Previous findings on relationships between infertility, infertility therapies, and autism spectrum disorders (ASD) have been inconsistent. The goals of this study are first, to briefly review this evidence and second, to examine infertility and its treatments in association with having a child with ASD in newly analyzed data. In review, we identified 14 studies published as of May 2013 investigating infertility and/or its treatments and ASD. Overall, prior results showed little support for a strong association, though some increases in risk with specific treatments were found; many limitations were noted. In new analyses of the CHildhood Autism Risk from Genetics and the Environment (CHARGE) population-based study, cases with autism spectrum disorder (ASD, n = 513) and controls confirmed to have typical development (n = 388) were compared with regard to frequencies of infertility diagnoses and treatments overall and by type. Infertility diagnoses and treatments were also grouped to explore potential underlying pathways. Logistic regression was used to obtain crude and adjusted odds ratios overall and, in secondary analyses, stratified by maternal age (≥35 years) and diagnostic subgroups. No differences in infertility, infertility treatments, or hypothesized underlying pathways were found between cases and controls in crude or adjusted analyses. Numbers were small for rarer therapies and in subgroup analyses; thus the potential for modest associations in specific subsets cannot be ruled out. However, converging evidence from this and other studies suggests that assisted reproductive technology is not a strong independent risk factor for ASD.
These analyses included 537 ASD cases and 381 TD controls. Parents of children with an ASD were slightly older than TD control parents, and case mothers were slightly more likely to have had a history of smoking; other demographic and lifestyle factors were similar between the groups (Table 2).
Nine percent of both ASD cases and TD controls had used at least one type of infertility treatment for the index birth. Numbers were small for rarer types of therapies, but overall, frequencies were remarkably similar between the ASD and TD groups (Table 3). The ASD case group also did not differ from TD controls by infertility diagnosis or according to hypothesized pathways.
In adjusted analyses comparing ASD cases to TD controls, fertility therapies and infertility continued to show no association with odds of ASD (Table 4). Overall, odds ratios were all close to 1 (OR for overall infertility treatment use = 1.16, 95% CI 0.70, 1.93), though for certain therapies, confidence intervals were imprecise due to small numbers. In particular, any male treatment had only nine exposed cases in the primary analysis, and while similar point estimates were similarly elevated across subgroup analyses for this treatment type (OR approaching 2), these results were not significant. Contrary to our hypotheses, no differences were noted according to hypothesized underlying pathways (Appendix, Table S1). Results were similar across models tested, including in weighted (Model 3, Table 4) and unweighted (Model 2, Table 4) analyses, and did not materially change when using reduced models including only maternal age, child year of birth, and matching factors, or conversely, when considering further adjustment for pre-pregnancy smoking and BMI, which have been associated with both the exposures and outcome under study in some investigations [30,34,35,36].
In subgroup analyses among mothers of advanced age (n = 237) and by diagnostic subgroup (367 autistic disorder and 170 broader ASD), results were very similar, and again non-significant for any associations with infertility and infertility treatments (data not shown; OR for any infertility treatment in the advanced maternal age group: 1.20, 95% CI 0.56, 2.59; in the autistic disorder case group, the corresponding OR was 1.27, 95% CI 0.73, 2.20). However, it should be noted that numbers were small within these groups; only 27 cases used any fertility therapies among the advanced maternal age subgroup, with numbers for individual types of therapies around 10 or fewer. Likewise, sensitivity analyses utilizing only self-reported information, or only information from medical records, also did not demonstrate any significant associations.
The results of this case-control study do not provide evidence for an association between fertility therapies and autism spectrum disorders. We examined a number of different types of therapies and conditions underlying the infertility being treated, and overall found remarkable similarity between ASD cases and typically developing controls. However, due to the low power to detect subtler effects in our study, we cannot exclude the potential for modest associations with rarer therapies or conditions. These topics should therefore be further explored in very large studies with standardized outcome ascertainment and rigorous exposure information.
