Biases in the mortality risk versus body mass index relationship in the NHANES-1 Epidemiologic Follow-Up Study

Int J Obes Relat Metab Disord. 2001 Jul;25(7):1071-8. doi: 10.1038/sj.ijo.0801648.


Background: Conflicting research findings have created uncertainty as to the ideal body weight.

Objective: To test whether average-adulthood BMI (body mass index, weight in kilograms per height in meters squared) is a more reliable predictor of mortality risk than the one-time baseline BMI measurement which has been used in previous studies.

Design: Prospective cohort study with a 20 y follow-up period, the First National Health and Examination Survey (NHANES 1) Epidemiologic Follow-up Study (NHEFS).

Subjects: A random probability sample of 14,407 adults aged 25-74 y in the 1971-1975 NHANES 1.

Measurements: Body-weight data were derived from measurements in the 1971-1975 and 1982-1984 NHEFS surveys, and current and recalled estimates by subjects in the 1971-1975, 1982-1984, 1986 and 1987 surveys. Height was measured in the 1971-1975 survey. Socio-economic and lifestyle covariates were derived from questionnaires administered in all four surveys. Almost all deaths were verified by death certificates.

Results: The shape of the morality risk vs baseline BMI curve varied due to a bias (late-life bias) caused by a steep decrease in BMI among the elderly toward the end of the lifespan. In a particular analysis, the portion of subjects at baseline who were deceased elderly with lower BMI than age-matched censored subjects indicated the size of the bias, and appeared to be the proximate cause. Strong evidence was found to support the proximate cause being the actual cause. Excluding deaths during early follow-up was only able to remove the late-life bias in limited circumstances. Using average-adulthood BMI as mortality predictor and baseline BMI as covariate was very effective in removing the late-life bias. It also appeared to remove the reverse-causation and regression-dilution biases, and yielded robustly positive mortality risk vs average-adulthood BMI curves in all analyses in which it was tested.

Conclusions: Average-adulthood BMI appears to be an appropriate predictor of mortality risk, provided baseline BMI is used as a covariate. Among non-elderly persons, being leaner meant a lower mortality risk, down to the lowest category of leanness in the study-<20 kg/m(2). Future survival analyses of the mortality-BMI relationship should account for the effects of the regression-dilution, reverse-causation and late-life biases.

MeSH terms

  • Adult
  • Aged
  • Bias
  • Body Mass Index
  • Body Weight*
  • Cohort Studies
  • Female
  • Follow-Up Studies
  • Humans
  • Male
  • Middle Aged
  • Mortality*
  • Nutrition Surveys
  • Prospective Studies
  • Risk Factors
  • Time Factors