Perinatal Factors and the Development of Autism: Title and subTitle BreakA Population Study

Autism is a developmental disorder that is characterized by severe impairmentin social interaction and communication and by the presence of stereotypicbehavior.¹ It has an estimated prevalence of10 to 20 per 10 000 individuals.^{2 - 4} Autismis diagnosed by clinical criteria,^{1 ,5} usuallyaround 3 to 4 years of age when social and communication milestones are notachieved. Symptoms are often noticed by 12 months of age,⁶ andaffected people show large individual differences in their symptomatology.⁷ Autism is part of a spectrum of disorders that includesAsperger syndrome and pervasive developmental disorder not otherwise specified(PDD-NOS).¹ The latter 2 diagnoses are madewhen affected individuals have fewer or milder symptoms. The prevalence ofall autism spectrum disorders is estimated at approximately 60 per 10 000people.⁸

The genetic basis of autism spectrum disorders is supported by clusteringin families, higher concordance in monozygotic twins than in dizygotic twins,and evidence for the broader phenotype, or variable expressivity, in otherfamily members.^{9 - 13} Theprecise mechanisms leading to the development of autism or autistic symptomsare unknown, but it is likely that many different genes are involved.¹⁴

In some cases, neurological differences,^{15 - 17} earrotation,¹⁸ and the distinct phalanx ratios¹⁹ suggest that a prenatal influence has acted in adirect or contributory manner. Specific prenatal factors have also been observed.^{20 - 30} Thedevelopment of autistic symptoms may be dependent on the timing of exposureand/or genetic predisposition to certain environmental stimuli before birth.

During the past 30 years, numerous investigations have attempted toidentify a pattern or causal pathway from the obstetric experience of peoplewho develop autism. Generally, an increase in the number of obstetric complicationshas been observed, with the most consistent findings being advanced maternalage, maternal bleeding during early pregnancy, maternal medication use duringpregnancy, and an increased risk of autism among firstborn infants. However,findings are inconsistent and often contradictory because of considerablevariation in methods, sample size, variable selection, analyses, data quality,and control groups. Sample size has typically ranged between 53 and 87 cases,^{31 - 35} withthe larger samples using retrospectively collected obstetric data from medicalrecords and/or maternal interviews.^{36 - 38}

The largest and most recent study compared 408 children identified forautism in the Swedish Inpatient Register with 2040 matched infants from theSwedish Medical Birth Register.³⁹ The autismgroup was characterized by maternal smoking during early pregnancy, a cesareansection delivery, small size for gestational age, low Apgar scores, maternalbirth outside Europe and North America, and congenital malformations. Whilethe Hultman et al³⁹ study is large and rigorous,no comparison was made with siblings, thereby restricting the interpretationof the influence of either genetics or the maternal environment. An increasein obstetric complications has also been observed in some nonautistic siblinggroups,^{31 ,40} suggesting that autismis not necessarily caused by the complications but rather may be an epiphenomenonof a strongly genetic disorder.³⁵

The population of Western Australia is approximately 1.9 million people,73% of whom reside in the metropolitan area of Perth.⁴¹ Forevery birth since 1980, obstetric data, including information on parentaldemographics, maternal obstetric history, pregnancy, delivery, and neonatedata, are recorded in the state Maternal and Child Health Research Database(MCHRDB).⁴² Diagnoses and service deliveryfor people with autism are centralized by the state government, and historically,diagnoses have been made at 5 centers. Diagnoses are made using the criteriaspecified in the DSM for the period.^{1 ,43 - 44} TheGillberg criteria⁴⁵ for Asperger syndrome aresometimes used as a supplement to, or instead of, the DSM criteria for Asperger syndrome because of the perceived value of thedetailed qualitative descriptions they offer for the diagnostic process.

The aim of the present study was to use the centralized resources containedin Western Australia to improve on the methods used in previous studies thathave investigated the correlation between obstetric experience and autismdevelopment. Although no single factor from the prenatal environment has beenidentified to cause autistic symptoms later in life, certain variables, suchas advanced maternal age, have sometimes emerged as risk factors. The methodsused by each study have varied significantly to the point where it is difficultto compare across samples. It is possible that the methodological differences,such as small sample size, have had an effect on the findings.

The present study used a large population-based sample with high-qualitydata and rigorous research methods to identify risk factors from prenataland perinatal experience for developing autism. It included a comparison groupof siblings to measure familial differences and compared diagnostic subgroupswithin the autism spectrum. It is unique for incorporating the largest numberof autistic cases born in a geographic region, using obstetric data collectedat the time of birth, comparing cases with a large population-based controlgroup and comparing cases with siblings.

