An artificial-intelligence algorithm pinpoints five blood components that point to someone’s risk of atrial fibrillation.
Summary
A blood test could warn of atrial fibrillation, a silent heart condition that can lead to devastating strokes if left untreated. Researchers created an AI algorithm to analyse thousands of biomolecules in the blood from nearly 47,000 UK Biobank participants. The algorithm pinpointed five blood components that together signal atrial-fibrillation risk slightly better than the method currently used in clinics.
Elevated levels of five blood components can warn of atrial fibrillation (AF), a silent condition that can lead to severe stroke if left untreated.
The discovery was made with the help of an artificial-intelligence (AI) algorithm that picked the five components out of thousands found in blood from almost 47,000 UK Biobank participants. Eventually, blood tests could point clinicians to people with the highest AF risk so that they can receive treatment before experiencing a potentially devastating stroke.
A silent condition
Sometimes, a stroke may be the first manifestation of atrial fibrillation.
Dr Yuichi Shimada, Columbia University, US
AF is an irregular heartbeat that can form blood clots in the heart. If these travel to the brain, they can cause disabling and sometimes fatal strokes. Blood thinners can lower the risk of blood clots and serious stroke, and other medications can regulate the heart rhythm.
But because AF often has no symptoms, around a quarter of AF cases remain undiagnosed. “Sometimes, a stroke may be the first manifestation of AF,” says Yuichi Shimada from Columbia University, US, who has done research on AF independent of the study team.
Doctors can estimate someone’s AF risk by ‘scoring’ factors such as age, weight, blood pressure and medical history. “But [it’s not] common practice to screen for atrial fibrillation based on a risk score, unfortunately,” says Akl Fahed from Harvard University, US, who co-led the study. “It is really more driven by symptoms.”
How to find the protein needle in the haystack
More accurate ways to estimate AF risk are urgently needed. Fahed and his colleagues turned to proteins, biomolecules produced by bodily processes. The amount and type of proteins in blood, for example, tell us a lot about a person’s health.
The problem is that there are thousands of proteins in the blood. Fahed and his colleagues tasked an AI algorithm to comb through the almost 3,000 proteins that had been measured in blood samples from nearly 47,000 UK Biobank participants.
The algorithm picked out five – including ones related to heart stress and inflammation – that together can warn of AF slightly better than the clinical scoring system. “It was quite unexpected for us because the clinical risk factor has been known as the gold standard,” says Fahed’s colleague Min Seo Kim, who also worked on the study.
Peering into patients’ future health
This is the first important step towards more accurate prediction using proteomics.
Dr Yuichi Shimada, Columbia University, US
Shimada says that the impressive size of the study gives it “statistical power – it has very low probability of missing important signals”. This also makes the method more likely to work for people beyond UK Biobank participants.
Although the improvement over the clinical scoring system isn’t huge, says Shimada, “this is the first important step towards more accurate prediction using proteomics”, meaning the large-scale analysis of the body’s proteins.
While a new AF prediction method won’t be available to patients anytime soon, proteomic tests could eventually become routine in clinics. They could let doctors estimate not only their patients’ risk of AF, but of multiple severe and chronic conditions, Fahed says. “We are very proud to have this [method] prepared for that era,” Kim adds.
