Using Twin Controls to Study the Effects of BMI on Mortality

Abstract

We investigated whether confounding from genetic factors could explain some of the observed associations between BMI and mortality.1 We used co-twin-control analyses, a well-known method2 that has contributed to many important findings in the past.3,4 We found that within monozygotic (MZ) twins there was no association between BMI and mortality from causes other than CHD, suggesting that part of the well-documented association between BMI and mortality may be due to genetic factors rather than causality. In their commentary, Kaufman and Glymour5 express concerns about potential sources of error—including violation of the equal-environment assumption, selection and survivor bias, and lack of representativeness—that may limit the validity of this finding. Kaufman and Glymour5 are concerned that MZ twin pairs may be more alike than dizygotic (DZ) pairs with regard not only to genetic factors but also to environmental factors, violating the “equal-environment assumption” underlying many methods used in twin research.6 In the article7 referred to by Kaufman and Glymour, the authors stated that “To date, most empirical findings have suggested that violations of the [equal-environment assumption] do not substantially bias estimates of heritability obtained from twin studies.”8,9 Furthermore, even though the equal-environment assumption is fundamental for estimating heritability, it is of little concern in co-twin control studies and would not result in biased estimates in MZ pairs. Rather, a higher degree of similarity in environmental factors within MZ pairs would imply that confounding from environmental factors is better controlled within BMI-discordant MZ pairs than within DZ pairs. As mentioned by Kaufman and Glymour,5 our data were based on BMI reported some 40 years ago. Exposure information was thus crude, and the proportion of obese people was low compared with the situation today. We agree that the generalizability of our findings could be questioned, although when we analyzed the twins as an ordinary cohort, the association between BMI and mortality was almost identical to findings from a recent meta-analysis based on data from 57 cohort studies.10 As stated in the paper, our findings cannot be extrapolated to obese subjects. Other potential concerns are potential survivor and selection bias. As noted by Kaufman and Glymour, the twins included in the co-twin control analyses (pairs where at least one twin had died during follow-up) were older than the full cohort of twins. The association between BMI and mortality was weaker in this subset, which is to be expected because the association is known to attenuate with age.11 To eliminate potential survivor bias, the twins would have had to be followed longer (ideally until all twins were dead). Still, it is difficult to see how this would lead to a stronger association between BMI and mortality in MZ twins. Extending the follow-up time means that the twins would be even older and, hence, the association between BMI and mortality would probably be attenuated further. As a consequence of the strong heritability of BMI, fewer MZ than DZ twin pairs were BMI discordant, and the mean difference in BMI was smaller in MZ twins. Still, when the co-twin control analyses were stratified by degree of BMI discordance, the association between BMI and mortality was consistently stronger in DZ than in MZ pairs. While we acknowledge that there may be unmeasured confounding systematic errors in our study, including those discussed above and these errors would have to differ between MZ and DZ twins to explain why BMI was related to non-CHD mortality only in the latter group. Also, if the systematic errors were more pronounced in analyses of MZ twins, why was the association between BMI and death from CHD—an association with stronger mechanistic data to support causality—similar in MZ and DZ twins? To evaluate causal effects, one must surely take into account the results from many different study designs, as each has biases and limitations. No study in itself is conclusive. One cannot selectively opt out of studies that contradict one's favorite hypothesis—if people had done so in the past, we would still be using leeches. We, therefore, still believe that it is worthwhile to consider the idea that previous concerns regarding the effect of overweight on non-CHD mortality might have been overemphasized.