Clinical performance and applicability of prediction models using wearable‐derived physiologic signals to anticipate migraine or headache attacks: A systematic review

OBJECTIVE: This study evaluated the performance, risk of bias, and clinical applicability of prediction models using wearable-derived or ambulatory physiologic signals to anticipate short-term migraine or headache outcomes in adults. BACKGROUND: A migraine or headache forecast can support self-management only when it is accurate, calibrated, actionable, and associated with an acceptable false-alarm burden. Wearable devices estimate autonomic, activity, and sleep-related features from surrogate signals, but their clinical readiness has not been systematically evaluated. METHODS: We reviewed prediction-model studies using objective wearable or ambulatory physiologic signals to anticipate migraine or headache outcomes in adults. PubMed/MEDLINE, Scopus, and Web of Science Core Collection were searched from inception to May 18, 2026. Reporting was structured according to Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis-Systematic Reviews and Meta-Analyses, with Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 used for study selection. We used Prediction Model Risk of Bias Assessment Tool/Prediction Model Risk of Bias Assessment Tool +artificial intelligence-compatible criteria and a task-specific narrative synthesis. RESULTS: Eight studies were included. Tasks were near-onset detection (20-40 min), night-to-next-day forecasting, and multi-day forecasting. Among comparable models, area under the receiver operating characteristic curve ranged from 0.62 to 0.84. In the largest study (146 adults; 21,550 prediction days), area under the receiver operating characteristic curve was 0.84 (95% confidence interval, 0.82-0.85) for next-day moderate-to-severe headache and 0.76 (95% confidence interval, 0.74-0.77) for a 3-day horizon. Five studies had high analysis-domain risk of bias and three had some concerns. No study reported external validation; calibration was reported once. CONCLUSION: Wearable-based prediction models remain early proof-of-concept studies, not clinically validated alert systems. Their reported discrimination may partly reflect between-person differences in headache frequency rather than within-person forecasting. External validation, calibration, threshold-specific alert burden, and evidence that acting on forecasts improves outcomes are needed before clinical use.