Ongoing debate concerns how many deaths or disease cases are linked to lifestyle-linked (modifiable) factors such as adiposity, smoking and physical activity. This paper considers mortality attributable to modifiable risks using US cohort data from the NHANES III survey, focussing on mortality in a cohort of adults under age 65 at baseline. A piecewise-exponential regression analysis is adopted, with predictor selection, so acknowledging model uncertainty. Attributable risks are estimated using a Bayesian approach, with risks estimated by gender and population sub-groups (ethnic, income). For the cohort considered, smoking has the highest attributable risk, especially for males.