In initial analyses, we obtained age- and smoking-adjusted rate ratios(RRs) of ARM by category of BMI in proportional hazards regression models adjusting for age, cigarette smoking (never, past, current <20 cigarettes per day, current ≥20 cigarettes per day), and, because subjects were participants in a randomized trial, randomized aspirin and beta carotene assignments. In these analyses, we allowed BMI to vary over time as a time-varying covariate in the proportional hazards models, using the nearest past BMI measurement available for each participant. In additional models, we further adjusted for height in categories of less than 170 cm, 171 to 178 cm, 179 to 183 cm, and 184 cm or more, as well as other potential risk factors including alcohol consumption and vitamin supplement use. To investigate the possibility of residual confounding, we also fit models in which we adjusted for pack-years of cigarette smoking as described previously,17 as well as alcohol consumption (1 or more drinks per day, 1 to 6 drinks per week, 1 to 3 drinks per month, and rarely or never), vitamin E (never, past, and current) and vitamin C (never, past, and current) supplement use, and mean daily servings of vegetables (sum of servings of broccoli, brussels sprouts, carrots, spinach, dark green lettuce, yellow squash, yams or sweet potatoes, tomato juice, and tomatoes), fruits (sum of servings of orange juice, cantaloupe, peaches, apricots, and nectarines), and cold breakfast cereal. Dietary information was obtained by means of a brief food frequency questionnaire.31 Finally, we explored whether relationships of BMI with ARM were different in younger vs older men by fitting separate proportional hazards models for those who were aged 75 years and older and those who were younger than 75 years. Similarly, we investigated whether the effect of BMI appeared to differ according to smoking status by fitting separate models for never, past, and current smokers.