Methods of calculating deaths attributable to obesity.

Previously reported estimates of deaths attributable to obesity in the United States have been based on a method that only partially adjusts for confounding and does not allow for effect modification. In this study, the authors investigated the possible magnitude and direction of bias in estimating deaths attributable to obesity when such a method is used. Hypothetical examples are based on 1991 US population data and published relative risks. Incomplete adjustment for confounding of the obesity-mortality relation by age and sex led to a 17% overestimation of deaths due to obesity. Additional bias resulted from slight differences between the derivation cohort and the target population. For example, a difference of three percentage points in the proportion of people 80 years of age or older led to a 42% overestimation of deaths due to obesity. In addition, these estimates appear to be sensitive to minor differences in relative risks between a derivation cohort and the target population. A difference of 0.20 in relative risks almost doubled the number of deaths (97% overestimation). Estimates of deaths attributable to obesity can be biased if confounding and effect modification are not properly taken into account or if the relative risks are not estimated accurately.