Reader Comments

Post a new comment on this article


Posted by plosmedicine on 31 Mar 2009 at 00:33 GMT

Author: Debbie Lawlor
Position: Professor of Epidemiology
Institution: MRC Centre for Causal Analyses in Translational Epidemiology , Department of Social Medicine, University of Bristol
Additional Authors: Nicholas J Timpson, Roger M Harbord, Sam Leary, Andy Ness, Mark I McCarthy,Timothy M Frayling, Andrew T Hattersley, George Davey Smith
Submitted Date: December 24, 2008
Published Date: December 24, 2008
This comment was originally posted as a “Reader Response” on the publication date indicated above. All Reader Responses are now available as comments.

Rebecca Reynolds and colleagues highlight three problems with our study or our interpretation of its results: (a) that the use of parental self report of weight and height is unreliable; (b) that our cohort of pregnant mothers are too lean to address the specific research question that we aim to address and (c) that our instrumental variables analysis is imprecise.[1] We shall reply to each of these points in turn.

a) We acknowledge that self report of parental weight and height may not be ideal.[2] However, for the greater magnitude of association of mother’s body mass index (BMI) with offspring fat mass compared to father’s BMI with offspring fat mass to be biased in a way that would alter our conclusions would require differential error in self-report by mothers and/or fathers. With respect to the outcome of offspring fat mass – a measure that was unknown at the time of parental self-report – we think it implausible that either self-report could be differential – how could parents have misreported in any way that was related to future offspring fat mass 9+ years later? Errors in both reports could plausibly be non-differential and therefore would have the expectation to bias associations towards the null (though we acknowledge in any given study this expectation may not occur). If error were more likely with mother’s self-report then the greater effect of mother-offspring association may be an underestimate (since the expectation would be that the association in mothers was weaker than the truth). However, when we compare mother’s self-reported ‘pre-pregnancy’ weight to actual measured weight at recruitment, in early pregnancy, we find very little evidence for error. For women who were measured at 10 weeks or less gestation (N=3301) the correlation coefficient between self-reported pre-pregnancy weight and measured weight in early pregnancy was 0.96, suggesting very little error in this maternal report. If error were more likely with fathers then our difference in association between mothers and fathers would be (by statistical expectation) an exaggeration of the true difference (since the expectation would be that the association in the fathers was weaker than the truth and the association in fathers might in reality be closer to that in mothers). We are unable to compare fathers self-report to any actual measurements, but any bias that resulted in an exaggeration of what is a modest higher effect in mothers would in fact strengthen our conclusion in the paper. The analyses with variation in FTO would not be affected by any measurement error since individuals will have no knowledge of their FTO status. This instrumental variable will measure an unbiased difference in mother’s adiposity.[3]

b) There is evidence that extreme maternal obesity during pregnancy is associated with offspring later obesity and that this association is likely to be due to intrauterine mechanisms. In a study of women who had undergone biliopancreatic surgery for extreme mobid obesity (mean BMI 48kg/m2), within 45 sibling pairs the prevalence of obesity at age 2-18 years was 52% lower (relative difference) in those siblings born after the mother had lost weight markedly following surgery than in their siblings born prior to their mother’s surgery.[4] However, we disagree with Reynold’s et al. that assessing graded associations between maternal BMI and offspring fat mass is inappropriate when our interest is in trying to understand the extent to which developmental overnutrition might affect population levels of obesity and hence population health. Furthermore, the various pieces of evidence that have been used to hypothesise the associations we tested would support such graded associations. Thus, there is strong evidence for a graded association between maternal glycaemia (at levels below that required for a diagnosis of gestational diabetes) during pregnancy and macrosomia,[5] maternal BMI (consistent with BMI in general in adults) has a graded linear association with levels of glycaemia,5 and whilst there is some evidence that the obesity epidemic is driven by a change in the shape of the BMI distribution (with greater right skewing) on the whole mean BMI has increased in populations experiencing the epidemic.

Reynolds argue that our population are too lean to contribute to this area of research and refer to a study of obesity in pregnant women in Middlesbrough in whom the percentage who were obese (>30kg/m2) at booking clinic (up to 16 weeks of gestation) increased from 9.9% to 16.0% between 1990 and 2004.[6] There are two important flaws with this argument. First, if we use equivalent data from ALSPAC – i.e. measured early pregnancy weight measured at booking clinic (when participants in ALSPAC were first recruited) and only including those assessed up to 16 weeks, as in the study cited by Reynolds, we find that 11% of our sample were obese, which is considerably greater than the estimate of 3% made by Reynolds et al. and refutes their assertion that ours is an unusually lean popution. Secondly, and most importantly, if the effects of developmental overnutrition are only seen in the extreme (for example those morbidly obese mothers in the study by Kral et al. referred to above[4]) then by definition this process cannot be driving the obesity epidemic and will have limited public health importance.

