Multivariable analyses were done using accelerated failure time models23 to determine which characteristics were predictive of remaining successful. The accelerated failure time model assumes that risk factors act to shorten or lengthen the time of remaining successful, such that the likelihood that an individual with a given risk factor remains successful until time t is equal to the likelihood that someone without the risk factor remains successful until time kt, where k is the acceleration factor. If k is greater than 1, the individual without the risk factor remains successful longer, and the person with the risk factor is said to have an accelerated failure time. If k is less than 1, the "risk" factor is beneficial to maintaining success, and if k is equivalent to 1, there is no difference in successful survival time between persons with and without the risk factor. The inverse of the acceleration factor k, provides an estimate of the proportion of years remaining successful in those with the given risk factor compared with those without. All models were adjusted for age, race, body mass index, years of education, depression score, and arthritis. Additional risk factors were allowed to enter in stages, with smoking, alcohol use, energy expenditure from physical activity, history of hypertension or diabetes mellitus, and blood pressure considered at stage 1. Once significant stage 1 variables were entered, the blood laboratory measurements were considered. In the final stage, measures of subclinical CVD were evaluated for entry. Skewed variables were log-transformed and tested linearly; results are presented in quintiles or clinical cutpoints to examine potential threshold effects. Multivariable models were generated separately for men and women, and sex interactions were tested for significance in a model including men and women. No significant interactions were found, and results of the stratified models were similar. We retained the model including men and women to estimate median years of successful life with the same set of predictors for both sexes. The median number of successful years was calculated from the final model for an idealized low-risk individual who had the best profile on each risk factor and an idealized high-risk individual who had the worst profile on each risk factor, for a selected age group and sex. The adjustment variables of years of education, body mass index, depression score, arthritis, and race were set to their mean values for the participants who were successful at baseline. Observed median survival was calculated from a life table.