You probably know the drill. The dull warning behind one eye, the overhead light suddenly too bright, the quiet knowledge that the next hours may be lost.
A new machine learning model has forecasted next-day migraine risk with 91.2% precision in a study published in Neurology Open Access.
When the tool told users a migraine was coming, it was right about nine times in ten.
Researchers trained the model on 770,473 daily symptom reports from 53,065 real-world users of a migraine wearable called Nerivio, collected between January 2020 and July 2025.
Also read: The same workout may affect men’s and women’s livers differently
That is the largest prospective dataset of its kind in migraine research.
The signal driving the forecast was not what the clinic has traditionally watched for.
Recent headache patterns outweighed classic warning signs
For decades, clinicians have watched for prodromal symptoms, meaning the vague cues like yawning, mood changes or food cravings that can precede an attack.
In this model, those day-of signals carried little weight.
Also read: Why fruit and vegetables are placed at the entrance of supermarkets
The strongest predictor was a person's own headache pattern over the previous 30 days, which accounted for about 56% of the model's predictive power.
Prodromal symptoms contributed only about 11%.
Precision and sensitivity do not mean the same thing
Precision of 91.2% means the model rarely cried wolf. When it forecast a migraine, that forecast held up most of the time.
Sensitivity was lower, at 80%. The model missed roughly one in five actual migraines, so a quiet forecast on your phone does not guarantee a quiet day.
Also read: Want to live longer? These foods may be worth eating more of
The team used a method called XGBoost, which combines many small decision rules into one stronger prediction.
They fed it diary entries, pre-treatment questionnaires, patient details and local weather patterns.
Overall accuracy came to 81%. The model also scored 0.893 on a standard measure of how well it separated migraine days from non-migraine days. A perfect score would be 1.0.
Performance did not vary significantly between women and men, or between adults and adolescents.
Also read: A change in how you speak could be an early warning sign of dementia, study finds
What a daily forecast could change for patients
Migraine affects roughly 1.2 billion people worldwide, according to the Global Burden of Disease analysis of data through 2021.
An evening warning could help with sleep timing, hydration, meal planning or stepping back from a stressful meeting the next morning.
Chia-Chun Chiang, a headache specialist at the Mayo Clinic in Rochester, Minnesota, called it "a meaningful shift in how we think about forecasting migraine risk," according to PR Newswire.
The study has limits. It drew only from people who already use a migraine wearable.
Also read: Your Halloween face paint could cause a skin reaction days later: Here's how to protect yourself
That group tends to have frequent attacks and to be actively engaged in treatment.
The authors say the next step is external validation in a different patient group. Only then could the forecast be used in everyday clinical care.
This article is made and published by Mie Hermansen, who may have used AI in the preparation.