A major strength of this study, and an improvement over a number of prior studies examining infertility and/or its treatments in association with ASD, is the confirmation of both case status and exposures through rigorous, gold standard measures. All children included in these analyses were evaluated at the UC Davis MIND Institute for diagnostic confirmation, and detailed interviews were conducted and medical records abstracted (in the majority of the study group) for exposure information. In contrast, none of the prior studies examining these factors have confirmed case and comparison group status at this level of detail. We also had information on a full range of infertility diagnoses and treatments, which is lacking in other studies. Our estimates of frequency of use of a wide range of therapies according to ASD status thus fill a needed gap in the literature. Despite using retrospective reporting, as had been previously utilized in a number of prior investigations, we also collected medical records in a large majority of the group for confirmation. We also carried out a thorough confounder identification strategy, whereas many of the prior studies of infertility treatments and ASD failed to adjust for even basic sociodemographic risk factors [16,20,21,22,25,27]. We further took advantage of the linkage of all our cases and controls to the population birth files that included all non-participants in order to account for potential differential participation (selection bias) through weighted analyses, which has not been done in previous case-control studies of this topic.
However, a number of limitations in our work should be noted. Despite a sample size of nearly 1,000 mother-child pairs, our study was limited by the relatively rare exposures, leading to small numbers in many categories. Thus, while we had sufficient power to detect odds ratios of at least 1.75 for treatments and diagnoses with prevalence over 5%, power was reduced to detect associations for specific therapy types with infrequent use. To date, only Hvitjorn and colleagues  have had adequate numbers to examine rarer therapies, but unfortunately, they did not have information on many different types of treatments. We cannot rule out bias due to participation, a common problem in case-control studies, by demographic factors that could be related to the exposures studied here; however, our use of sampling weights strove to mitigate any such biases. While we did rely on retrospective reporting, between 70–80% of our exposures were confirmed in medical records. Another potential limitation, not restricted to our own study, is the definition of infertility itself; how different couples perceive “regular intercourse” is open to interpretation, and timing, diet, lifestyle and cultural factors all may influence reported infertility in ways not related to hypothesized biological pathways.
Consistent with our results, the majority of prior work suggests that use of assisted technologies does not increase risk of adverse child outcomes (with the notable exceptions of multiple births, pre-term birth and low-birth weight). A handful of studies have suggested increased risks of autism, or developmental delay, cerebral palsy, and imprinting disorders with use of ART [15,19,22]. However, our study and four other investigations [17,18,25,27], including the largest study of ASD and assisted conception to date, with over 3,600 cases and approximately 33,000 children exposed to assisted conception , found no association between ART and risk of ASD specifically. We also did not see associations with IVF or other ART subtypes, though numbers were small. Two prior studies have also found no association between ASD and IVF or ICSI [25,26], while results from few others have been inconsistent for more broadly defined developmental outcomes and ART subtypes [19,22,24,27].
For other types of infertility treatments, there is limited information on associations with ASD specifically. A handful of prior studies have suggested associations with ovulation drugs or medications (three studies, each of which found associations only in different subgroup analyses) [17,18,28], specific hormones (two studies) [17,28], and artificial insemination/intrauterine insemination (two studies) [18,28]. Specifically, an investigation in the Nurses’ Health Study II found a significant association between ASD and OID in an advanced maternal age subgroup , which was a larger subgroup than the current study; thus, smaller numbers here could account for the differences seen. The Danish study conducted by Hvitjorn and colleagues  found significant associations with ASD for female offspring exposed to OID as well as for use of follicle-stimulating hormone (FSH)-containing medications, while another study saw an association with urofillitropin, a purified form of FSH, only among multiple births . Given that FSH-containing medications are indicated for a range of underlying problems, the meaning of these findings is not immediately evident. We did not see an association with FSH specifically, and a larger investigation than ours will be needed to replicate results. Another recent study found no association with a general category of infertility medications (that included OID) and ASD among singleton births , but did find a significant association among multiple births. Our results did not differ in multiple or singleton births, though as in the work by Grether and colleagues, exposed numbers among multiples were small, thereby limiting conclusions.
