N=1 statistical approaches to examine risk factor profiles of ICHD-3beta classified migraines within individuals.

Objectives: The present investigation compares the
strengths and limitations of two distinct analytic
approaches to understand both incidence and severity patterns
within individuals in relation to daily exposure to a
wide spectrum of risk factors that included emotions,
sleep qualities, environmental and weather, lifestyle, and
diet. The two approaches used were Cox regression to
define incidence and a form of hierarchical linear modeling
to identify severity that is tailored for intensive withinperson
analyses. These two analytic techniques were compared
in terms of which risk factors were identified as
possible ‘‘triggers’’ of migraine onset as opposed to
being associated with severity of a migraine.
Methods: Participants were 750 individuals with migraine
identified by clinician referral or via the internet and registered
to use a novel digital platform (Curelator
HeadacheTM). Participants completed baseline questionnaires
and then entered daily data on headache occurrence
and severity (level of pain), ICHD-3beta migraine criteria,
and exposure to 70 migraine risk factors. Nearly 88% of
the sample was female. Risk factors spanned emotions,
sleep qualities, environmental and weather, lifestyle, diet,
substance use, travel, and three additional triggers selected
by each patient. Cox regression analysis is models the
binomial incidence of migraine attacks (versus no headache).
Hazard ratios from Cox regression tested and computed
strength of associations between occurrence of a
migraine (binomial) and the triggers. These associations
were re-tested for severity of migraine headache using
mixed model trajectory analysis (MMTA), a form of hierarchical
linear modeling analyses severity of migraine headaches
(a continuum). MMTA statistically controlled for
patient-specific time-related trends in pain severity, autocorrelation,
and used statistical tests that generate conservative
estimates for N ¼ 1 analyses.
Results: Overall, a greater number of risk factors were
associated with severity of migraine headaches (MMTA)
than incidence of migraines (Cox regression). However,
Cox regression also detected unique triggers that were
associated only with incidence (not severity) of migraine
attacks. Consistent with past evidence, the profile of risk
factors that were associated with incidence and severity of
migraines varied considerably among patients, demonstrating
that comprehensive clinical research on migraines
requires analytics at the N ¼ 1 level.
Conclusion: Cox regression of migraine incidence and
MMTA of migraine severity each provide unique insights
regarding within-person patterns and correlates of
migraine attacks. The power to detect associations may
be greater for MMTA by virtue of the continuous pain
severity outcome rather than the binomial outcome used
in Cox regression. However, the fact that Cox regression
detected unique risk factors for occurrence of migraine
headaches suggests that different risk factors are associated
with occurrence of migraine attacks versus severity
of migraine pain