A meta-analysis of published studies was conducted to identify factors
which explained variation in estimates of migraine prevalence. Twenty
-four population based studies contributed a total of 168 gender and a
ge specific estimates of migraine prevalence. In linear regression ana
lysis, 70.6% of the variation in these prevalence estimates was explai
ned by gender, age (AGE + AGE(2)), a binary variable for case definiti
on, and an interaction term between age and the case definition. Initi
ally, we identified five groups of case definitions among the 24 studi
es. Only the definition of Waters (any 2 of warning, nausea, or unilat
eral pain) was associated with statistically significant differences i
n prevalence estimates among studies; accordingly the other 4 groups w
ere combined. Several other factors were examined as predictors of mig
raine prevalence including the method of selecting the study populatio
n, the source of the population, the response rate and whether diagnos
es were confirmed by a clinical assessment. None of these factors subs
tantially increased explained variance. We conclude that after taking
sociodemographic factors and case definition into account, estimates o
f migraine prevalence are remarkably stable among studies.