The Nurses’ investigation also saw an association with AI, which the CHARGE study did not replicate; however, the source of infertility treatment information in the Nurses’ study was not as rigorous as in the current study. We did find increased odds ratios for male treatments in our study; though non-significant, use of AI is sometimes indicated for male factors. Again, Grether and colleagues’ study found an association with IUI and ASD only among multiple births but not in singletons, providing mixed results. Given that Hvitjorn and colleagues’ definition of OID included use with and without AI, future studies should also investigate AI in association with ASD, both in singleton and multiple births.
Our analyses of these infertility treatments considered adjustment for a number of potential confounders. Prior studies examining potential effects of infertility treatments have adjusted for or stratified on multiple births in attempt to assess the effect of the treatments on the various outcomes studied, not mediated by multiplicity. For comparison to this work, we examined exposures stratified by singleton and multiple births and found that results did not differ (data not shown; nor did results materially change when adjusting for birth order, which has similar issues when considering effects of infertility and its treatments). However, conditioning on a downstream consequence of exposure can introduce bias. Another example is adjustment for birth weight, a common flaw in studies of prenatal exposures; again, we did not include birth weight in our multivariable models for this reason. Future large studies may benefit from the use of more sophisticated statistical methods, such as marginal structural models (MSMs) [37,38], to determine controlled direct effects not mediated by these factors. Alternatively, using mediator analyses [39,40] may also be useful in determining the impact of factors that may be downstream of infertility therapies, assuming confounding of the intermediate-outcome association is adequately accounted for. Given the null findings for exposures in our study, we did not see significant associations with potential mediators when we conducted such analyses (results provided in Appendix, Table S2); however, pregnancy complications, low birth weight, and multiple births had fairly large estimates of percent mediation. Little prior work has investigated underlying infertility, rather than just its treatments, in association with ASD. While two small studies reported increased prevalence of infertility among mothers of affected children [16,21], and a third study reported an association only for multiple births , two larger studies (one registry-based and one nested case-control) have not found associations between maternal infertility and risk of ASD [17,18]; our work is consistent with these recent findings.
Infertility treatments have been hypothesized to influence ASD through a number of mechanisms, including hormonal influences of the medications, effects of invasive procedures on DNA methylation or other direct effects of the procedure/treatment, impaired egg quality, influences of the underlying infertility, or simply through associations with downstream consequences of the treatment (such as multiple birth, pregnancy complications, low birth weight, or pre-term birth) [15,18,41]. While we had hypothesized that hormonal or inflammatory pathways may be involved, we did not see associations with these pathways as related to infertility and its treatments. However, power was limited to detect modest associations (i.e., those on the order of OR = 1.5 or less), given the number of exposed cases in each of the pathway groups. Continued investigation of such pathways and groupings as conducted here may be useful in learning more about potential underlying mechanisms.
Our work and that of others highlights the need for very large studies in order to fully address the topic of infertility and its treatments in association with ASD. Overall, the evidence to date suggests that women using infertility therapies do not need to be concerned about strong increases in risk of ASD. However, the known risks associated with infertility therapies, such as prematurity, low birth weight infants and multiple births, remain as concerns associated with use of these therapies, and evidence suggests the need for continued long-term follow-up of children conceived using these procedures . Women using these therapies appear to also be at higher risk for pregnancy complications, although this increased risk could be a result of the primary infertility and its root causes. Thus, further investigations are needed to disentangle the complex role of underlying infertility, its treatments, and possible mediators of hypothesized effects on risk of ASD. The limited power to detect modest associations in our study suggests further work may be required to (a) detect subtler risks associated with specific infertility therapies and underlying infertility pathways, and (b) better understand associations in groups such as multiple births and women with advanced maternal age, for whom these treatments and issues are more common.