The main outcome variable was death during the study period. The date of death was typically reported to the BCN by family members, employers, or health care professionals. To assess potential confounding variables, we captured data on age, sex, self-reported race, insurance type, glaucoma type (open angle, narrow angle, suspected, neovascular, other, or >1 type), glaucoma surgery (laser trabeculoplasty, iridotomy, trabeculectomy, glaucoma drainage device, and cyclophotocoagulation), oral β-blocker use, and the following comorbid conditions: diabetes mellitus, congestive heart failure, cancer (any type), chronic liver disease, chronic kidney disease, asthma, bradycardia, hyperlipidemia, arterial disease, ischemic heart disease, cerebrovascular disease, osteoporosis, hypotension, atrioventricular block, and depression. eTable 2 provides the International Classification of Diseases, Ninth Revision, Clinical Modification, and Current Procedural Terminology424 billing codes used to identify the type of glaucoma diagnosis, surgical procedures, and comorbid medical conditions. Participant characteristics were summarized for the entire sample using means and standard deviations for continuous variables and frequencies and percentages for categorical variables. Cox regression was used to estimate the hazard of death associated with various glaucoma medications. Using age as the time axis, the Cox model was left truncated at the age of first recording of glaucoma in the database. Participants were followed up until death or were censored at the age of disenrollment or at the end of the study period (December 31, 2007). Initially, univariable models were run to test potential predictors individually. Multivariable models were adjusted for sex, glaucoma surgery, glaucoma type, and chronic medical conditions. Because glaucoma surgery, glaucoma type, chronic medical conditions, and prescription medications purchased could change across time, these variables were entered into the Cox models as time-dependent covariates. These time-dependent covariates were indicator variables, for example, taking the value of 1 when the patient was prescribed a particular medication and the value of 0 when the medication was not prescribed. Multiple medication use at a given time was accommodated using this method because each medication was entered as a separate indicator. Statistical analyses were performed by using a commercially available software program (SAS version 9.1; SAS Institute Inc, Cary, North Carolina). This study was approved by the institutional review boards of the University of Michigan, Ann Arbor, and the BCN.