The current study is the first meta-analysis of the relationship between perinatal and neonatal factors and risk of autism. Over 60 perinatal and neonatal factors have been studied in relation to autism in 64 epidemiologic studies, of which 40 studies were eligible for meta-analysis. Few of these factors have been examined in multiple well-designed studies. Therefore, attempted replication in methodologically rigorous studies remains necessary for many perinatal and neonatal variables. Most perinatal and neonatal factors examined in multiple studies have shown inconsistent results, and the preponderance of findings overall have not been statistically significant. The factors with the strongest evidence for an association with autism risk included abnormal fetal presentation, umbilical-cord complications, fetal distress, birth injury or trauma, multiple birth, maternal hemorrhage, summer birth, low birth weight, small for gestational age, congenital malformation, low 5-minute Apgar score, feeding difficulties, meconium aspiration, neonatal anemia, ABO or Rh incompatibility, and hyperbilirubinemia. In contrast, the factors with the strongest evidence against a role in autism risk included anesthesia use during delivery, assisted vaginal delivery, postterm birth, high birth weight, and head circumference.
Heterogeneity in effect estimates across studies was observed for several of the risk factors examined, and variability in study design characteristics may have accounted for the observed heterogeneity for many of these variables. There were some potentially important early risk factors for autism that were not found in most individual studies but are highlighted in these stratified analyses, including oxygen resuscitation and respiratory distress at birth.
Although there is insufficient evidence to implicate any 1 perinatal or neonatal factor in autism etiology, the studies using optimality scales provide some evidence to suggest that exposure to multiple neonatal complications may increase autism risk. It also is important to note that the observed association between perinatal and neonatal complications and risk of autism may actually reflect the consequences of previous prenatal complications. As mentioned, we previously published a meta-analysis of prenatal risk factors for autism, in which we found that few prenatal risk factors were associated with the risk of autism. The strongest prenatal factors included advanced maternal and paternal age at birth, maternal gestational bleeding, gestational diabetes, being first born versus third born or later, maternal prenatal medication use, and maternal birth abroad.7
The perinatal and neonatal complications identified in the current analysis may be the result of previous prenatal complications and/or may operate in combination with prior prenatal complications to impact autism risk. Additional research that considers the joint and independent effects of adverse conditions during these various time periods is required to address these possibilities.
This study suggests that several perinatal and neonatal complications may be related to autism risk, either alone, in combination, or perhaps only in those who are genetically vulnerable. However, the correlated occurrence of many of these complications limits the ability to determine which factors, if any, are independently associated with autism. For example, Cesarean deliveries are more common in pregnancies with abnormal fetal presentation, fetal distress, and multiple birth.80,81
Congenital malformations, low birth weight, abnormal presentation, and low Apgar score also are interrelated.36
Most studies did not use multivariate analyses to simultaneously control for all obstetrical factors examined, and a different set of factors was examined in each study. It is possible that increasing rates of some obstetrical factors, such as Cesarean delivery, low birth weight, multiple birth, and neonatal resuscitation, may be contributing factors to the rising prevalence of autism.81
The obstetrical complications that have emerged as significant risk factors for autism in the current meta-analysis suggest a possible role of fetal and neonatal hypoxia. In particular, growth retardation, fetal distress, umbilical-cord wrapping around the neck, low Apgar score, respiratory distress, resuscitation, meconium aspiration, and Cesarean delivery are all potential risk factors that also may be associated with an increased risk of hypoxia.6,24,26,36,38,82
Although some brain abnormalities observed in individuals with autism may reflect a potential role of oxygen deprivation during development, this possibility requires additional examination. Hypoxia also has been shown to increase dopaminergic activity, and there is evidence for dopamine overactivation in autism.83
The current meta-analysis indicates that low birth weight and being small for gestational age are significantly associated with an increased risk of autism. The association with low birth weight was particularly strong among prospective studies. Birth weight is commonly studied as an indicator of fetal growth and development and is affected by many prenatal factors, only some of which may be etiologically relevant. In addition to autism, low birth weight has been associated with a variety of psychiatric, cognitive, and behavioral problems.84–93
Season or month of birth was significantly related to the risk of autism in 6 of 12 studies. Although the seasonal trends varied across studies, March10,23,60,64
were both suggested as birth months associated with an elevated risk of autism. The meta-analysis of season of birth only included a small subset of the studies that examined this association, and it indicated a potential elevated risk of autism associated with summer birth. Although the biological mechanism underlying this association is unclear, a relationship may be caused by seasonal variation in viral or other infections, nutritional factors, or vitamin deficiencies. Therefore, maternal infections and vitamin and nutrient consumption during pregnancy should be examined further in future prospective cohort studies.
