AI-ECG Model Predicts Vascular Risk in Migraine Patients

MINNEAPOLIS — Artificial intelligence (AI) may help identify patients with migraine at an increased risk for atrial fibrillation (AF) and other vascular complications, new research showed.

Results of a retrospective study that evaluated more than 29,000 patients who had migraine with and without aura showed the AI prediction model output identified patients at higher risk for paroxysmal AF as well as a composite outcome of vascular events that included acute myocardial infarction (MI), acute ischemic stroke (AIS), and venous thromboembolism using patients’ ECG data.

The study builds on previous research from the same group showing that migraine with aura is associated with an increased risk of developing stroke and adverse vascular events, study investigator Chia-Chun Chiang, MD, associate professor of neurology at the Mayo Clinic in Rochester, Minnesota, said in a presentation at the American Headache Society (AHS) Annual Meeting 2025, as reported to Medscape Medical News.

“Our studies showed that ECG, a widely available test, coupled with AI algorithms, has the potential to be used in clinical practice to screen and identify migraine patients at risk for adverse vascular events,” Chiang said.

Vascular Event Prediction

The researchers evaluated 29,928 patients with migraine who received one or more standard 12-lead ECGs between 2000 and 2020, which included 13,250 patients with aura and 16,678 without aura. Patients had a mean age of 44.3 years and a median follow-up of 54 months.

Researchers input data from each patient’s index ECG into an AI model to assess the risk of developing AF. They also calculated each patient’s delta age — defined as the AI-predicted age based on ECG data minus the individual’s actual chronological age. Higher delta age values have previously been associated with endothelial dysfunction and increased cardiovascular mortality, the researchers noted.

The study’s primary outcome was the incidence of acute MI, AIS, venous thromboembolism, or death as a composite outcome, with a separate outcome of new-onset AF.

Overall, 4662 patients (15.6%) developed adverse events, and new-onset AF occurred in 1384 patients. Using a multivariate Cox regression model that adjusted for factors such as age, sex, aura, and vascular risk factors, researchers found every 10% increase in output of the AI model (hazard ratio [HR], 1.15; 95% CI, 1.12-1.18) and a 10-year increase in delta age (HR, 1.16; 95% CI, 1.12-1.21) was associated with a higher likelihood of developing adverse vascular events.

There was also a higher likelihood of developing new-onset AF with a higher AI model output (HR, 1.31; 95% CI, 1.26-1.37), and among those with new-onset AF, there was a higher risk for adverse vascular events (HR, 2.43; 95% CI, 2.17-2.73).

Other risk factors identified by the AI-ECG AF model output included coronary artery disease (HR, 2.02; 95% CI, 1.79-2.27; P < .001), congestive heart failure (HR, 1.89; 95% CI, 1.53-2.34; P < .001), diabetes mellitus (HR, 1.67; 95% CI, 1.50-1.86; P < .001), hypertension (HR, 1.42; 95% CI, 1.32-1.52; P < .001), tobacco use (HR, 1.38; 95% CI, 1.27-1.51; P < .001), male sex (HR, 1.19; 95% CI, 1.11-1.27; P < .001), and migraine with aura (HR, 1.12; 95% CI, 1.06-1.18; P < .001).

The researchers used Contal and O’Quigley’s method to determine the value at which patients would be categorized into groups at high and low risk for adverse vascular events. They found that an AI model output of 1% or more (HR, 1.45; 95% CI, 1.36-1.54) and a delta age of 1 year or more (HR, 1.09; 95% CI, 1.02-1.16) were the optimal cutoffs to categorize patients into high-risk and low-risk groups.

Chiang noted her group is validating the study’s results and also evaluating factors such as echocardiography results and migraine characteristics that could aid in risk stratification and prediction in patients with migraine.

“The AI models could potentially be used in clinical practice as a screening tool to identify patients at risk, helping clinicians with risk stratification, and facilitate early prevention of adverse vascular events,” Chiang said.

Innovative Research

The study’s large sample size, validated outcomes, and risk adjustments lend credibility to its results, Adriana Rodriguez-Barrath, MD, a neuro-ophthalmologist with University of Iowa Health Care, Iowa City, Iowa, told Medscape Medical News.

“The study is a meaningful step forward in our understanding of migraine as a systemic, rather than solely neurologic, condition, particularly by highlighting its association with elevated risks of AF and stroke,” she explained. “What sets this study apart is its innovative use of AI-ECG models to estimate two novel cardiovascular biomarkers.”

The AI model estimating the probability of AF and delta age “offers noninvasive, scalable, and inexpensive tools that go beyond traditional risk stratification methods,” she said.

Migraine is not an isolated neurological condition, and research has previously shown migraine with and without aura has been linked to conditions such as AF and stroke, Rodriguez-Barrath noted.

“By applying AI-ECG to a migraine population, the study provides compelling evidence that subclinical cardiovascular risk may already be encoded in the ECG signals of patient with migraine, especially those with aura, even before overt cardiovascular disease develops,” she added.

Also commenting on the research, Deborah Reed, MD, associate professor of neurology and director of the Headache Program at University Hospitals and Case Western Reserve University School of Medicine in Cleveland, said the result shows the model accurately predicts and identifies risk factors for adverse events in patients with migraine.

“The strength of this study is the large population that was studied and the development of these tools to predict [AF] and adverse vascular events. I am excited to see where this leads,” Reed told Medscape Medical News.

However, many of the risk factors identified are already known to clinicians, she said.

“Although there was increased vascular risk in patients who were older, male, and had migraine with aura, these are all well-known risk factors for vascular events,” said Reed.

Reed said she discusses vascular risk in patients who have migraine with aura as they develop other risk factors and in patients considering oral contraceptives.

“As we follow our patients over a lifetime, most physicians will discuss risk factors for vascular events, neurologists included,” she explained. “When I discuss this with my patients, I add migraine with aura to the risk factors including family history, diabetes, lifestyle, cholesterol, and exercise.”

AI in Clinical Practice?

Reed suggested that confidently informing patients of their elevated risk for vascular outcomes could help motivate them to prioritize lifestyle changes, despite the challenges involved.

“I think it would be great if we could just get an [ECG] and have a computer tell us our individual risk of [AF] or other vascular events. At the end of the day, there are risk factors you can treat and there are those you cannot. Migraine with aura is quite preventable,” she added.

AI tools have the potential to identify both high-risk groups earlier and offer personalized preventative strategies, Rodriguez-Barrath said, but should be seen as “adjuncts, not replacements, for clinical judgment.”

“All algorithms are trained on population-level data and may miss the nuances of individual patient contexts, such as comorbidities, psychosocial factors, or atypical presentations,” she said.

“As clinicians, we bring human experience, ethical reasoning, and the ability to integrate complex clinical narratives, something no algorithm can replicate. Therefore, AI can enhance our decision-making, but it is our clinical acumen that ensures safe, patient-centered care,” she added.

This study was funded by the American Heart Association. The Mayo Clinic has licensed several AI-ECG algorithms to Anumana, Eko Health, and AliveCor. Several coauthors are inventors of the algorithms and may benefit financially if they are commercialized. Reed and Rodriguez-Barrath reported having no relevant conflicts of interest.