Theranica’s machine learning model forecast next-day migraine with 91.2% precision. It was trained on app data from 53,065 users of the Nerivio wearable. The peer-reviewed results appear in Neurology Open Access.
Which signal did the work is the finding that matters. Prodromal symptoms, the aura, mood shifts and yawning that clinicians have tracked for decades, contributed roughly 11% of the model’s predictive power. The stronger input was the rolling average of a patient’s own headache severity across the preceding 30 days.
Chia-Chun Chiang, a headache specialist at the Mayo Clinic, said the largest dataset reported to date suggests the most informative signal sits outside the narrow pre-attack window. She called that a meaningful shift in how the field thinks about forecasting risk.
Nerivio is an FDA-cleared neuromodulation device worn on the upper arm. Its smartphone app delivers 45-minute electrical pulse treatments and separately collects symptom, timing, demographic and weather data. The Your Day Ahead feature turns that history into a next-day risk estimate so patients can plan around an attack.
The study reports an area under the curve of 0.893. Theranica sponsored the work, and the feature already ships in the app.
Two caveats belong beside the number. The output is a planning aid that neither diagnoses migraine nor changes prescribed treatment. And sponsor-funded validation on a company’s own user base is a different bar from an independent trial. Even so, it points to a useful pattern for chronic disease AI: the durable signal is often the patient’s own longitudinal record.
