Information theory and headache triggers

Objectives This secondary analysis evaluated the information content exhibited by various measurement strategies of commonly perceived causes, or “triggers,” of headache attacks. Background When evaluating triggers of primary headache attacks, the variation observed in trigger candidates must be measured to compare against the covariation in headache activity. Given the numerous strategies that could be used to measure and record headache trigger variables, it is useful to consider the information contained in these measurements. Methods Using previously collected data from cohort and cross‐sectional studies, online data sources, and simulations, the Shannon information entropy exhibited by many common triggers was evaluated by analyzing available time‐series or theoretical distributions of headache triggers. The degree of information, reported in bits, was compared across trigger variables, measurement strategies, and settings. Results A wide range of information content was observed across headache triggers. Due to lack of variation, there was little information, near 0.00 bits, in triggers like red wine and air conditioning. Most headache triggers exhibited more information when measured using an ordinal scale of presence/degree (e.g., absent, mild, moderate, severe) versus a present/absent binary coding. For example, the trigger “joy” exhibited 0.03 bits when assessed using binary coding but 1.81 bits when coded using an ordinal scale. More information was observed with the use of count data (0.86 to 1.75 bits), Likert rating scales (1.50 to 2.76 bits), validated questionnaires (3.57 to 6.04 bits), weather variables (0.10 to 8.00 bits), and ambulatory monitoring devices (9.19 to 12.61 bits). Conclusions Although commonly used, all binary‐coded measurements contain ≤1.00 bit of information. Low levels of information in trigger variables make associations with headache activity more difficult to detect. Assessments that balance information‐rich measurements with reasonable participant burden using efficient formats (e.g., Likert scales) are recommended to enhance the evaluation of the association with headache activity.