An AI-Enabled Electrocardiogram Model Predicts Poor Vascular Outcomes in People with Migraine

Higher output from a model that estimates the likelihood of atrial fibrillation was associated with increased risk of vascular events. The model also estimates physiologic aging, which was a significant predictor, too.

Migraine, particularly migraine with aura, is associated with a higher risk of cardiovascular and cerebrovascular disease, including heart attack and stroke. Despite this strong connection, doctors have lacked tools to identify which patients are at greatest risk for these poor vascular outcomes.

But now, to tackle this problem, researchers at Mayo Clinic are using an artificial intelligence (AI) model, based on a patient’s electrocardiogram (ECG), that predicts the probability of atrial fibrillation (AF), an irregular heartbeat, and that estimates a patient’s physiologic age. They found that higher output from the model predicted a higher risk of adverse vascular events, as did accelerated biological aging.

Antoinette Maassen van den Brink, who studies neurovascular aspects of migraine at Erasmus MC University Medical Center, Rotterdam, the Netherlands, but was not involved with the new work, said that, in a time when “AI seems to be everywhere,” the researchers’ model stands out.

“When AI is used as a tool to support human thinking, it can be a very powerful tool,” she said. “Here, the authors have a clear hypothesis and show that these AI models can detect trends in cardiovascular risk that we may not otherwise be able to detect.”

The paper was published online ahead of print in Headache on February 20, 2026.

AI uses for a common test
As both a headache and stroke specialist at Mayo Clinic, Rochester, US, Chia-Chun Chiang, who led the new investigation, said she is very interested in the strong but not well-understood link between migraine and stroke, particularly in women who have been diagnosed with migraine with aura.

“We see women with migraine who are younger and without traditional vascular risk factors like hypertension, diabetes, and high cholesterol who end up developing stroke without a clear cause,” Chiang told Migraine Science Collaborative in an interview. “Right now, there is not a good way to identify who these patients are, and it’s a big clinical problem. The question I would like to answer is how to better predict the patients who are at the highest risk.”

Chiang and colleagues at Mayo Clinic have been working for several years on AI models to help predict cardiovascular risk in patients living with migraine, as well as other populations. In 2022, they used AI-enabled ECG in a study of 40,000 people with migraine. They found that the model predicted an increased risk of AF in people who had migraine with aura, compared to those who had migraine without aura (see MSC related news). That work showed that aura was an independent risk factor for subclinical or undiagnosed AF, most notably in patients younger than 55 years of age. It also suggested a potential mechanism of action for this issue – an AF-mediated cardioembolism, or stroke caused by a blood clot.

The success of the model in predicting AF risk raised the question of whether ECG-based models could be used in additional ways. Chiang’s colleagues at  Mayo Clinic’s Department of Cardiovascular Medicine, for example, created an AI model based on ECG to determine physiologic age (sometimes called biological age), a measure of how well the body functions for its chronological age.

“ECG is just a snapshot of your heart rhythm,” she explained. “It’s a very common test for many patients even during a regular health check-up, and we are learning we can predict a lot of things using it…. You can think of AF risk as a reflection of potential cardioembolism and the physiologic age as a marker or reflection of endothelial dysfunction, both of which we know can lead to vascular problems.”

Higher model output means higher risk
For the new study, the researchers asked whether their AI-enabled ECG model could predict which migraine patients would go on to have adverse vascular outcomes. To answer this question, the investigators began with an index ECG, defined as the first ECG received after the age of 18 that had a normal sinus rhythm. Patients who had AF or adverse vascular outcomes before or within three days after the index ECG were excluded from participating. The study would include 29,928 patients (mean age of 44.3 years) diagnosed with migraine who were seen at Mayo Clinic between 2000 and 2020.

The ECG data were then used in the AI model to predict AF risk, as well as delta age (defined as physiologic age estimated by the model minus chronological age). The researchers then asked how well the two AI-ECG algorithms could predict adverse vascular events, which were identified in patients’ electronic health records. The study’s primary outcome was a composite endpoint that included stroke, heart attack, deep vein thrombosis, pulmonary embolism, and all-cause mortality. Newly diagnosed AF was a secondary outcome.

The investigators found that both algorithms predicted adverse vascular outcomes. Those with higher AF prediction model output showed a higher risk on the study’s composite endpoint, with a 15% higher risk of adverse vascular events for every 10% increase in model output. A larger delta age, which reflects accelerated biological aging, produced a similar increase in risk, with a 16% higher risk of adverse vascular outcomes for every 10-year increase in delta age.

“We adjusted for multiple vascular risk factors, including age, and the models still could independently predict the adverse outcomes,” Chiang said. “The other thing we found is the AF algorithm could also predict new development of AF, as we excluded any patients who already had a diagnosis of AF from the study. This provides additional evidence that we can use ECG data to reliably predict AF.”

With regard to new-onset AF, higher model output translated to a 31% higher likelihood of having that event. And those with new-onset AF had roughly 2.5 times the risk of having adverse vascular events.

Finally, for each algorithm, the researchers were also able to identify the best cutoffs to divide patients into high- and low-risk groups for adverse vascular events.

In search of mechanisms
Chiang believes these algorithms could be used as risk stratification tools to help doctors determine which patients with migraine may be at highest risk of developing vascular disease later in life.

While Chiang sees the AI-ECG AF prediction model output and delta age as likely markers of cardioembolism and endothelial dysfunction, respectively, these models cannot definitively say what the mechanism is that underlies the association between migraine and cardiovascular disease. As a result, said Maassen van den Brink, there’s only so much physicians can do right now with the information gleaned from the AI models.

“We have an association here, but we don’t know from these data what the causative factor is or the actual mechanism involved,” she said. “It would be good to have studies that complement these models that show us the mechanisms involved and, even more importantly, how we might intervene with those mechanisms to prevent a stroke or blood clot.”

If doctors today were to use the model to put patients diagnosed with migraine into different vascular risk groups, it’s unclear what actions they should recommend beyond what’s already offered, added Maassen van den Brink.

“We already know patients living with migraine have an increased risk of cardiovascular issues. We already tell those patients to look at their blood pressure, exercise, look at their body weight, stop smoking, and reduce other risk factors,” she said. “But we tell all patients, whether they live with migraine or not, to do that. So while these models are very interesting, until we investigate the mechanisms so we can do something about the risk, it’s tricky to apply these results at the individual patient level.”

Chiang agrees that the mechanism explaining why people with migraine have a higher risk of cardiovascular disease remains uncertain. But she said the study provides an important advance by showing there are opportunities to use AI to identify those patients with migraine who are at risk of vascular events before symptoms appear. Future AI models, she believes, may also be able to offer new clues about potential mechanisms and help identify the best ways to prevent vascular disease.

“There is unlimited potential for using AI in different ways in medicine that can help us predict outcomes,” she said. “And we are investigating other data points, and other modalities like echocardiogram, and migraine characteristics to understand how those features might play into outcomes. They could help us develop a multimodal model to predict vascular outcomes and also help us identify the right preventive measures to help reduce risk for our patients and what actions they should take to avoid vascular events. Those are the next steps.”

Kayt Sukel is a freelance writer based in Kansas.

Image credit: 123RF Stock Photo.

Reference
Higher artificial intelligence-ECG atrial fibrillation prediction model output and estimated physiologic aging predict higher risk of adverse vascular events in patients with migraine.
Yang et al.
Headache. 2026 Feb 20. Online ahead of print.