AI finds migraine clues beyond the headache itself

There is no single definitive biomarker routinely used to diagnose migraine, leaving doctors dependent largely on headache characteristics and associated symptoms. Researchers publishing in Neurology tested whether machine learning could identify a migraine pattern without receiving information about the headaches themselves.

According to Neurology, the models were trained and tested using data from 43,197 participants in the Trøndelag Health Study. Researchers included 60 variables covering areas such as age, sex, genetics, medication use and broader health information, including constipation and back pain.

The strongest model achieved an area under the curve of about 0.80. The measurement indicates that it distinguished migraine cases from people without headache substantially better than chance, but the result does not make the system a diagnostic tool for routine clinical use.

Withholding headache information was a central part of the experiment. Instead of giving the model the symptoms already used by doctors to diagnose migraine, researchers examined whether combinations of other health characteristics contained a detectable signal associated with the condition.

Several patient profiles emerged

A separate part of the Neurology study examined 12,185 participants with different forms of headache. Machine learning divided them into two large clusters without initially relying on conventional diagnostic categories.

One group contained 1,425 people, more than 90 percent of whom met migraine criteria. The other included 10,760 participants and was dominated by people with other headache presentations, including headaches that only partly met migraine criteria.

Researchers then analyzed the migraine-heavy group more closely. Four profiles emerged: A group consisting entirely of men, another with prominent neck pain, one characterized by greater musculoskeletal pain together with anxiety and depression, and a fourth that more closely resembled established migraine presentations.

Machine-learning-based genetic risk scores also distinguished the profiles more effectively than the conventional polygenic risk scores tested in the research. The findings do not establish four separate diseases. Further studies will have to determine whether these divisions prove useful outside the dataset or influence how patients are diagnosed and treated.

Treatment outcomes vary widely

The differences experienced by migraine patients are already apparent in clinical care. NRK reported that Malin Abelsen Wigdahl, 26, has tried blood-pressure medication, preventive drugs, CGRP inhibitors and Botox without obtaining sufficient relief.

Wigdahl, who normally works in a kindergarten, has been away from work for more than a year because of migraine. According to NRK, she has experienced symptoms since upper secondary school, with more than ten attacks a month during the past two years.

Marit Bergh, 31, has had a different response to treatment. She previously experienced around 20 migraine attacks a month. After beginning Botox and CGRP inhibitors, her attacks became substantially less frequent and she has been able to continue working full time.

The wider effect of severe headache disorders can extend into employment. Citing Norwegian welfare data and Headache Norway, NRK reports that around 800,000 people in Norway have migraine or other headache-related conditions. About 40 percent of those with chronic migraine are fully or partly outside working life.

The new study cannot determine which treatment will work for an individual patient. What it provides is a different way of examining migraine, using health and genetic patterns beyond the headache symptoms traditionally central to diagnosis.

Further research will need to show whether those patterns can be reproduced in other populations and whether they eventually have practical value for patient care.

Sources: NRK, Neurology

This article is made and published by Kenneth Glad, who may have used AI in the preparation