According to these results, we divided the sample into subgroups of homogeneous ages at onset. Distributions of sociodemographic and clinical correlates of the subgroups were compared using chi-square. The odds ratio and confident intervals were estimated through a logistic regression. Sociodemographic covariates in the regressions included age at interview, race, marital status, educational level, household income, level of urbanicity and region. The appropriate statistical weight was employed when mentioned to ensure the data were representative of the population.
