In order to bring you the best possible user experience, this site uses Javascript. If you are seeing this message, it is likely that the Javascript option in your browser is disabled. For optimal viewing of this site, please ensure that Javascript is enabled for your browser.

Latest Science: Machine-learning prediction algorithm identifies patients at high risk of undiagnosed AF

28 Aug 2021
Presented science not to be missed

The intermittent and symptomatic nature of atrial fibrillation (AF) means that many cases are undiagnosed. An AF risk prediction algorithm was developed using machine-learning techniques from a UK dataset of 2,994,837 patients and was found to be more effective than existing models at identifying patients at risk of AF.1 However, whether this risk prediction algorithm is accurate at identifying undiagnosed AF in a real-world primary-care setting is unknown.

As part of the on-demand programme, Doctor Alexander Cohen (Guy's and St Thomas' NHS Trust Hospitals, King’s College London, UK) presents the results from the PULsE-AI randomised trial, which was designed to assess the real-world ability of the machine-learning AF risk prediction algorithm coupled with diagnostic testing for the identification of patients with AF compared with routine care.

Eligible participants (aged >=30 years and without an existing AF diagnosis) registered at participating UK general practices were randomised to intervention and control arms. Intervention arm participants identified at highest risk of developing AF (algorithm risk score >=7.4%) were invited for a 12‑lead electrocardiogram (ECG) followed by 2 weeks of home-based ECG monitoring with a KardiaMobile device. Control-arm participants were managed routinely. The primary endpoint was the combined number of diagnoses of AF, atrial flutter and fast atrial tachycardia during the trial in participants judged by the algorithm to be at high risk of AF. The number of diagnoses across arms were compared using penalised logistic regression analyses, adjusted for baseline characteristics.

A total of 23,745 participants from six general practices were randomised. Of the 906 in the intervention arm with a high risk score, 255 (28.1%) consented to diagnostic testing. AF or related arrythmias were diagnosed in 5.63% and 4.93% of participants in the intervention and control arms, respectively (odds ratio [OR] 1.15; 95% confidence interval [CI] 0.77 to 1.73; p=0.486). Twice as many patients attending the clinic for diagnostic testing were diagnosed with AF or related arrhythmias compared with high-risk participants in the control arm (9.41% vs. 4.93%, respectively) (OR 2.23; 95% CI 1.31 to 3.73; p=0.003).

The findings indicate that the risk prediction algorithm might be a valuable tool for use in the primary-care setting to identify patients at a high risk of undiagnosed AF who could benefit from diagnostic testing – further evaluation of the cost-effectiveness of the intervention is warranted.


Latest Science - On Demand.


1. Hill NR, et al. PLoS One. 2019;14:e0224582.