Assisted Reproduction and Offspring Neurodevelopment—Untangling Confounding by Indication

The relation between assisted reproductive technology (ART) use and neurodevelopmental outcomes in offspring has been a longstanding focus of epidemiological research. Evidence from large, population-based studies suggests that the associated higher risk is generally modest in effect size. Furthermore, the observed elevated risk is more likely explained by having underlying subfertility and its predisposition to obstetrical complications, including multifetal gestation and preterm birth, rather than a product of the ART procedures themselves [1, 2]. In this issue of Paediatric and Perinatal Epidemiology, Delahanty and colleagues [3] examined the relationship between fertility treatment and autism spectrum disorder (ASD) within the Study to Explore Early Development (SEED), a large US population-based case–control study that employed rigorous methods to ascertain ASD. The authors' major methodological contribution is their explicit consideration of female-factor infertility for addressing confounding by indication—a challenge that has been overlooked in prior studies. SEED recruited children aged 2.5 to 5 years across eight US states from 2007 to 2020. ASD case status was ascertained through standardized, in-person assessments, rather than administrative codes from inpatient or outpatient health encounters. Exposure to fertility treatment was captured via structured maternal interviews, thereby optimally classifying ovulation-inducing medications, ART and their combination. The authors found no association between ovulation-inducing medications and ASD, either in the overall sample or in the subset who reported female-factor infertility. Point estimates for ART and combined treatments suggested small elevations in the full cohort, which were further attenuated upon restricting to female-factor infertility. Confounding by indication occurs when a patient factor, such as the decision to receive a specific treatment, also predicts the outcome. What then arises is two exposure groups with different baseline risks for the outcome. In the ART context, individuals who proceed to in vitro fertilisation (IVF) often tend to be older and have a higher burden of comorbidity (e.g., obesity), and also more severe or persistent subfertility. This, in turn, leads to a longer time to pregnancy and a greater number of prior pregnancy losses. These age, comorbidity and latency factors may affect child neurodevelopment through other pathways, regardless of the ART itself. Minimising confounding by indication requires deliberate design choices. The first is the use of an active comparator, such as women with subfertility and no ART, or those receiving ovulation induction/intrauterine insemination (OI/IUI). The second is by restricting or stratifying by the reason for the infertility. The third design choice is the careful use of timing (i.e., temporal order) that avoids conditioning on mediators such as multifetal pregnancy, preterm birth, or caesarean delivery. Besides these design strategies, it is also important to apply appropriate analytic strategies, such as adequate covariate adjustment and the use of advanced causal inference methods to minimise confounding [4]. When mechanistic “causes” are of interest, mediation analysis can help decompose total effects into an effect that is direct versus one that operates through a mediating component (i.e., indirect effect) [1]. Defining female factor infertility based on self-reporting or a prolonged time to pregnancy may have introduced exposure misclassification, as these measures can misclassify male or combined infertility as female factor, potentially biasing associations toward the null. Furthermore, the study by Delahanty and colleagues did not explicitly explore male factor or unexplained infertility, which may carry distinct risks from those attributed to the future mother. Unexplained infertility, typically diagnosed by exclusion after standard evaluation, cannot be readily captured in datasets lacking detailed clinical information. This limitation underscores the potential for residual confounding, as heterogeneity in infertility aetiology may influence both treatment decisions and pregnancy outcomes. Second, statistical precision was low within the subgroup analyses, leaving much uncertainty about any small but potentially meaningful clinical effects. Third, potential mediators—multifetal gestation, preterm birth and caesarean birth—were not modeled explicitly, despite their strong links to both fertility treatment [5-7] and adverse neurodevelopmental outcomes in the child. Formally separating confounding from mediation [1] may have sharpened the study's interpretation. The conclusions of the study by Delahanty et al. align with those of recent population-based studies. For example, a study of over 1.3 million births in Ontario observed a modestly increased risk of ASD among children of parents with infertility—whether treated with ART or not—compared to children of unassisted conception [1]. Notably, mediation analysis further suggested that much of the observed association was attributed to obstetrical and neonatal complications. For example, among children conceived through IVF or intracytoplasmic sperm injection (ICSI), 78% of the excess risk for ASD was mediated by being part of a multifetal pregnancy, 50% by being born preterm, and 25% to 29% by caesarean birth and experiencing severe neonatal morbidity [1]. In another Canadian population-based study, infertility itself was associated with an increased risk of attention-deficit hyperactivity disorder (ADHD) in the offspring, which was not amplified by the use of fertility treatments [2]. Hence, both the biological causes of infertility and the ensuing potential obstetric complications among women with infertility may drive much of the observed association with neurodevelopmental disorders, rather than the receipt of ART itself. There is an ongoing need to reduce adverse pregnancy outcomes among women with infertility. Strategies should continue to use ART while avoiding high-risk multi-fetal pregnancies [5], optimising antenatal care in pregnancies conceived with ART [8] and managing modifiable preconception and early pregnancy risk factors [9]. Future research should integrate mechanistic approaches to clarify potential pathways linking infertility and ART with ASD. Beyond the careful control of confounding by indication, studies should better capture metabolic and inflammatory factors (e.g., polycystic ovary syndrome, endometriosis, obesity and glycemic and inflammatory markers) in the preconception and early gestational periods, as well as possible epigenetic changes that might be measured in the fetus or newborn. Because epigenetic differences have been reported after prolonged infertility and use of ICSI [10], designs should explicitly evaluate plausible treatment-related mechanisms. The latter can be done on embryo culture media and gamete/embryo micromanipulation. Importantly, paternal characteristics remain under-studied, such that future work should ascertain male-factor infertility and measure paternal age, semen parameters, and sperm epigenetic marks, recognizing that ICSI is often performed for male indications. Finally, and most relevant to the current study [3] and this journal, ongoing population-based surveillance is warranted as the characteristics of patients undergoing ART evolve and new ART techniques are introduced. M.P.V. was invited to write the commentary. She contributed to conceptualisation and manuscript writing. J.G.R. Manuscript editing. Dr. Maria P. Velez is supported by a Chercheur Boursier Clinicien award from the Fonds de recherche du Québec—Santé (FRQ-S). M.P.V. has received honoraria from Ferring Pharmaceuticals and EMD Serono outside the submitted work. The authors declare no conflicts of interest. Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.