OBJECTIVES: The PULsE-AI trial sought to determine the effectiveness and cost-effectiveness of a machine learning risk prediction algorithm in conjunction with diagnostic testing for identification of undiagnosed atrial fibrillation (AF) in primary care.

METHODS: Eligible participants were randomised into intervention and control arms and risk assessed using the AF risk prediction algorithm. Intervention arm participants – identified at high risk of undiagnosed AF – were invited for a 12-lead electrocardiogram (ECG) and two weeks of home-based ECG monitoring. Intervention arm participants who did not accept the invitation to participate, and control arm participants were managed routinely. Analysis considered the costs associated with intervention implementation (including administrative, clinical contact and device costs) and lifetime treatment and events (stroke, major bleed, myocardial infarction, intracranial haemorrhage) in patients diagnosed with AF compared with lifetime cost of events in undiagnosed patients who did not receive the intervention. Estimated treatment and event costs, and quality-adjusted life years (QALY) were based on the mean of all direct oral anticoagulant therapies. Incremental net monetary benefit (NMB) was calculated based on a willingness to pay threshold of £30,000.

RESULTS: Estimated cost per diagnosis in the intervention arm was £947. Despite lower incidence of events in diagnosed patients, estimated lifetime costs were higher compared with no diagnosis, driven by treatment acquisition costs (total intervention and lifetime costs were estimated at £22,943 for a diagnosed patient vs. £12,956 in a patient without diagnosis). However, after accounting for QALY gains (+1.40 years vs. undiagnosed), diagnosis of AF as a result of the trial intervention resulted in an incremental NMB of £31,012 compared with patients remaining undiagnosed.

CONCLUSIONS: This health economic impact assessment demonstrates that diagnosis of AF in previously undiagnosed patients via application of this machine learning AF risk prediction algorithm combined with diagnostic testing is cost-effective in a UK primary care setting.