The question of which environmental risk factors have a causal effect is extremely important for the field of child psychopathology, but one that is frequently overlooked or inadequately tested. Putative environmental risk factors, such as maternal pre-pregnancy obesity, which can be altered or avoided by the individual, are especially important to identify. However there is a need to rigorously investigate whether associated risks are causal, to ensure that preventative interventions and public health messages target factors which will actually lead to improvement in child outcomes. The paper by Chen and colleagues1 moves the field forward with just such an investigation of the relationship between maternal pre-pregnancy body mass index (BMI) and offspring attention deficit/hyperactivity disorder (ADHD).
The elegance of this study is that first, in a large population cohort (n = 673 632), the authors demonstrate association between maternal pre-pregnancy overweight/obesity and increased risk for offspring ADHD (HRoverweight = 1.23, P = 0.01, HRobesity = 1.64, P = 0.01) as found in previous studies, and apply the same statistical controls for measured covariates. Using a subset of the sample (n = 272 790), Chen and colleagues then test whether the observed associations are due to intrauterine effects by comparing the risk of ADHD in full siblings discordant for exposure to maternal pre-pregnancy overweight/obesity (n = 19 814). Their finding that the previously observed association was attenuated (HRoverweight = 0.98, P = 0.82, HRobesity = 1.15, P = 0.38) suggests that familial confounding, rather than a direct intrauterine effect, accounts for the previous associations.
By first showing the previously observed association between maternal pre-pregnancy overweight/obesity and offspring ADHD and then demonstrating in the same sample that this association does not appear to be due to direct intrauterine effects, Chen and colleagues neatly highlight the ability of the discordant sibling design to test causal hypotheses in a way that is not possible in traditional study designs. As such, they also demonstrate that statistical control of measured confounders is inadequate, resulting in incorrect conclusions regarding risk factors, because it fails to take account of unmeasured familial confounding.
Chen and colleagues predict that the familial confounding found by their study may be genetic in origin. This may well be the case; however, there may be other possible explanations. One of the main limitations of sibling comparison designs (as noted by the authors) is that they cannot control for time-varying confounding factors. It has also been noted2 that some factors might vary systematically over time and therefore risk of exposure may differ with birth order or for participants born in the later rather than earlier years of the study (so later-born siblings). For example, the general societal trends towards increased rates of obesity might result in earlier-born siblings having a lower exposure risk to maternal high BMI during pregnancy. Similarly, as identification and diagnosis of ADHD has increased with time, later-born siblings might be more likely to be diagnosed with the disorder than those born earlier. These systematic time-varying factors might introduce uncontrolled bias in this and other sibling-comparison studies.
There also may be stable risk factors which have time-varying expression. For example, social stressors could be present across all pregnancies, but the expression of this stress through weight fluctuation may vary over time. Thus one might observe weight changes between pregnancies even though the same stressor is present at all time points.
Similarly, obesity has been associated with depression as well as other mental health problems3 and it may be maternal depression, rather than obesity per se, which is related to the risk for ADHD (it should be noted that the mechanism for any observed association between maternal depression, obesity and offspring ADHD may be the consequence of genetic or environmental factors).
A further limitation of sibling-comparison studies is that of exposure misclassification. It has been highlighted that misclassification of exposure will lead to a greater bias for sibling-comparison studies than for other between pair designs.4 Further, it has been suggested that, where misclassification occurs, it is more likely to bias findings towards the null.5 As BMI measurement in this study was based on maternal recall, such misclassification is a possibility.
Additional studies are needed to further investigate the nature of the genetic or other familial confounding factors that account for the observed association between maternal pre-pregnancy BMI and offspring ADHD. It is testament to this paper, however, that attempts were made via study design to challenge causal assumptions, and the discussion turns to considering the true nature of the risk pathways.
This paper is well designed, certainly adds to our knowledge and demonstrates the importance of testing for, rather than assuming, causal mechanisms by using a ‘quasi-experimental’ design. However, maternal pre-pregnancy obesity is not one of the most widely studied or established putative environmental risk factors for offspring ADHD; there are few previous studies6–9 investigating the relationship (even without testing causal inference), with some mixed findings depending on who rates the child’s symptoms and the symptom dimension studied.7 Therefore, whereas the authors should be applauded for their study, it is important to note that there are other, perhaps better established, putative environmental risk factors for ADHD and other disorders which would also benefit from similar investigations.
Following the findings of this paper, there is a necessity for further studies using similar and alternative quasi-experimental designs to replicate these findings, test other environmental risk factors and also further investigate what genetic or familial confounding factors account for observed associations if associations are not the result of direct intrauterine effects. For example, we have previously used a cohort of children born following in vitro fertilization, including those who are genetically related and unrelated to the birth mother, to tease apart the influence of genetic and other factors from intrauterine effects.10 Indeed, this paper highlights the benefit of this and other quasi-experimental designs that are able to pull apart likely risks and confounding factors, overcoming the fundamental disadvantage of observational studies in their inability to do so. Such quasi-experimental designs seem especially pertinent to the field of child and adolescent mental health, where exposures to putative risks are likely to be influenced by the behaviour of both parent and child.
Other work that has used alternative designs to test links between maternal prenatal risk factors and offspring ADHD has been undertaken. For example, such studies have investigated the relationship between exposure to maternal smoking during pregnancy and ADHD. These findings also suggest that statistical control of measured confounds is not sufficient and that familial, genetic or other confounding is involved (see Thapar et al. 201311 for a more specific discussion). Although it is important to highlight, as Chen and colleagues do, where a putative risk factor does not appear to be genuinely causal, it is also necessary to make efforts to find environmental risk factors that have direct causal risk effects. Not only will this enable researchers to further elucidate the aetiology of childhood psychopathology such as ADHD, but it is also important to identify suitable targets for prevention and intervention strategies that have an impact on children.
Conflict of interest: None declared.