c) We are very explicit in our paper about the imprecision of the instrumental variable estimate.[2] We are currently working on a number of studies that aim to further test these associations with greater statistical power. We were a little uncertain about the specific points being made in the fifth paragraph of this letter. The association of offspring FTO with their fat mass has been previously reported:[7] for each additional at risk allele, fat mass at age 9 to 11 increases by 0.12SD (95%CI: 0.08, 0.16). This was not reported in our PloS Medicine paper since the concern of that paper was using variation in maternal FTO as an instrumental variable for maternal adiposity in order to assess its association with future fat mass in offspring as a test of the importance of the developmental overnutrition hypothesis for population health. The point made by Reynold’s et al. about the small contribution of FTO to variation in BMI is also well established and we have discussed elsewhere the relevance of this to Mendelian randomization/instrumental variable studies in general.[3,8] In brief, this is not disimilar to randomised controlled trials (RCT). For example, in RCTs of antihypertensives, randomisation to a single class of antihypertensive will explain a very small proportion of variation in blood pressure (~2%) in the whole study sample. However, such RCTs can and have demonstrated the unbiased, unconfounded causal effect of blood pressure on stroke and other cardiovascular disease outcomes.[3,8]

As we discussed in full in our original paper, we agree that more work in this area is required to understand any likely impact of maternal adiposity during pregnancy on future offspring risk of greater adiposity. However, we believe that our interpretation of this set of results is correct, i.e. they provide little evidence that developmental overnutrition has made an important contribution to the obesity epidemic to date. Indeed if Reynold’s et al. believe that very high levels of obesity (such as those reported for 2004 in Middlesborough) are required for developmental overnutrition to have an important impact on population levels of obesity, then they must also agree that this process can not have contributed to the current obesity epidemic that we are experiencing, since the start of the childhood and adult obesity epidemic in the UK, and other developed countries, clearly predates 2004.[9]

Debbie A Lawlor (1,2), Nicholas J Timpson (1,3), Roger M Harbord (2), Sam Leary (4), Andy Ness (4), Mark I McCarthy (3,5), Timothy M Frayling (6,7), Andrew T Hattersley (6,7), George Davey Smith (1,2)

1 MRC Centre for Causal Analyses in Translational Epidemiology, University of Bristol, UK
2 Department of Social Medicine, University of Bristol, UK
3 Wellcome Trust Centre for Human Genetics, University of Oxford, UK
4 Department of Oral and Dental Science, University of Bristol, UK
5 Oxford Centre for Diabetes, Endocrinology and Metabolism, University of Oxford, UK
6 Genetics of Complex Traits, Institute of Biomedical and Clinical Sciences, Peninsula Medical School, Exeter, UK
7 Diabetes Genetics, Institute of Biomedical and Clinical Sciences, Peninsula Medical School, Exeter, UK


1. Reynolds R, Seed P, Poston L. Response to Lawlor et al: Exploring the developmental overnutrition hypothesis using parental-offspring associations and FTO as an instrumental variable. Plos Medicine 2008
2. Lawlor DA, Timpson N, Harbord R, Leary S, Ness A, McCarthy MI, Frayling TM, Hattersley AT, Davey Smith G. Exploring the developmental overnutrition hypothesis using parental-offspring associations and the FTO gene as an instrumental variable for maternal adiposity: findings from the Avon Longitudinal Study of Parents and Children (ALSPAC). PLoS Medicine 2008;5:e33
3. Lawlor DA, Harbord RM, Sterne JAC, Timpson NJ, Davey Smith G. Mendelian Randomization and Instrumental Variables. Statistics in Medicine 2008;27:1133-1163
4. Kral JG. Biron S, Simard S, Hould F-S, Lebel S, Marceau S, Marceau P. Large maternal weight loss from obesity surgery prevents transmission of obesity to children who were followed for 2 to 18 years. Pediatrics 2006;118:e1644-1649.
5. Metzger BE, Lowe LP, Dyer AR, Trimble ER, Chaovarindr U, Coustan DR, Hadden DR, McCance DR, Hod M, McIntyre HD, Oats JJ, Persson B, Rogers MS, Sacks DA. Hyperglycemia and adverse pregnancy outcomes. N Engl J Med. 2008;358:1991-2002
6. Heslehurst N, Ells LJ, Simpson H, Batterham A, Wilkinson J, Summerbell CD. Trends in maternal obesity incidence rates, demographic predictors, and health inequalities in 36,821 women over a 15-year period. BJOG 2007;114:187-194.
7. Frayling TM, Timpson NJ, Weedon MN, Zeggini E, Freathy RM, Lindgren CM, Perry JRB, Eilliot KS, Lango H, Rayner NW, Sheilds B, Harries LW, Barrett JC, Ellard S, Groves CJ, Knight B, Patch A-M, Ness AR, Ebrahim S, Lawlor DA, Ring SM, Ben-Shlomo Y, Javelin M-R, Sovio U, Bennett AJ, Meltzer D, Ferrucci L, Loos RJF, Wareham NJ, Karpe F, Owen KR, Cardon LR, Walker M, Hitman GA, Palmer CNA, Doney ASF, Morris AD, Davey Smith G, The Wellcome Trust Case Control Consortium, Hattersley AT, McCarthy MI. A common variant in the FTO gene is associated with body mass index and predisposes to childhood and adult obesity. Science 2007; 316:889-894
8. Davey Smith G. Randomised by (your) god: robust inference from an observational study design. J Epidemiol Community Health 2006; 60:382-388
9. Kipping R, Jago R, Lawlor DA. Obesity in children. Part 1: Epidemiology, measurement, risk factors, and screening. British Medical Journal 2008; 337:a1824.

No competing interests declared.