We used univariate analyses of covariance to compare baseline characteristics of participants and nonparticipants at the follow-up examination, adjusted for age and sex, when applicable. Differences in the distribution of the ER haplotypes were evaluated with Kruskal-Wallis tests. Logistic regression analyses were used to calculate odds ratios with corresponding 95% confidence intervals, which can be interpreted as relative risk. We tested statistical significance for trends in increasing exposure by adding categorical determinants continuously in the model. Because estrogens and ERs are thought to have different effects in men and women, we stratified the analyses by sex, adjusted for age and follow-up time. Analyses were additionally adjusted for the following possible confounders: mean perfusion pressure (calculated as times diastolic blood pressure plus ⅓ times systolic blood pressure minus intraocular pressure), body mass index (calculated as weight in kilograms divided by height in meters squared), diabetes mellitus (defined as use of antidiabetes medication, a random or postload glucose value ≥200 mg/dL [to convert to millimoles per liter, multiply by 0.0555], or both), smoking status (categorized as current, former, or never smoker), ratio of total cholesterol to high-density lipoprotein cholesterol, and intraocular pressure–lowering treatment. All analyses were performed with commercially available software (SPSS for Windows, version 11; SPSS Inc, Chicago, Illinois).