Year: 2014
Type: editorial
Abstract: It has long been recognized that the composition of the human body, especially the amount and distribution of adipose tissue, plays an important role in the development of many diseases. However, accurately and precisely quantifying these features requires fairly onerous assessments such as those that involve body scans, hydrostatic weighing, air displacement plethysmography, or doubly labeled water. Thus, for very practical reasons, simple anthropometric indices are often favored for the estimation of body corpulence. Using weight alone is often insufficient, as big people are sometimes very lean and small people are sometimes very fat; this is particularly so early in life, when children are growing up and out at variable rates. Thus, controlling weight for height is necessary in most circumstances where overall body adiposity is the phenotype of interest, and especially so in studies of children and adolescents. The body mass index (BMI: body weight in kilogram divided by the square of height in meters) is probably the most commonly used metric for determining whether a person's weight, given their height, is likely to pose health risks. The BMI equation was initially proposed in the mid-1800s by the Belgian polymath Adolphe Quételet, then termed the Quételet index and later renamed “BMI” by Keys et al (1); since then, the BMI equation has been embraced by a wide variety of end users, including insurance companies, health care providers, the military, school boards, and scientists, typically for the purpose of prognosticating about future health. Recent technological and analytical advances in population genetics and emphasis on hypothesis-free analyses have demanded studies that include tens or even hundreds of thousands of participants. The nature of such large studies is that compromises are made on the quality of phenotypic data, and rather than focusing on objective measures of adiposity, studies of obesity genetics including those in children (2) often analyze genetic associations with anthropometric estimates of body composition (3). Despite correlating reasonably well with objective measures of total adiposity (4) and various disease endpoints (5), there is a range of limitations associated with BMI (6). For example: (i) it does not characterize compartment-specific adiposity; (ii) it may misclassify people who are short and muscular as overweight or obese and tall, fatter people as normal weight, which might lead to inaccurate assumptions about health prognosis and may confound epidemiological analyses, particularly those that focus on interactions between genotypes and lifestyle; (iii) because BMI is correlated with height and an imperfect proxy for total adiposity, it is likely that genetic associations studies focused on this trait will fail to detect some loci truly associated with the latter phenotype, particularly in pediatric cohorts where height is a dynamic trait strongly linked with growth and development; and (iv) people of the same BMI sometimes have starkly different underlying body compositions (7), which reduces statistical power in association analyses when BMI is used in place of a more accurate measure. In a landmark paper published in this edition of the Journal (8), Stergiakouli et al. report analyses demonstrating that when body weight is statistically disassociated from height (BMI[x]), and this variable is used as the outcome in genome-wide association analyses, those results differ markedly from those where BMI is used as the outcome. The authors initially studied 5,809 children (mean age 9.9 years) from the Avon Longitudinal Study of Parents and Children, a prospective cohort study from the southwest region of the United Kingdom (9); replication analyses were performed in 2,729 Dutch children (mean age 6.1 years) from Generation R, a prospective cohort study set in Rotterdam, The Netherlands (10). The authors identified a locus (ADCY3—adenylate cyclase 3) that harbors a number of variants that are robustly associated with BMI[x] and objectively assessed adiposity, but which are not robustly associated with BMI. Although ADCY3 has previously been associated with prepubertal growth (11), it has not been specifically reported as a childhood obesity locus. Aside from unveiling and characterizing a novel childhood obesity locus, Stergiakouli et al.'s paper is important because it highlights the often-overlooked need to fully account for different rates of linear growth in genetic association studies of body corpulence in children. Why is BMI correlated with height? The BMI equation involves division by height (in meters squared). Some might assume that division neutralizes the dependency of the numerator (in this case body weight) on the denominator (height). This assumption does not always hold unfortunately, as has long been known. Indeed, more than 60 years ago, Tanner discussed the use of ratios for the characterization of oxygen consumption and cardiac output (12), physiologic traits that were and still are commonly expressed per unit body weight or body surface area. The objective of dividing a physiological measurement by body size is often to provide an estimate that is independent of the latter; yet the approach is flawed. As Tanner stated when discussing the use of ratios in the clinical scientific literature, “…investigators had drawn positive conclusions not justified by their data, had been confused by a seemingly uninterpretable phenomenon in their results, had proposed a less effective and more biased normal standard in preference to a more effective and less biased one, had reduced a correlation between two physiological functions from a very high to a medium value, and had invented a new clinical syndrome. All these events were wholly or in major part due to the incautious use of ratios.” (12) Although most investigators who use BMI in genetic association analyses are no doubt cognizant of the limitations of the measure, there is a prevailing and understandable view that in the absence of a widely available and more accurate and precise index of adiposity, BMI is good enough. However, as Stergiakouli et al. clearly demonstrate (8), the BMI equation can easily be optimizing using available data, such that weight bears no correlation to height. Although the residual correlation of BMI with height is probably of most concern in pediatric obesity studies, one wonders if the reanalysis of existing adult datasets using BMI[x] or other anthropometric indices of body size (13, 14) might also reveal novel adult obesity loci.
Source: Obesity
DOI: 10.1002/oby.20858
Author: Paul W. Franks
Language: English
Subfield: Genetics
Domain: Life Sciences
Open Access status: bronze
Funder: Medical Research Council
Award: G4492458937
OpenAlex ID: W1955553090