The need for genetically‐informative designs in developmental science: Commentary on Gustavson et al. () and Myers et al. ()

The papers in the July issue of JCPP Advances from Gustavson et al. and Myers et al. draw further attention to the critical issue of familial confounding. Familial confounding has been known to be an important consideration in epidemiology for some time now. Many supposedly environmental risk factors are themselves heritable, the so‐ called ‘nature of nurture’ (Plomin & Bergeman, 1991). Some of the genetic and shared environmental risks that influence a particular outcome may themselves also be causes of certain exposures shown to be associated with that outcome, thus confounding their association. While randomized clinical trials provide a gold standard for addressing confounding and establishing causal effect, such study designs are unfeasible or unethical for most potential risk factors. As such, powerful alternative study designs are needed, such as studies that utilize differentially exposed relatives (or include individuals discordant for the outcome), to test whether an association persists beyond familial confounding. The issue is nicely illustrated in Gustavson et al.'s article. They note that prior studies have reported an association between maternal use of acetaminophen (i.e., paracetamol) during pregnancy and attention‐deficit/hyperactivity disorder (ADHD) in their children. However, both of these phenotypes are known to be heritable, and so they tested whether their association was due to familial confounding. They used data from the population‐based Norwegian Mother, Father, and Child Birth Cohort Study (MoBa), which recruited pregnant women in Norway between 1999 and 2008. Mothers were asked to report on medications they had used for medical conditions at gestational weeks 17 and 30, and again 6 months after giving birth. ADHD diagnoses were ascertained from the Norwegian Patient Register. Their results provide a striking illustration of familial confounding. They found that although short‐term use of acetaminophen during pregnancy was not associated with an increased risk for ADHD, longer term use (defined as over 29 days) was associated with an approximately twofold increase in the risk of ADHD. On face value, then, these results could be taken to mean that using acetaminophen during pregnancy could increase the risk of ADHD in children. However, they then studied this association within sibling pairs where mothers had taken acetaminophen during one pregnancy but not the other. The association attenuated drastically in these analyses, thus suggesting that familial confounders are likely to explain the association between maternal acetaminophen during pregnancy and ADHD in children. As acetaminophen was most often reported to be used to manage pain, it is possible that these results reflect shared causes between pain disorders and ADHD. Gustavson et al.'s findings are important, particularly for helping clinicians to provide accurate information to pregnant women about the risks associated with particular medications during pregnancy. Similar studies have been conducted for other medications during pregnancy, such as SSRIs, and report supportive results (Morales et al., 2018). Gustavson et al.'s results do, however, highlight an important cautionary note when it comes to conducting sibling comparisons. Their initial sample size was very large. However, in a sibling comparison design, only the sibling pairs who are discordant for exposure and outcome contribute to the analyses. As a

genetic and shared environmental risks that influence a particular outcome may themselves also be causes of certain exposures shown to be associated with that outcome, thus confounding their association. While randomized clinical trials provide a gold standard for addressing confounding and establishing causal effect, such study designs are unfeasible or unethical for most potential risk factors. As such, powerful alternative study designs are needed, such as studies that utilize differentially exposed relatives (or include individuals discordant for the outcome), to test whether an association persists beyond familial confounding.
The issue is nicely illustrated in Gustavson et al.'s article. They note that prior studies have reported an association between maternal use of acetaminophen (i.e., paracetamol) during pregnancy and attention-deficit/hyperactivity disorder (ADHD) in their children.
However, both of these phenotypes are known to be heritable, and so they tested whether their association was due to familial confounding. They used data from the population-based Norwegian Mother, Father, and Child Birth Cohort Study (MoBa), which recruited pregnant women in Norway between 1999 and 2008. Mothers were asked to report on medications they had used for medical conditions at gestational weeks 17 and 30, and again 6 months after giving birth. ADHD diagnoses were ascertained from the Norwegian Patient Register.
Their results provide a striking illustration of familial confounding. They found that although short-term use of acetaminophen during pregnancy was not associated with an increased risk for ADHD, longer term use (defined as over 29 days) was associated with an approximately twofold increase in the risk of ADHD. On face value, then, these results could be taken to mean that using acetaminophen during pregnancy could increase the risk of ADHD in children. However, they then studied this association within sibling pairs where mothers had taken acetaminophen during one pregnancy but not the other. The association attenuated drastically in these analyses, thus suggesting that familial confounders are likely to explain the association between maternal acetaminophen during pregnancy and ADHD in children. As acetaminophen was most often reported to be used to manage pain, it is possible that these results reflect shared causes between pain disorders and ADHD. consequence, and as seen in Gustavson's paper, the sample size soon declines. Accordingly, confidence intervals are very wide for the sibling analyses, and so conclusions around familial confounding should be drawn with this in mind. Studies that pool data from multiple cohorts and perform meta-analyses of the results across cohorts have the potential to provide solutions to any issues with statistical power that may arise (see e.g., Huybrechts et al., 2018 (Carlsson et al., 2020). Many of these disorders are not necessarily rarer than autism or ADHD either. A UK populationbased study reported that the prevalence of language disorders with no known origin was 7.58%, for example, which is substantially higher than the reported prevalence for autism and similar to prevalence rates reported for ADHD (Norbury et al., 2016). Given that other neurodevelopmental disorders appear to have similar heritability estimates to autism and ADHD, the designs highlighted by Gustavson et al. and Myers et al. have great potential to cast light on the causes of disorders about which little is known.

ACKNOWLEDGMENT
The author has no competing interests to declare. Thanks to Henrik Larsson for comments on a draft of the commentary. Mark Taylor is funded by a Fellows Award from MQ Mental Health Research (MQF20/19).

CONFLICT OF INTEREST
The author is a member of the Editorial Advisory Board for JCPP