Forecasting migraine with time-series machine learning from mobile health data.

Area of Science:

• Neurology
• Data Science
• Digital Health

Background:

• Machine learning (ML) models complex patterns in high-dimensional datasets for predicting health events.
• Mobile health (mHealth) data offers a valuable resource for real-world health monitoring and prediction.
• Forecasting migraine attacks is crucial for effective patient management and treatment.

Purpose of the Study:

• To forecast headache days in episodic migraine patients using ML and mHealth data.
• To evaluate the performance of different ML architectures, including standard, foundation, and time-series models.
• To identify key features predictive of migraine onset.

Main Methods:

• Analysis of data from the BioCer clinical trial (NCT05616741) involving app-based biofeedback for episodic migraine.
• Collection of daily wearable data (muscle tension, heart rate variability, skin temperature) and headache diary entries.
• Training and optimization of ML models using area under the receiver operating characteristics curve (AUC), with time-series models showing the best performance.

Main Results:

• The best time-series model achieved a test set AUC of 0.84 for next-day headache prediction and 0.76 for three-day prediction.
• Standard ML and foundation models showed lower predictive accuracy, with AUCs ranging from 0.55 to 0.59.
• Key predictors for headache forecasting included headache intensity, duration, and heart rate variability.

Conclusions:

• Time-series ML models demonstrate good accuracy in forecasting moderate-to-severe headaches in episodic migraine.
• The study highlights the potential of mHealth data and advanced ML techniques for migraine prediction.
• Findings suggest that ML-driven forecasting can aid in proactive migraine management.