Algorithm shown to detect those at high risk of atrial fibrillation

An artificial intelligence (AI)-based machine learning (ML) algorithm developed by the Bristol Myers Squibb-Pfizer Alliance has been shown to help support identification of patients at high risk of atrial fibrillation (AF), as validated in both retrospective datasets and the PULsE-AI trial. 

If implemented in primary care, the algorithm may be a cost-effective method to identify those at highest risk of undiagnosed AF who should undergo further assessment. Results from the trial were announced at ESC Congress 2021, organised by the European Society of Cardiology (ESC).

AF is prevalent in ~3% of the UK population, increases the risk of stroke by five times and is associated with heart failure. Early detection and management is likely to improve outcomes for patients, but detection is challenging because some people with AF may experience minimal or no symptoms at all. As a result, as many as 300,000 people are living with undiagnosed AF in the UK.

The PULsE-AI trial, conducted with 1,880 people aged 30 and over during a 20-month period, was designed to assess the effectiveness and cost-effectiveness of a ML risk prediction algorithm in conjunction with diagnostic testing for the identification of patients at high risk of AF in primary care settings in the UK. Participants at high risk of AF were identified using the ML algorithm, and then split into two groups for the study:

• The intervention group was invited to attend a research clinic for diagnostic testing including a 12-lead ECG; and two weeks of twice-daily monitoring using a KardiaMobile
• The control group had no direct contact with the investigators, but had access to usual clinical practice for diagnosis of AF.

Based upon assumed AF diagnosis rates, it was anticipated that 2.4% of people in the intervention group and 0.7% of people in the control group would be diagnosed with AF over a six-month period. However, following identification through the AI algorithm, 4.97% of the intervention group, and 4.93% of the control group were diagnosed with the condition. The trial did not meet its primary endpoint (which was the number of AF diagnoses in the intervention arm vs. the control arm) possibly attributed to higher than predicted background diagnosis rates brought about by the trial extension owing to the COVID-19 pandemic. 

Nonetheless, in the subgroup analysis looking at the high-risk participants who attended the research clinic for diagnostic testing, the intervention was found to be superior to routine practice where 8.63% AF diagnoses occurred compared to the control group (4.93%). The authors concluded that the algorithm may be an effective tool in narrowing the population at high risk of undiagnosed AF who should undergo diagnostic testing.

Usman Farooqui, executive director - head of medical affairs, Central Eastern Europe, Turkey, Israel & India (CEETII) at Bristol Myers Squibb, said: “We are pleased that these results confirm the effectiveness of the algorithm in helping to identify people at risk of AF in both the control and intervention groups.

“We aim to share the PULsE-AI algorithm as a tool that could help health providers in the near future. When implemented in a clinical setting, it has the potential to identify more people who need AF management so they can receive the care that they need, when they need it.”

Further results of the health economic impact assessment will be shared later in the year.