Several investigators have questioned the causal nature of the observed relationship between perinatal and neonatal complications and autism. Confounding by birth order has been suggested because prenatal, perinatal, and neonatal complications are more often observed in first-, fourth-, and later-born offspring, and an increased risk of autism has been reported among those who are born first, fourth, and later.50,71
Although some studies have shown that associations were attenuated and no longer significant after adjusting for parity,39,59
have shown that the positive relationship persists. A second noncausal hypothesis is that obstetrical and neonatal complications may occur as a result of the autistic condition in the offspring or as a consequence of other factors (eg, genetic factors) that are the true causal determinants of autism.50
In this epiphenomena explanation, perinatal complications simply reflect the abnormalities of autistic fetal development (autism causes suboptimality) or the same familial factors cause both autism and obstetrical and neonatal complications. The study conducted by Bolton et al50
provided strong evidence in support of the shared-risk hypothesis because there was an association between composite prenatal, perinatal, and neonatal suboptimality and measures of autism severity and familiality, and the suboptimality scores in the cases were highly correlated with that of their affected siblings. In addition, probands with increased prenatal, perinatal, and neonatal complications had more family members with the broader autism phenotype, which was characterized as similar yet milder impairments in language, communication, and social skills or repetitive stereotyped interests and behaviors compared with diagnostic criteria for autism. However, this finding was not replicated in a second study by Zwaigenbaum et al.71
The shared-risk hypothesis also was supported by the findings in the Zwaigenbaum et al study that indicated more composite prenatal, perinatal, and neonatal adversity among unaffected siblings of children with pervasive developmental disorders that had high familial loading for the broader autism phenotype.71
Methodological limitations that may have impaired the precision and validity of the results of the studies in this review include small sample size, nonnormal control groups (eg, Down's syndrome), broad disease definition, and retrospective parental recall of exposures. Of 64 studies included, only 19 had over 80% power to detect an RR of 2 for an exposure with 10% prevalence. A total of 19 studies used broad diagnostic criteria, resulting in the inclusion of case subjects with other autism spectrum disorders, which may limit the ability to detect associations due to etiologic heterogeneity. A total of 21 studies assessed the exposure variables retrospectively, resulting in the high possibility of recall bias. However, the use of medical records also has the limitation of being incomplete. Finally, the majority of studies included only univariate analyses and did not assess potential confounding. These methodological weaknesses also were likely sources of heterogeneity of effects across studies. A possible additional source of heterogeneity of effects across studies was variability in the definitions of risk factors across studies. In fact, risk factor definitions and criteria often were lacking in the manuscripts reviewed. As mentioned, when possible we analyzed both broad variable definitions (eg, abnormal presentation at birth) and narrow variable definitions (eg, breech presentation).
This meta-analysis has a few limitations. First, only published data were used. Second, of 64 studies reviewed, only 40 reported the data necessary for inclusion in the meta-analysis. Within these 40 studies, the investigators did not report the necessary data for a meta-analysis on all factors examined. And although 40 studies were included in the meta-analysis overall, for each factor there generally were fewer than 6 studies included, limiting the statistical power to detect heterogeneity across studies and potential effect modification by study characteristics. Third, because of the rarity of many of the exposures examined and the small sample sizes in many studies, there were several instances of 0 cell counts within studies. The relatively small addition of 0.5 to the cell counts may have had an impact on the overall results because of the small sample sizes. Fourth, a few studies only reported an effect estimate and an indication of whether the results were statistically significant. In these cases, the CIs were estimated on the basis of assumptions regarding the actual P value (P = .05 if significant, P = .50 if not significant). In the case of statistically significant findings, these assumptions resulted in estimated confidence limits that were wider, less precise, and therefore more conservative than might have been expected. Fifth, the tests of publication bias were underpowered because of the limited number of studies in each meta-analysis. Finally, many studies simply examined all available perinatal and neonatal data using designs with methodological weaknesses and without a priori hypotheses or knowledge about reproductive epidemiology. As a result, significant associations observed due to chance are possible in this meta-analysis.
The current review and meta-analysis was not restricted to studies with particular methodological strengths. In addition, individual study characteristics were examined in meta-regressions rather than assigning studies aggregate quality scores. These strategies are consistent with the recommendations proposed by the Meta-Analysis of Observational Studies in Epidemiology Group, which advocated the use of broad inclusion criteria for studies along with regression analyses to relate specific study design characteristics to outcome.94
This maximizes the amount of data available for review. In addition, different methodological considerations are relevant for each exposure. However, the increased probability for heterogeneity of results using the broad inclusion criteria is important to note.