For the analysis of the predictive factors with the appearance of a ‘de novo’ headache, we used binary logistic regression methods by steps backwards, to maximise sensitivity; variables with a univariate association of p<0.200 were included as candidates in the multivariate model.