Artificial intelligence can predict migraines based on biofeedback data

A new study published in The Journal of Headache and Pain shows that artificial intelligence (AI) can be used to predict migraine attacks several days in advance. The study is based on data from the BioCer study and draws on the biofeedback data from that study. The research is still in its early stages, but this is an important step toward more personalized and preventive migraine management. 

In the BioCer study, participants underwent daily biofeedback sessions over a three-month period while recording their headaches in a digital diary. The sensors measured, among other things:

• Muscle tension in the neck
• HRV (heart rate variability)
• Finger temperature

In total, over 26,000 days of data were collected, of which approximately 21,000–25,000 days were used in the various analyses, providing a solid foundation for understanding patterns preceding migraine attacks.

 What did the researchers find?

 By analyzing this data, the researchers found that:

• It is possible to predict a migraine one day in advance with a fairly high degree of accuracy
• It is also possible to make predictions about risk up to three days in advance
• The most important indicators are previous headaches (their intensity and duration) and changes in the body’s physiology

The body begins to change before a migraine attack—and these changes can be detected.

Findings from the study

The researchers based their analysis on two main types of data. The first consisted of information the participants themselves entered into the app, such as whether they had a headache, how severe it was, and how long it lasted. The other was physiological measurements from the biofeedback sessions, such as HRV, finger temperature, and neck muscle tension. By combining these data sources on a day-to-day basis, they obtained a more comprehensive picture of the condition before a seizure occurs.

A key objective of the study was to determine which type of analysis actually works best. The researchers tested both simpler methods that examine individual days in isolation and more advanced models that analyze trends over several days. It became clear that the latter yielded far better results. When the model had access to several days of history—up to about a week—it was able to detect small, gradual changes in the body that often precede a migraine attack. This resulted in significantly greater accuracy than the models that only assessed single points in time.

When it came to which signals were actually most important, three factors in particular stood out: how intense the headache had recently been, how long it had lasted, and various measures related to heart rate. This suggests that both the user’s own subjective experience and more “hidden” physiological changes play a role in the period leading up to an attack. At the same time, no single variable could explain everything—it is the combination and the progression over time that are decisive.

Although the results are promising, the study also clearly shows that this is a complex problem. The models were generally better at predicting days without migraines than days with attacks. This means that they are better at identifying when the body is stable than when an attack is actually on its way. This is likely due to the fact that migraines vary greatly from person to person, and that not all relevant factors—such as stress, sleep, or hormonal changes—are fully captured in this dataset.

Future work will therefore focus on improving both the data set and the models. The researchers particularly emphasize the need for more comprehensive and continuous data, so that important triggers and early symptoms are not overlooked. At the same time, the goal is to make the models more personalized, so that they can better learn the patterns specific to each individual user.

The study shows that the data from the BioCer biofeedback study not only provide an overview—they can also be used to look ahead. Although the technology is still under development, this is an important step toward more personalized and preventive migraine treatment.