Time-series features of headache: individual distributions, patterns, and predictability of pain

This preliminary study employed "within-patient" time-series methodology to examine headache symptom distributions (peak, sum) to assess predictability of day-to-day headache based on overall pattern, and to better characterize distributional properties of headache as they relate to study design and statistics in future research.

Headache symptoms for a given patient may vary widely from one day to the next, ranging from days when the headache sufferer is asymptomatic to days when he or she is functionally disabled with severe headache. Although day-to-day variations are well recognized and appreciated clinically, headache is seldom studied using daily (eg, time-series) methodology that can elucidate potentially important individual and temporal variations. Instead, most headache research has relied upon cross-sectional designs aimed at examining group rather than individual effects which may serve to mask important individual differences. Additional within-patient time-series studies are needed to help identify symptom patterns, clinically meaningful patient subgroups, and relationships among precipitants, interventions, and outcomes. Such research would be aided by delineation of the distributional and temporal properties of key scales (eg, daily headache peak, sum). The present research provides such information derived from clinical samples of migraine and tension-type headache sufferers, laying a foundation for future time-series research.

Twenty-five migraine and 24 tension-type headache sufferers prospectively recorded daily headache activity for 1 month. Individual headache distributions were generated and examined individually and by diagnostic group. Predictability for headache was analyzed case by case and summarized by diagnostic group.

The study determined the degree to which individual patients' headache activity distributions deviate from normality (skewness, kurtosis). For migraine and tension-type headache, individual patients' headache patterns were distinctly bimodal in nature. Notably, headache patterns with the lowest frequency were the most bimodal, and as frequency increased, the distributions tended to more closely approximate normal. Patterns were also detailed according to measures of predictability for headache (trends, autocorrelation). Generally, headache days tended to cluster together for both tension-type and migraine (positive autocorrelation) with headache on day 1 being a good predictor of headache on day 2.

To the authors' knowledge, this is the first study to utilize time-series methodology to characterize individual patients' headache distributions and temporal patterns and to empirically address predictability in this manner. The bimodal distributions noted among less frequent headache patterns would suggest that basic assumptions underlying the use of inferential statistics may be violated when examining intra-individual relationships. Time-series research promises to yield unique insights into patterns, precipitants, and impact of headache disorders, but future research must address both the large degree of individual differences in headache and the account for the unique types of statistical distributions among individual headache sufferers.