Additional known and hypothesized risk factors for fracture and diabetes were included in models as possible confounders or mediators for the association of diabetes and fractures. Cox proportional hazards regression models determined the independent effect of DM or IFG on risk of incident fracture, while adjusting for age, race, sex, site (Memphis or Pittsburgh), total hip BMD, bone volume, falls, lean and fat mass, visceral fat, fasting insulin, serum creatinine, smoking, drinking, Health ABC performance battery categorical score (0-12), weekly physical activity from walking and exercise, gain or loss of 2.25 kg or more in past 12 months, adult weight changes, ankle-brachial index, and bone-active medication use. Diabetes-related complications (history of cardiovascular disease, transient ischemic attack or stroke, kidney disease, retinopathy or retinal disease, cataracts, and nerve function [pain or numbness in legs or feet, monofilament detection, vibration threshold, and nerve conduction amplitude and velocity]) were entered in the model to assess their role as potential mediators. Date of first fracture or last follow-up if no fracture occurred was used to censor data. Models were built progressively by entering variables in the following order: demographics, hip BMD, body composition factors, other potential confounders, and diabetes-related complications. Sex, race, clinic site, and diabetes variables were forced into the model, and other variables were removed at P>.10. Given the low number of fractures, multivariate analyses for fracture risk including only diabetic participants were not performed. Data were analyzed using SPSS statistical software (SPSS Inc, Chicago, Ill).