The Problem With Casting a Wide Net
Right now, guidelines say everyone 65 and older should be considered for AFib screening. That sounds simple, but in practice it means testing a huge number of people, most of whom will not have the condition. It is a bit like searching every house on a street for a lost dog when you already know it tends to stay near the park.
But what if you could narrow the search?
That is exactly what researchers set out to do. They wanted to know whether an artificial intelligence (AI) model — one trained to detect subtle patterns in a standard heart-tracing test — could identify which older adults were most likely to actually be diagnosed with AFib during a screening program.
Teaching a Computer to Read the Heart's Signals
The standard 12-lead ECG records the heart's electrical activity from 12 different angles. To the untrained eye, these lines of peaks and dips look almost identical for many patients. But AI can detect tiny differences that predict future rhythm problems — differences too small and complex for any human to reliably spot.
Think of it like audio software that can hear a guitar string going slightly out of tune before your ear ever catches it.
The researchers tested three different risk scores on data from the VITAL-AF trial, a large study run at 16 primary care clinics connected to Massachusetts General Hospital. One score relied on standard clinical information (age, health history). A second used only the AI-reading of a 12-lead ECG. A third combined both approaches into a single tool called CH-AI.
All three tools were better than chance at predicting who would develop AFib over two years, but the AI-based models outperformed the clinical score alone. The combined CH-AI tool had the highest accuracy — correctly ranking patients by true risk in roughly 79 out of 100 cases.
More importantly, the screening program made the biggest difference for people in the top 10 percent of risk according to CH-AI. In that group, the screening caught about 2.3 extra AFib cases per 100 people per year compared to no screening. To put it plainly: for every 43 high-risk people screened for one year, one extra case of undetected AFib was found.
This does not mean AI heart screening is available at your local clinic today.
In the rest of the population — those at average or lower risk — the screening program showed no meaningful benefit over usual care. This is the trade-off the researchers are honest about: focusing on high-risk people finds more cases efficiently, but it also means some lower-risk people with AFib may be missed.
Where This Fits in the Bigger Picture
This study adds important evidence to an ongoing conversation in cardiology about how to make screening programs smarter. Rather than applying the same test to everyone, a targeted approach could save time and resources while catching the cases that matter most. The findings also confirm that AI tools trained on ECG data can do something that simple age-based rules cannot — they can read the subtle electrical "fingerprint" that hints at future rhythm trouble.
If you are 65 or older, especially if you have other heart risk factors like high blood pressure or diabetes, ask your doctor whether an ECG has been done recently and what your overall AFib risk looks like.
The study had real-world strengths: it used data from a large, randomized trial across many primary care practices. But it also had limits. Only patients who already had a 12-lead ECG on file before the study could be included, which may have left out some people. The findings come from one academic health system in the northeastern United States, so results may differ elsewhere.
Researchers now need to test whether acting on these AI-identified risk scores — by watching high-risk patients more closely, for example — actually reduces strokes and other AFib-related harm. The next step is trials designed not just to find AFib earlier, but to prove that earlier detection, guided by AI, leads to better patient outcomes.
