Use of self-reported height and weight biases the body mass index-mortality association

Abstract

Background: Many large-scale epidemiological data sources used to evaluate the body mass index (BMI: kg/m(2)) mortality association have relied on BMI derived from self-reported height and weight. Although measured BMI (BMI(M)) and self-reported BMI (BMI(SR)) correlate highly, self-reports are systematically biased.

Objective: To rigorously examine how self-reporting bias influences the association between BMI and mortality rate.

Subjects: Samples representing the US non-institutionalized civilian population.

Design and methods: National Health and Nutrition Examination Survey data (NHANES II: 1976-80; NHANES III: 1988-94) contain BMI(M) and BMI(SR). We applied Cox regression to estimate mortality hazard ratios (HRs) for BMI(M) and BMI(SR) categories, respectively, and compared results. We similarly analyzed subgroups of ostensibly healthy never-smokers.

Results: Misclassification by BMI(SR) among the underweight and obesity ranged from 30-40% despite high correlations between BMI(M) and BMI(SR) (r>0.9). The reporting bias was moderately correlated with BMI(M) (r>0.35), but not BMI(SR) (r<0.15). Analyses using BMI(SR) failed to detect six of eight significant mortality HRs detected by BMI(M). Significantly biased HRs were detected in the NHANES II full data set (χ(2)=12.49; P=0.01) and healthy subgroup (χ(2)=9.93; P=0.04), but not in the NHANES III full data set (χ(2)=5.63; P=0.23) or healthy subgroup (χ(2)=1.52; P=0.82).

Conclusions: BMI(SR) should not be treated as interchangeable with BMI(M) in BMI mortality analyses. Bias and inconsistency introduced by using BMI(SR) in place of BMI(M) in BMI mortality estimation and hypothesis tests may account for important discrepancies in published findings.

Figure 1. Parts A-B. Weighted mortality hazard ratios by BMI_M and BMI_SR

Scheme for associating RNA sequence features with splicing outcomes. Top left: More than 1000 diverse features were used; the examples shown here were chosen to illustrate their diversity. Each feature was also defined by the region in which it occurs, as indicated on the map on the lower left, where the alternatively spliced exon is red. Upper right: Exon inclusion data were originally measured in 27 mouse tissues or cell lines using microarrays and then consolidated into four tissue types: C, central nervous system; M, striated and cardiac muscle; D, digestion related tissues; E, embryonic tissue and stem cells. A machine learning algorithm was devised to associate particular features with particular splicing outcomes; the latter being categorized as increased exon inclusion, increased exon exclusion, or no difference in comparing two tissue types. After training on a set of ∼3000 exons, the algorithm was able to reliably predict these splicing outcomes in a set of test exons.