AI finds bodywide migraine clues in more than 43,000 people

A migraine diagnosis still begins with a conversation. A person describes throbbing pain, nausea, vomiting or sensitivity to light and sound; the clinician has no blood test to confirm it. At NTNU, researchers used artificial intelligence to search for another kind of evidence, analyzing health information from more than 43,000 people in Norway's HUNT countywide survey and finding patterns that extended beyond the headache itself.

The model was kept blind to participants' headache symptoms. It worked instead with 60 variables, including age, sex, constipation, medication use, back pain and general health information. According to the study, published in Neurology, AI could distinguish people with migraine from people without headaches. The researchers say that signal suggests migraine may leave a biological signature across the body, not only during an attack.

The team then examined more than 12,000 people with different forms of headache. Its pattern-finding analysis first separated a group of 1,425 people, more than 90% of whom met migraine criteria, from 10,760 people with other or partially matching headache types. Inside the migraine group, it found four profiles: an all-male subgroup; one marked by prominent neck pain; one combining widespread musculoskeletal pain, anxiety and depression; and a group with more classic migraine features, including aura.

Genetic analysis added support to the split. The researchers compared established genetic risk scores with newer AI-based models, and the AI approach was better at distinguishing the migraine groups. Anker Stubberud, a physician and headache researcher at NTNU, says the result strengthens the idea that migraine is not one single disease but a diverse set of biological conditions.

So what changes in practice? Not yet a new test in the doctor's office. The immediate contribution is a route toward an additional tool alongside clinical judgment and a patient's account, potentially helping clinicians make a more consistent diagnosis and eventually match treatment more closely to a person's migraine profile. The findings are based on survey data and an AI model; no clinical diagnostic deployment is reported.