Malpresentation and autism spectrum disorder in the study to explore early development: Methodological challenges and considerations

In this issue of Paediatric and Perinatal Epidemiology, Zhang and colleagues report the results of a case–control study examining the association between malpresentation at delivery and autism spectrum disorder (ASD), and whether this association differs by gestational age1. The authors suggest that malpresentation (including breech presentation as well as shoulder, compound, face, and brow presentations) could act as an early indicator of problems in foetal development that manifest in ASD during childhood. They found that malpresentation was associated with a higher odds of ASD (adjusted odds ratio [OR] 1.31, 95% confidence interval [CI] 1.02, 1.68), after adjustment for maternal age, poverty level, hypertensive disorders of pregnancy, and smoking. When results were stratified by type of malpresentation, the associations were similar for breech presentation and other types of malpresentation. Furthermore, the associations did not differ by gestational age, although the magnitude of the effect estimate was larger for term babies compared to preterm babies. Finally, the authors conducted a subgroup analysis by mode of delivery. While an association was observed among the vaginal delivery group, there was no association between malpresentation and ASD among those who delivered via caesarean. The authors conclude that early neurodevelopmental monitoring of children born with malpresentation could identify children with ASD sooner, leading to early intervention and enhanced opportunities that may improve developmental outcomes. Here, we outline some of the paper's key strengths and the methodological challenges that researchers often face when examining associations between pregnancy complications and child morbidity. Additionally, we propose using multidisciplinary efforts to establish potential causal associations to allow for targeted interventions. The study has some key strengths. The analysis included a large sample size, with 4056 participants comprising 1879 cases of ASD and 2177 population controls. Additionally, multiple data sources were used when classifying data on malpresentation, and the evaluation of ASD was conducted using gold standard assessment tools, reducing the likelihood of misclassification of both the exposure and outcome.2 However, disentangling associations between pregnancy risk factors and child outcomes is complex and comes with several methodological challenges and considerations. Perhaps the biggest issue with such studies is that it is not possible to randomise pregnancy complications, and thus, there is a lack of comparability between the study groups. Therefore, great care and consideration must be taken during the design and analysis phase in order to ensure methodological rigour and thus improve causal inference. In particular, in-depth consideration must be given to potential bias by confounding. While Zhang et al. controlled for maternal age, poverty level, hypertensive disorders and smoking (and ethnicity and maternal education through sensitivity analyses), unmeasured confounding (including, but not limited to, maternal body mass index, parental mental health and morbidity) may be contributing to the observed association. For example, as highlighted by the authors, confounding due to maternal thyroid dysfunctions (among other factors) may also play a role in the observed results. Results of a Finnish study3 including 737,788 singleton births examining the association between maternal hypothyroidism and breech presentation suggested a 1.53-fold increase in odds (OR 1.53, 95% CI 1.28, 1.82). In contrast, a systematic review and meta-analysis4 suggested that maternal hypothyroidism was associated with a 1.41-fold increased odds of ASD (OR 1.41, 95% CI 1.05, 1.90). To examine the potential extent of unmeasured confounding by maternal thyroid dysfunction, we calculated an E-value of 1.95 based on the adjusted effect estimate of 1.31 (95% CI 1.02, 1.68) reported in the study. Therefore, assuming appropriate measures are taken to minimise all other types of bias, the minimum strength of association that maternal thyroid dysfunction would need to have with both malpresentation and ASD is 1.95 to explain away the observed odds ratio of 1.31.5 Similarly, gestational diabetes mellitus (GDM) has been shown to be associated with a 42% increase in odds of breech presentation,3 as well as a 42% increase in odds of ASD in a systematic review and meta-analysis,6 which may also partly explain the association suggested by the authors. Therefore, it would be important to examine this research question further using large cohorts that account for maternal thyroid dysfunction and GDM, as well as other potential confounding factors. The role of familial confounding and shared genetics are also important to consider. Previously, we, and others, have used data from Scandinavian national registers to conduct sibling-control studies when examining the relationship between pregnancy risk factors and child outcomes. This allowed to indirectly control for shared familial factors, including genetics, and