Enhancing Migraine Trigger Surprisal Predictions: A Bayesian Approach to Establishing Prospective Expectations.

Area of Science:

• Computational neuroscience
• Migraine pathophysiology
• Predictive modeling

Background:

• Surprisal theory links unexpected trigger exposure to migraine onset within 12-24 hours.
• Previous studies used retrospective expectations, limiting real-time prediction.
• Bayesian methods offer dynamic updating of expectations for prospective surprisal calculation.

Purpose of the Study:

• To develop real-time methods for estimating migraine trigger likelihood using surprisal theory.
• To operationalize prospective surprisal estimation under limited personal observation.
• To investigate the impact of prior specification on dynamic surprisal modeling.

Main Methods:

• Prospective daily diary study (N=104) over 28 days, collecting data on stress, sleep, and exercise.
• Application of Bayesian models to estimate daily expectations for each trigger variable.
• Calculation of dynamic surprisal based on predictive distributions and comparison with static empirical values.

Main Results:

• Dynamic Bayesian surprisal estimates differed significantly from retrospective values, especially early in data collection.
• Uninformative priors led to more variable and biased surprisal trajectories compared to empirically informed priors.
• Individual variability in trigger exposure, particularly exercise, was substantial.

Conclusions:

• Real-time migraine prediction using prospective surprisal modeling is feasible.
• Model calibration is sensitive to prior specification, with empirical or individual priors enhancing early accuracy.
• These methods provide a basis for real-time headache forecasting and understanding brain-environment interactions.