Migraine Forecasting Model Reaches 91% Precision Using App-Based Headache Data

A machine-learning model using patient-reported data from 53,065 Nerivio app users predicted next-day reported migraine outcomes with 91.2% precision, according to a study published in Neurology Open Access. The findings indicated that the strongest predictive signals came from patients’ headache history over the preceding 30 days, rather than symptoms reported shortly before an attack.¹

For clinicians, the results suggest that longitudinal headache records may provide information beyond a single day’s premonitory symptoms. However, the study assessed forecasting performance—not whether receiving a forecast improves patient outcomes or changes migraine management. The app feature, called Your Day Ahead, provides next-day likelihood information but does not diagnose migraine or recommend changes to prescribed treatment.

What this analysis suggests, in the largest dataset reported to date, is that the strongest predictive signal isn't in that narrow pre-attack window, instead, a patient's own recorded pattern of headache severity over the preceding month turns out to be the most informative signal in the model," Chia-Chun Chiang, MD, Associate Professor of Neurology, Headache Specialist of Mayo Clinic in Rochester, Minnesota said in a statement. "That's a meaningful shift in how we think about forecasting migraine risk, and it matters for patients: consistent, longitudinal patient-reported data isn't just a record of what's happened — it may be a window into what's likely to come next.”

Study Overview

The researchers analyzed 770,473 daily reports from 53,065 individuals who used the Nerivio app between January 2020 and July 2025. Nerivio is a prescription remote electrical neuromodulation device used for acute and preventive migraine treatment. The study drew on patient-reported app data, including diary entries and pretreatment questionnaires, as well as demographic information and location-based weather conditions.²

Seven machine-learning algorithms were evaluated to forecast the following day’s reported migraine outcome. The investigators assessed model performance using precision, accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). The analysis was an observational assessment of app records, rather than a clinical trial evaluating forecast-guided care.

Model Performance and Predictive Features

The customized XGBoost classifier achieved 91.2% precision, 81.0% accuracy, 80.0% sensitivity, 83.0% specificity, and an AUC of 0.893, according to the researchers.¹ Although other algorithms performed better on certain individual measures, the XGBoost model demonstrated the strongest overall performance across the reported predictive metrics.²

Precision indicates the proportion of days identified by the model as likely positive days that corresponded to positive outcomes in the evaluated data. Thus, a precision of 91.2% does not mean that the model correctly classified 91.2% of all days; overall accuracy was 81.0%.

The rolling mean of headache severity over the preceding 30 days was the model’s most informative individual feature, accounting for 37.6% of the reported feature contribution. When grouped by assessment window, features from the preceding 30 days accounted for 56.3% of the contribution. By comparison, the prodromal symptom feature group accounted for 11.1%.

These findings describe how the model used the recorded variables and do not establish that the identified features cause subsequent migraine outcomes. They also do not indicate that prodromal symptoms lack clinical value in individual patients.

The company announcement reported no statistically significant performance differences between female and male users (P = .42) or between adults and adolescents (P = .28). However, these comparisons do not establish equivalent performance across all patient subgroups. In addition, the forecasted outcome included mild headache pain as well as moderate and severe pain, which is an important consideration when interpreting the model as a tool for predicting migraine attacks.¹

Clinical Context and Interpretation

A next-day forecast could potentially help patients anticipate symptom recurrence and plan daily activities. However, the study did not evaluate whether providing forecast information affected patient behavior, treatment adherence, medication use, disability, or quality of life. Whether the forecast reduces anticipatory anxiety or improves migraine-related outcomes remains untested.¹

The analysis is notable for its large number of users and daily reports, but the study population was drawn from individuals using a specific therapeutic device. As a result, the cohort may overrepresent people who are engaged in migraine treatment and symptom tracking, including those with more frequent symptoms. Reliance on app-reported outcomes also presents a consideration when interpreting the findings.

The investigators separated users across the training, validation, and test datasets. Nevertheless, external validation in an independent clinical population is needed to assess whether the model performs consistently across different patient groups, care settings, and patterns of migraine reporting.

Future prospective studies should evaluate the forecast in broader migraine populations and determine whether providing next-day risk estimates improves clinical or patient-reported outcomes without causing unnecessary concern or changes in medication use. Until such findings are available, the results support further investigation of longitudinal headache data as a forecasting tool rather than a basis for changing treatment decisions.

REFERENCES
1. Theranica's AI Achieves 91% Precision on Next-Day Migraine Likelihood in Largest Study of Its Kind. Theranica. News Release. September 23, 2026. Accessed September 23, 2026. https://www.prnewswire.com/news-releases/theranicas-ai-achieves-91-precision-on-next-day-migraine-likelihood-in-largest-study-of-its-kind-302887129.html
2. Rabany L, Markina A, Cowan R, et al. Machine Learning Assessment of Next-Day Migraine Likelihood Using Data From 53,000 App Users Living With Migraine. Neurl. Open Access. 2026:2 (3) e000144 doi:10.1212/WN9.0000000000000144