This study provides evidence that IPIs of <24 months are associated with increased odds of autism in second-born children. Using pairs of full-sibling singleton births drawn from a large California population, we found an inverse association between autism and IPIs, with the greatest risk occurring for IPIs of <12 months. The association was not mediated by preterm birth or low birth weight and persisted across categories of sociodemographic characteristics, with some attenuation in the oldest and youngest parents. A case-sibling control analysis additionally suggests that although confounding by family-level factors may be present, it does not entirely account for the association, particularly for IPIs of <12 months. To our knowledge, this is the first study to specifically address the association between IPIs and autism risk.
One possible explanation for this association is some alteration of maternal physiology associated with IPI. The association of short IPIs with adverse pregnancy outcomes including low birth weight and preterm birth has been investigated previously, with maternal nutritional depletion, in particular of folate, representing a leading hypothesis.8
Folate is required during pregnancy for DNA synthesis and cell division. Without supplementation, serum and erythrocyte levels decrease from midpregnancy.14
O'Rourke et al15
found that erythrocyte folate levels, which are more representative of long-term status than serum levels, declined through at least 12 months postpartum. Our observation of the greatest relative odds of autism in second-born siblings in pregnancies after intervals of <1 year, with a rapid decline thereafter, is consistent with the hypothesis that folate status might also be related to autism risk. However, other possible mechanisms, such as maternal levels of iron and polyunsaturated fatty acids16
should also be considered. Some attenuation of the IPI–autism association occurred among younger and older parents, which suggests an explanatory mechanism that is less correlated with IPIs in these groups. Dilution by greater baseline risk may also occur among children of older parents.12,18
Use of administrative records of the California DDS for identification of autism represents a strength of the study, facilitating population-based analyses over 11 years of birth records from this populous and diverse state. However, inclusion as a case subject depends on seeking services and receiving a qualifying diagnosis, with previous reports estimating that 75% to 80% of people with autism in California register with the DDS.19
This could potentially bias estimates if there was differential diagnosis with autism or use of DDS services by IPI. For example, parents of children similar in age may be more attuned to typical development and any delays in the second. In some,20,21
although not all,22
studies parents of children with autism have reported first having concerns about their child's development earlier when there was an older sibling. If this explains some of the association observed here, it may indicate a need for increased attention to developmental surveillance, especially where parents are less familiar with patterns of typical development. If this were the primary explanation for our results, we could expect a stronger effect among less severe cases, with the reasoning that more severe cases are less likely to escape diagnostic attention. Case subjects with a comorbid diagnosis of mental retardation had an OR for IPIs of <12 months of 3.51 (95% CI: 2.73–4.49) compared with 3.25 (95% CI: 2.82–3.75) for those with no recorded mental retardation diagnosis, suggesting that this is not the case. However, because a child with autism and mental retardation may not have both diagnoses in the DDS record,23
this merits additional investigation with more accurate measures of severity. Additional limitations related to the administrative nature of the data include potential errors in matching siblings and misestimation of IPIs because of incorrect reporting of gestational age. Although we took steps to minimize these sources of error, we cannot rule them out. To the extent to which such errors lead to nondifferential misclassification of IPIs, we would expect estimates to be biased toward the null. This may explain why ORs for IPIs of <6 months are lower than those between 6 and 12 months ( and ), if the shortest IPIs are more likely to be misclassified values.
Another concern is that autism diagnosis or symptoms in a first child may impact the decision to have a second child (“stoppage”) and that this would be more likely with longer IPIs. To avoid potential bias this would induce, we included in logistic regression analyses only subjects whose first sibling did not have an autism diagnosis; if the firstborn child does not have autism, stoppage cannot effect whether a second child is born. Our results, therefore, pertain to simplex families; whether differences would be observed in multiplex families or whether affected families in our sample later included additional children with autism is not addressed by our analyses. This approach could not be applied to the case-sibling control analysis (), which rests on comparing diagnoses in first- versus second-born siblings. If stoppage among families with longer IPIs resulted in the observation of fewer pairs with affected firstborn children, we would expect ORs to be biased upward at longer IPIs. Because we observe elevated ORs at short, rather than long, IPIs, such a bias is unlikely to explain the results.
As in any epidemiologic study, this association could arise through uncontrolled confounding if families who tend to have shorter IPIs are also at greater risk of autism for other reasons (ie, parental genetic or hormonal factors or social influence on both autism diagnosis and decisions regarding family structure). We were able to analytically address family-level confounders (those to which all siblings within a family are equally exposed). If an unmeasured family-level factor associated with short IPIs and risk for autism was primarily responsible for the observed association, we would expect equal risk in first- and second-born children, because they are equally exposed. The fact that instead we see second-born siblings at disproportionately greater risk for short IPIs indicates that this is not the case. On the other hand, we cannot rule out confounding by unmeasured individual-level factors that vary between children within families. Social24
exposures have been associated with autism diagnosis in this population and may vary between children in families who move. However, we have no specific reason to believe that these are causally associated with IPIs.
Only first- and second-born children were included in this study. IPI as a preconceptional “exposure” is defined only for the second children; however, it may be correlated with measures that are relevant for all children or may serve as a marker at the family level of unmeasured factors associated with autism risk. An exploratory analysis of 122 202 third-born children whose previous 2 siblings did not have autism yielded an adjusted OR of 2.09 (95% CI: 1.47–2.97) for IPIs of <12 vs ≥36 months (ORs for 12–23 and 24–35 months did not significantly differ from 1.0), which suggests that an increased risk of autism at shorter IPIs persists for later-born children.
Finally, sibling pairs were included only if they were both born in California, with the same parents, within the time window from 1992 to 2002. These results may not pertain to half-siblings or other geographic locales. If nutritional factors are involved, the association may depend, for example, on background nutritional status of the population, use of supplements, and prevalence of breastfeeding that may vary across populations. If differential recognition is involved, it may depend on medical and educational practices and general awareness among parents. We did not have the data to address these factors in the current analysis.