Prediction of Undiagnosed AF using a ML Algorithm: PULsE-AI Trial
Research type
Research Study
Full title
A randomised controlled trial for the identification of undiagnosed atrial fibrillation patients using a machine learning risk prediction algorithm and diagnostic test
IRAS ID
252934
Contact name
Alexander Cohen
Contact email
Sponsor organisation
Bristol-Myers Squibb Pharmaceuticals Ltd.
Clinicaltrials.gov Identifier
ClinicalTrials.gov, In progress
Duration of Study in the UK
0 years, 10 months, 25 days
Research summary
Summary of Research
The purpose of this study is to investigate the accuracy of a risk prediction algorithm (similar to a computer programme) to detect patients with atrial fibrillation (AF). AF is a common disorder that disrupts the normal beating rhythm of your heart. It can be difficult to diagnose, meaning more than 400,000 people in England may be living with undiagnosed AF. Patients with AF are at higher risk of developing conditions such as stroke and heart failure; therefore, early diagnosis and management of AF is important.
During the study, patients will be randomly allocated to one of two groups; intervention or control. The intervention group will use the algorithm, and the control arm will continue to receive routine care. For patients in the intervention arm, non-identifiable patient data will be used to predict patients’ risk of developing AF, by use of the risk prediction algorithm. If patients are identified as high risk, they will be invited to a research clinic to be screened for the disease using an electrocardiogram (ECG). At this stage, patients may be diagnosed with AF (positive ECG result); or invited for further screening using an at-home heart monitor, which patients will use twice daily for two weeks.
From the steps outlined above, patients in the intervention group will have one of three outcomes: normal; AF diagnosed; or undetermined. If undetermined, patients may undergo further investigation as determined by the healthcare professional.
In the control arm, patients will only be diagnosed through clinical practice, and will be identified by comparison of medical records at the start and the end of the study.
To find out whether it is more effective to use the risk prediction algorithm rather than usual care, the number of patients diagnosed with AF will be compared in the intervention and control arms.
The purpose of this study was to investigate the ability of a risk prediction algorithm (similar to a computer programme) to find patients at higher risk of atrial fibrillation (AF). AF is a common disorder that disrupts the normal beating rhythm of the heart. It can be difficult to diagnose, meaning more than 400,000 people in England may be living with undiagnosed AF. Patients with AF are at higher risk of developing conditions such as stroke and heart failure; therefore, early diagnosis and management of AF is important. Patients from six general practices in England who participated in the study were assigned to intervention and control groups. Patients assigned to the intervention group and predicted to be at high risk of undiagnosed AF (based on data included in their medical records and the risk prediction algorithm) were invited to undergo testing for AF. Those who accepted the invitation for testing received an electrocardiogram (ECG – the test for AF) and – if they had access to a compatible smartphone or tablet – underwent two-weeks of home-based ECG monitoring with a portable ECG device. The ability of the risk prediction algorithm (in combination with ECG testing) to find patients with undiagnosed AF was determined by comparing the percentage of patients diagnosed with AF in the intervention group, with those in the control group – who received routine care. There was little difference in the percentage of patients at high risk of undiagnosed AF who were diagnosed with AF during the study (5.63% in the intervention group and 4.93% in the control group). However, only 28% of patients in the intervention group who were at higher risk of undiagnosed AF accepted the invitation to participate and therefore underwent ECG testing. Of this smaller intervention group who underwent testing, 9.41% were diagnosed with AF compared to 4.93% of patients in the control group, indicating that patients who received the intervention (risk prediction algorithm + ECG testing) were twice as likely to have their AF identified than those who did not. This risk prediction algorithm for AF could be a valuable tool to select patients at high risk of undiagnosed AF who may benefit from ECG testing.
REC name
Wales REC 5
REC reference
19/WA/0069
Date of REC Opinion
28 Feb 2019
REC opinion
Favourable Opinion