FIND-AF: a widely applicable artificial intelligence algorithm to target systematic screening for atrial fibrillation in older individuals through primary care electronic health records. (19th May 2022)

• Record Type: Journal Article
• Authors: Wu, J; Nadarajah, R; Raveendra, K; Cowan, JC; Gale, CP
• Is Part Of: Europace. Volume 24:Supplement 1(2022)
• Publication Date: 2022-05-19
• Subjects: Arrhythmia -- Treatment -- Periodicals; Cardiac pacing -- Periodicals; Catheter ablation -- Periodicals; Heart -- Physiology -- Periodicals; Electrophysiology -- Periodicals; 617.4120645
• DOI: 10.1093/europace/euac053.565
• Languages: English
• ISSNs: 1099-5129

Abstract

Abstract: Funding Acknowledgements: Type of funding sources: Foundation. Main funding source(s): British Heart Foundation Background: Systematic screening for atrial fibrillation (AF) in people ≥75 years of age can improve detection rates, anticoagulant prescription and clinical outcomes but is inefficient.(1) A large proportion of European populations are registered in primary care with a routinely-collected electronic health record (EHR).(2) An algorithm embedded in this system to identify people at higher risk of incident AF could facilitate targeted AF screening. Purpose: To develop and internally validate a widely-applicable artificial intelligence (AI) algorithm for predicting incident AF in people ≥75 years of age using primary care EHRs. Methods: We identified people who were ≥75 years of age (1998 – 2018), in the nationwide Clinical Practice Research Datalink (CPRD)-GOLD primary care EHR dataset and followed them until a diagnosis of AF, or withdrawal from CPRD, or 6 months. Each subject had 81 features including age, sex, ethnicity and comorbidities. Algorithms developed with random forest (RF) and multivariable logistic regression (MLR), were compared by area under receiver operating curve (AUROC) and the proportion of patient EHRs to which the algorithms could be applied Results: 440, 000 patients were studied, with 3922 occurrences of AF. The RF algorithm achieved an AUROC of 0.77 after 10 fold cross-validation, 12% better than the MLR algorithm (0.68). Notably the RF algorithm could be applied to all EHRs. At 75% sensitivity, the RF algorithm would reduce the potential number needed to screen for one new case of AF to 11, an improvement of over 6-fold compared to using age alone. Conclusions: This study showed a novel AI algorithm that can be widely applied in nationwide European primary care EHRs to target screening for AF in a population that derives clinical benefit.