Information on demographic characteristics and other potentially relevant factors was compared among individuals with and without a history of hypertension. χ2 Tests were used for categorical data, and analysis of variance was used for continuous variables. Multivariate Cox proportional hazards models were used to estimate the association of hypertension to incident all-cause MCI, AMCI, and NAMCI. Because the period between the follow-up assessments in this cohort is relatively short, the time-to-event variable was age at onset of MCI (ie, the age at the assessment at which the research diagnosis was made). Among individuals who did not develop MCI, those who developed dementia were censored at the dementia diagnosis and those who did not develop dementia, who died, or who were lost to follow-up owing to relocation before development of MCI were censored at their last evaluation. Information on covariates was obtained at baseline. We initially adjusted for sex and age, then we adjusted for sex, age, ethnic group, years of education, and APOEε4 genotype in a second model. In a third model, we adjusted for sex, age, ethnic group, years of education, APOEε4 genotype, stroke, diabetes, heart disease, and plasma low-density lipoprotein cholesterol level. The additional covariates in the third model are theoretically in the pathways linking hypertension and MCI. Thus, any attenuation of hazard ratios observed in this model should be interpreted as evidence of mediation and not of confounding. We checked the proportional hazards assumption that the effect of variables of interest is constant in time, by creating time-dependent variables that we then added to the model. When the variable tested added significant information (eg, proportional hazard assumption not satisfied), the model was adjusted for this variable. To explore the association between blood pressure levels and risk of MCI, we finally repeated all analyses using the continuous measure of blood pressure as the independent variable. We estimated the risk of conversion to dementia among persons with MCI using logistic regression. Generalized estimating equations39 were used to examine changes in neuropsychological domains over time, represented by cognitive scores, and compare them between persons with and without hypertension. The dependent variables were the cognitive scores, and the independent variables were hypertension and time (included as a continuous variable). Generalized estimating equation analyses yield coefficient values that represent associations between factor scores and variables included in the model. A significant coefficient for hypertension indicates a difference between 2 groups at baseline or at any subsequent interval. A positive value for the coefficient indicates that the group with a specific variable performed better than the group without that variable. A significant time coefficient would indicate a significant change in a score over the total duration of follow-up. A significant interaction term would indicate a difference in the rate of change in cognitive score between persons with and without hypertension. Data analysis was performed using 2 commercially available software programs (SPSS, version 13.0 [SPSS Inc, Chicago, Illinois], and SAS statistical software, version 9.1 for Windows [SAS Institute Inc, Cary, North Carolina]).