Sources and severity of bias in estimates of the BMI-mortality association.

Estimates of mortality differences by body mass index (BMI) are likely biased by: (1) confounding bias from heterogeneity in body shape; (2) positive survival bias in high-BMI samples due to recent weight gain; and (3) negative survival bias in low-BMI samples due to recent weight loss. I investigate these sources of bias in the National Health and Nutrition Examination Survey (NHANES) 1988-94 and 1999-2006 linked to mortality up to 2015 (17,784 cases; 4,468 deaths). I use Cox survival models to estimate BMI differences in all-cause mortality risks among adults aged [45-85) in the United States. I test for age-based differences in BMI-mortality associations and estimate functional forms of the association using nine BMI levels. Estimates of the BMI-mortality association in NHANES data are significantly affected by all three biases, and obesity-mortality associations adjusted for bias are substantively strong at all ages. The mortality consequences of overweight and obesity have likely been underestimated, especially at older ages.