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

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Higher surprisal (unexpected trigger exposure) predicts headache onset 12-24 hours later.
• Previous analyses used retrospective expectations, limiting real-time application.
• Bayesian methods can dynamically update expectations for prospective surprisal estimation.

Purpose of the Study:

• To extend surprisal theory for migraine attack risk prediction.
• To develop real-time methods for estimating trigger variable likelihood with limited personal observation.
• To compare prospective dynamic Bayesian surprisal with static retrospective estimates.

Main Methods:

• Prospective daily diary study (N=104) over 28 days, collecting data on stress, sleep, and exercise.
• Applied Bayesian models to estimate daily variable expectations using uninformative and empirical priors.
• Calculated dynamic surprisal based on predictive distributions and compared with static empirical values.

Main Results:

• Dynamic Bayesian surprisal values systematically differed from retrospective estimates, especially early in observation.
• Divergence was greater with uninformative priors but decreased over time.
• Empirically informed priors yielded more stable, lower-bias surprisal trajectories; individual variability was notable.

Conclusions:

• Prospective surprisal modeling is feasible but sensitive to prior specification, particularly with sparse data.
• Using empirical or individually informed priors can enhance early model calibration.
• These methods provide a foundation for real-time headache forecasting and brain-environment interaction modeling.