undiagnosed or unmeasured maternal and lifestyle factors that remain constant across pregnancies.7 Often, any associations observed using traditional methods were attenuated once sibling-controlled analyses were applied.7, 8 For example, using Swedish National Registry data, we estimated that children born by elective caesarean were 21% more likely to be diagnosed as having ASD after controlling for known confounders. However, in the sibling control analysis, estimates for the association between mode of delivery and ASD were obtained using pairs discordant on exposure and outcome, and no association was found, suggesting that previous associations may portend strong familial confounding by shared environmental and genetic factors.8 Moreover, Obel et al.7 examined the association between maternal smoking during pregnancy and the risk of hyperkinetic disorder in the offspring. In the cohort analysis, children prenatally exposed to maternal smoking had twice the risk of being diagnosed with hyperkinetic disorder than children of non-smokers; in the sibling-matched analyses, the hazard ratio was substantially reduced to 1.20. It is important to note, however, that it is possible that unshared factors between siblings that are related to confounders would not be accounted for in the sibling-matched analysis and may lead to biased results, further highlighting the complexities of examining pregnancy risk factors and child outcome relationships. In addition, misclassification bias may be an issue in the current study. For example, maternal smoking was self-reported 3–5 years after pregnancy and therefore subject to recall bias, while use of questionnaire data alone may lead to slight underestimates in the true prevalence of smoking and potentially distort the results.9 Furthermore, as alluded to by the authors, the subgroup analysis by delivery mode should be interpreted cautiously. While the results may indicate an effect modification between malpresentation and vaginal delivery, the mode of delivery is a potential mediator, which makes the interpretation difficult. Additionally, confounding by indication may be an issue in the association between malpresentation and ASD among children delivered vaginally. Therefore, the association observed between malpresentation and ASD should be further explored using mediation analysis and exploring the indications for vaginal delivery when there is malpresentation. We agree with the author's conclusion to a certain extent that introducing early neurodevelopmental monitoring in those exposed to pregnancy complications could identify children with ASD sooner and enhance opportunities to provide support to optimise developmental outcomes. While there is consensus that early identification and intervention can improve long-term neurodevelopmental outcomes, there is debate about recommending such screening for children exposed to adverse pregnancy outcomes.10 For example, the results suggested by Zhang et al. may not warrant the introduction of standardised screening, and it is important to avoid causing unnecessary stress for parents who experienced pregnancy complications. Conversely, early screening is non-invasive and may be worthwhile. However, multidisciplinary efforts combining methodologically sound, large-scale epidemiological studies with animal and/or genetic studies are needed to clarify the purported associations. This approach would provide increased opportunities for establishing potential causal associations between pregnancy complications and neurodevelopment which, in turn, would allow for more targeted interventions to improve outcomes. ASK was invited to write the commentary. GMM drafted the initial version of the manuscript. ASK revised the manuscript for important intellectual content. Gillian M. Maher is a Postdoctoral Researcher and Epidemiologist at the INFANT Research Centre, University College Cork, Ireland. Her research aims to make pregnancy safer and improve outcomes for mothers and babies using epidemiological methods and advanced statistical techniques. This includes secondary analyses of large datasets, including data from Swedish National Registers, Growing Up in Ireland, Millennium Cohort Study and MN-CMS data in Ireland. Her most recent work, which is funded by the Irish Research Council, involves developing a standardised and automated approach to harnessing data from Ireland's MN-CMS, paving the way for the creation of a data warehouse, encompassing information from all pregnancies and deliveries in Irish maternity services. Ali S. Khashan is a Professor in Epidemiology at the School of Public Health, and Principal Investigator at the INFANT Research Centre, University College Cork, Ireland. He has a PhD in Perinatal Epidemiology, MSc in Medical Statistics and BSc in Mathematical Sciences. For the past 20 years, he led several projects in perinatal epidemiology using data from Denmark, Sweden, the UK, Ireland and Tanzania. Open access funding provided by IReL. GMM is supported by the Irish Research Council Government of Ireland Postdoctoral Fellowship Programme [GOIPD/2023/1297]. The authors have no conflicting interests. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.