagulation at baseline. Ten selected predictors formed a 0-23-point score. The derived score achieved a C-index of 0.661 (95% CI, 0.636-0.685), higher than CHA2DS2-VA (0.626; P < 0.001), CHA2DS2-VASc (0.615; P < 0.001), and the unweighted score (P = 0.016), with no significant difference from the full Cox or machine learning models. In external validation (8,309 patients; 147 strokes), the C-index was 0.652 (95% CI, 0.614-0.690) versus 0.616 for CHA2DS2-VA (95%CI, 0.598-0.634; P < 0.001). IDI/NRI, calibration, and decision-curve analyses supported improved risk differentiation, close calibration, and generally greater net benefit than CHA2DS2-VA. The score retained higher discrimination than CHA2DS2-VA among patients without baseline anticoagulation (P < 0.001).
CONCLUSION
The GLORIA-AF Stroke Weighted Risk Score provides risk refinement beyond CHA2DS2-VA while retaining discrimination consistent with more complex models. External validation supports its transportability and potential adjunctive role in guideline-directed thromboembolic risk assessment.
Late diagnosis of Heart failure (HF) is associated with worse outcomes. We aimed to develop a scalable tool to identify those at high risk of undiagnosed HF using routine electronic health records (EHR). We developed and internally validated a logistic regression (FIND-HF) model for incident HF diagnosis within one year in United Kingdom primary care EHRs (CPRD-Aurum, n=3 520 186), with good prediction performance (area under the receiver operating characteristic curve (AUC) 0.79), equal to more complex modelling techniques. We externally validated FIND-HF in United Kingdom (CPRD-GOLD, n=570 850, AUC 0.72), Japan (JMDC, n=6 820 694, AUC 0.73), United States of America (Epic Cosmos, n=7 710 398, AUC 0.78), and Taiwan (NTUH, n=170 518, AUC 0.85). In a cohort who had undergone HF diagnostics an optimised FIND-HF threshold had a positive predictive value of 21.4% and a negative predictive value of 96.9%. Amongst patients with HF who had undergone cardiac magnetic resonance imaging, high FIND-HF risk compared with low FIND-HF risk as reference, was associated with increased risk of a primary composite outcome of heart failure hospitalisation or cardiovascular death and more advanced adverse remodelling including lower left ventricular ejection fraction. FIND-HF is a scalable EHR-based model which has the potential to help rule out undiagnosed HF in low risk cases, whilst high risk cases are associated with more advanced cardiac dysfunction and worse prognosis.
Y. Nakao, R. Nadarajah, F. Shuweihdi et al.· Scientific Reports· 0 citations
This data-driven, interpretable XGBoost model enables individualized AF risk assessment in middle-aged and older CHD patients, offering a practical tool for early identification and targeted intervention in clinical practice.
Feng Chen, Qin Fu, Ling Li et al.· Frontiers in Cardiovascular...· 0 citations
Background: The Cohorts for Heart and Aging Research in Genomic Epidemiology - Atrial Fibrillation (CHARGE-AF) score is a validated tool for estimating 5-year risk of atrial fibrillation (AF). We aimed to evaluate the utility of repeated CHARGE-AF scores for improving AF risk prediction. Methods: We analyzed participants from the Atherosclerosis Risk in Communities (ARIC) study with complete data from the first four clinic visits (9-year period) and with no prevalent AF by visit 4 (analysis baseline; N = 10,188). CHARGE-AF scores were calculated for each visit using clinical and demographic variables. Incident AF was determined from electrocardiograms, hospital discharge codes, and death certificates over a median follow-up of 19.5 years. Four Cox regression models were assessed: model 1 included only the visit 4 CHARGE-AF score, and subsequent models added prior CHARGE-AF scores in stepwise fashion. C-statistics were used to evaluate model discrimination, and comparison of observed versus predicted risk was employed to evaluate calibration. Secondary analysis restricted follow-up to five years. Results: During follow-up, 2,519 participants developed AF (14.2 cases per 1,000 person-years). The mean age of participants at start of follow-up was 62.8 (standard deviation 5.6) years. In the primary analysis, each 1% increase in the visit 4 CHARGE-AF score was associated with incident AF (Model 1 HR = 1.14, 95% CI 1.13-1.15). Addition of scores from prior visits did not significantly improve model discrimination (C-statistic: 0.702-0.703 for all models). Sex modified the association between a 1% increase in CHARGE-AF score and incident AF, with a stronger association among females (Model 1 HR = 1.21, 95% CI: 1.19-1.22) than among males (HR for model 1 = 1.12, 95% CI: 1.11-1.13). Similar patterns were observed in the secondary (5-year restricted) analysis. Conclusions: A single measurement of the CHARGE-AF score provided strong predictive value for incident AF, with the addition of prior scores offering limited incremental benefit. These findings suggest that, in clinical settings with longitudinal data, the most recent assessment is sufficient for AF risk prediction.
Eduardo Gonzalez Villarreal, F. Norby, L. Y. Chen et al.· medRxiv· 0 citations
Background Artificial intelligence–enabled electrocardiography (AI-ECG) has emerged as a potential method for identifying atrial fibrillation (AF) from sinus rhythm. However, its clinical utility and interpretability in routine practice remain uncertain. Objectives The objective of the study was to assess the performance and explainability of an AI-integrated ECG system for AF risk stratification in a multicenter real-world cohort. Methods We enrolled 665 patients aged ≥40 years who underwent 12-lead ECGs using an AI-enabled electrocardiograph (FCP-9900). The device automatically assigned AF risk into 4 categories (low, mid-low, mid-high, and high). Machine-learning models—support vector machine, adaptive boosting, and artificial neural networks—were developed using clinical and ECG-derived variables. Internal validation used stratified 10-fold cross-validation, and external validation used an independent cohort. Feature contributions were assessed with SHapley Additive exPlanations. Results AF prevalence increased across AI-ECG risk categories, with significantly higher odds in the mid-high and high groups vs low. Model 2, which incorporated CHADS2 and CHA2DS2-VASc scores, achieved strong discrimination in internal and external validation (support vector machine AUC 1.00; adaptive boosting 0.97-0.98; artificial neural network 0.89-0.95), outperforming AI-ECG alone (area under the receiver operating characteristic curve: 0.64-0.69). SHapley Additive exPlanations analysis showed CHA2DS2-VASc as the most influential predictor, whereas AI-ECG provided modest incremental value. Conclusions AI-ECG provides rapid, low-cost AF risk estimation from a single sinus rhythm ECG; however, its predictive performance is modest compared with clinical scores. At present, AI-ECG may complement, but not replace, traditional risk stratification.
Kouki Matsuo, Y. Sobue, T. Miyake et al.· JACC: Advances· 0 citations