Correcting biases in estimates of mortality attributable to obesity

Objective: To assess whether a recent study that found a relatively small number of excess deaths attributable to obesity may have underestimated by not correcting for statistical biases. Research Methods and Procedures: This prospective cohort study used data from the First National Health and Nutrition Examination Survey Epidemiologic Follow-Up Study. Survival analyses were conducted using 9690 individuals 32 to 87 years of age and 1886 all-cause deaths during a 9.1-year follow-up. Corrections were made for the reputed regression-dilution bias by using the average BMI during the decade before follow-up as predictor. Corrections for the reputed reverse-causation bias were made by excluding participants with a history of serious illness. Attributable fractions were calculated and used to estimate excess deaths. Results: The uncorrected estimate of excess deaths attributable to obesity (BMI >= 30) was 41.9, using 18.5 to 25 kg/m(2) as ideal-weight category. Using average BMI as predictor increased the estimate to 93.3. Correcting for reverse-causation effects increased the estimate further to 131.1 (range, 93.3 to 169.0). The uncorrected hazard ratio, 1.25, was increased to 1.41 by using average BMI as predictor, and then to 2.40 by correcting for reverse causation. Using BMI 21 to 25 kg/m(2) and 23 to 25 kg/m(2) as ideal-weight categories increased the corrected estimates to 144.6 (range, 80.5 to 177.2) and 164.1 (range, 103.8 to 194.9), respectively. Larger increases were found for overweight and Grade 2 to 4 obesity (BMI :35 kg/M2). For overweight, the uncorrected estimate using 18.5 to 25 kg/m(2) as ideal-weight category was - 88.3 and the corrected estimate using 23 to 25 kg/m(2) as ideal-weight category was 205.4 (range, 114.5 to 296.3). Discussion: Correcting for statistical biases and using higher ideal-weight categories increased the estimate of excess deaths attributable to obesity by similar to 400% and changed the negative estimate for overweight to a large positive estimate.