Primary Care Corner with Geoffrey Modest MD: BMI, Height and Socioeconomic Status

By Dr. Geoffrey Modest

One clear limitation of observational epidemiologic studies is that they are only capable of showing an association, not causality. One way to assess possible causality is through a technique called “mendelian randomization”.  There was a recent study looking at the association between BMI or height and socioeconomic status (see BMJ 2016;352:i582​). In this study, one might suspect that the well-known association between BMI and socioeconomic status (SES) could be bidirectional (high BMI might predispose a person, through conscious or unconscious bias, to be less likely to be hired or make much money, hence the lower SES; and, low SES might predispose people to a higher BMI since they are less likely to be able to afford or even have access to more expensive vegetables and fruit, have less ability to exercise because of unsafe neighborhoods, etc.). Mendelian randomization is a process where one looks at genetic differences (which are randomly allocated at conception) as an unconfounded proxy for the risk factor (in this case, BMI and height). So, since the outcome of SES cannot influence the underlying genetic variation which is set at birth, by looking at the genetic code itself instead of the complexities of BMI or height (which may reflect the genes but also be confounded by the social environment), one might see that, for example, the genes associated with increased BMI or height really are the potential drivers of low SES. The ability to do this genetic testing is now greatly enhanced because of the widespread availability of cheap genome-wide studies, currently identifying 10’s of genetic variants associated with BMI and 100’s associated with height.

Details of this study:

• 119,669 British men and women, aged 37 to 73 in a UK Biobank that had recruited 500,000 people from across the UK from 2006-2010, and included demographics, health status, lifestyle, blood pressure and stored samples of urine, sputum and blood. This analysis was limited to white people with “British ancestry” because those from other ethnic groups did not have enough representation to be adequately powered for a statistical association.
• Main outcome: the relationship between BMI or height and the following 5 markers of SES: age completed full-time education, level of education, job class, annual household income, and Townsend deprivation index (a composite of unemployment, non-car ownership, non-home ownership, and household overcrowding).

Results:

• Overall analyses:
  • Pretty much all of the five SES markers were very highly correlated with coronary artery disease, hypertension, type 2 diabetes and “long illness”
  • Tall stature was related inversely to all of the above SES indicators individually, equivalently for men and women; by genetic analysis, these associations were only significant for men
  • Higher BMI was related directly to all of the above SES indicators individually, equivalently for men and women​; by genetic analysis, was only significant for women, and only for the annual household income and the Townsend deprivation index
• Genetic analysis found that these relationships were partly causal:
  • A 1-SD (6.3cm) taller stature “caused” :
    • 06 year older age of completing full-time education (0.02-0.09, p=0.01)
    • 12 times higher odds of working in a skilled profession (1.07-1.18, p<0.001)
    • $1615 higher annual income ($970-2260, p<0.001), with stronger association in men
  • A 1-SD higher BMI (4.6 kg/m²) “caused”:
    • $4200 lower annual income ($2400-6000, p<0.001) in women
    • 10 higher level of deprivation (0.04-0.16, p=0.001), but only in women

So, what are the implications of this study:

• We know from many studies that lower SES is association with poorer health and shorter life expectancy
• And, we know that adult height and BMI are associated with SES, with many studies in richer countries finding that taller people and those with lower BMI have higher SES and better health
• I cited a previous mendelian randomization trial for alcohol and cardiovascular events, assessing the often-found association between drinking more and less CAD. Mendelian randomization found the opposite: those drinkers with a genetic variant for alcohol dehydrogenase had lower alcohol consumption and lower levels of CAD risk factors (blood pressure, inflammatory markers, etc.) compared to those without this allele, and that the lower risk of CAD events was present in those with this allele who actually were lower alcohol consumers (see BMJ 2014; 349:g4164 4 doi: 10.1136/bmj.g4164,as well as the editorial in Addiction: doi:10.1111/add.12828​). So this suggests that alcohol consumption by itself is not cardioprotective, but there was likely some confounding or bias inherent in the observational studies. And this bias was the cause of the prior held association of increased alcohol consumption and cardioprotection.
• The current study did find that the observational associations between BMI and height with SES were in fact significantly stronger than the genetic analyses, suggesting that genetics played some role, but actually only a small part in the association (i.e., a significant part of the association between BMI/height and SES may well be social and possibly bidirectional)
• The authors do acknowledge several limitations: even if there is a genetic predisposition to a higher BMI, do people with higher BMI move to areas where they are more comfortable? (Areas with others having higher BMI, and perhaps these are areas of lower SES, leading to their children being born in areas of lower SES, getting less good education, etc. And the higher BMI itself does have pretty clear adverse health effects, leading to more illness and less ability to be employed and decreasing SES. So, the genetic association may well be dwarfed by the social issues. A little harder to make equivalent arguments about height.
• Although this is a huge dataset, it seems to me that there are a couple of limitations to the interpretation (I should add that these are my assessments based on what I glean of the mandelian randomization methodology, and would defer to anyone more statistically or biologically sophisticated than I — would welcome feedback and would like to send around contrary interpretations). One is: do we really know all of the important genetic “determinants”, and which really are the important ones? And is there any significant interaction between the many identified genetic markers (i.e., does it make sense to look at individual genetic markers singly as in this study or in combination?).  Second: while it is undoubtedly true that social circumstances cannot change one’s genes fundamentally, it is also true that the function of genes can change at pretty much anytime through epigenetics by DNA methylation and histone modification, leading to changes in the expression of these genes. For example, rats exposed to marijuana have changes in DNA methylation, which can affect function (see https://stg-blogs.bmj.com/bmjebmspotlight/2015/07/29/primary-care-corner-with-geoffrey-modest-md-marijuana-passing-through-the-generations/ ). This blog goes through some of the basic principles of epigenetics, raising the concern that this post-conception phenomenon may have very profound clinical effects. In the case of the rats, where DNA methylation was actually passed to a subsequent generation, those infant mice were more prone to opioid addiction. And there are a variety of epigenetic changes associated with the environment, lifestyle, and chronic diseases themselves. For example, there are animal studies showing that there is a pretty strong association between nutrition and DNA methylation (e.g. J Nutr Biochem 2012; 23: 853); in fact there is a new journal “Environmental Epigenetics” with articles on such things as epigenetics as a mediator between air pollution and preterm birth, chemical exposures and autism, etc.
• So, bottom line: there may well be genetic variants leading to obesity and short stature, and these may predict a small part in their determining decreased SES. But overall, I think this study reinforces that the relationship between BMI or height with SES is predominantly a social one: there are social biases, conscious or not, which directly affect the overall ability of overweight women and short men to achieve their full human potential, and this is reflected in their lower SES. And that lower SES may well contribute to higher BMI. I bring up this article because there are more and more studies using mendelian randomization, including a recent one on HDL-cholesterol, and that it seemed reasonable to think about the potential value and shortcomings of this technique.