Approval to obtain diagnostic information was received from the ethicscommittees of the 5 diagnostic centers, as well as the University of WesternAustralia, Perth. Diagnostic reports of all persons born after 1980 identifiedas having an autism spectrum disorder or autistic features before 1997 werepersonally examined by a pediatric registrar (G.C.) involved in autism assessments.Those who had been assessed using the DSM criteriafor an autism spectrum disorder according to the version used in that periodwere included in the study.^{1 ,43 - 44} Thecollection of data by the pediatric registrar was limited by what was containedin the files, so descriptive information, such as IQ, was not available forevery case. Diagnoses made between 1997 and 1999 from the 2 main diagnosticcenters that are responsible for at least 90% of all diagnoses and registrationsand who were subject to centralized diagnoses during this period were added.Since June 1997, for a child to receive government funding for early interventionservices in Western Australia, autism spectrum diagnoses must incorporateassessments by a pediatrician, speech pathologist, and a clinical psychologistwho must all agree on the diagnosis. This ensured a comprehensive listingof people born since 1980, diagnosed between 1986 and 1999.

Cases were electronically linked to the MCHRDB using probabilistic recordlinkage methods⁴⁶ and manual checking to confirmpossible matches. Based on the number of cases reviewed by the pediatric registrar,it was predicted prior to case extraction that approximately 350 autism caseswould be identified. For such a sample size, to detect a 2.2-fold increasein the risk of bleeding during pregnancy (one of the more consistent findingsin previous studies with a prevalence of 3.7% for all births in the MCHRDB)at 80% power and 5% level of significance, 3 control subjects per case wouldbe needed. Five hundred one autism cases were identified and linked to theirbirth record, and sibling data were also extracted. The control group (n =1503) was matched for sex but otherwise randomly selected across the samerange of birth years as the cases (1980-1997). After case ascertainment, itwas apparent that cases born in 1996 and 1997 were few in number (n = 36)because they were diagnosed at an especially young age (younger than 3 years).The pattern of symptoms that appears in very young children with autism maydiffer from that seen at the more prototypic age of 4 or 5 years. Includingthese children may have introduced a bias, particularly because most childrenborn in these birth years who develop autism but were not yet diagnosed wouldhave been omitted. Therefore, these cases were excluded from the study alongwith any control subject or sibling born in the same 2-year period.

The following variables were selected from the MCHRDB:

Analyses were performed using the Statistical Package for the SocialSciences version 10.0 (SPSS Inc, Chicago, Ill). Comparisons were made on eachvariable using χ² tests; odds ratios (ORs) were used for categoricalvariables and t tests for continuous variables. Toinvestigate the possible effect of several variables simultaneously, casesand control subjects were compared using binary logistic regression, and ORsand 95% confidence intervals (CIs) were calculated, with values of P<.05 considered significant.

A total of 465 linked cases was identified (314 autism, 67 Aspergersyndrome, and 84 PDD-NOS) (Table 1).There were 1313 control subjects and 481 siblings of cases. One hundred thirtycases were singletons, 200 had 1 sibling, 97 had 2 siblings, and 30 had 3or more siblings. Eight cases had a sibling who also had been diagnosed withan autism spectrum disorder, including 1 set of concordant male twins, andthese siblings were all included in the case group. Three of the sibling pairsdid not have any other siblings, and the remaining 5 pairs had 1 unaffectedsibling. The random removal of 1 of each of the sibling pairs from the casegroup or the removal of all familial cases from the case group did not alterthe research findings.

Case parents were significantly older than control subject parents.The mean age of case mothers was 28.62 years, and the mean age of controlsubject mothers was 27.01 years. (t = 5.71; P<.001); the mean age of case fathers was 31.74 years,and the mean age of control subject fathers was 30.31 years (t = 4.19; P<.001) (Table 2). Among case pregnancies, 175 mothers (37.6%) had at least1 pregnancy complication recorded compared with 426 control subject mothers(32.4%). Case mothers were more likely to experience a threatened abortionbefore 20 weeks' gestation (OR, 2.41; 95% CI, 1.56-3.73) (Table 3) and significantly more "other" pregnancy complicationsthan mothers of control subjects (OR, 1.36; 95% CI, 1.04-1.78). Case and controlsubject mothers experienced 92 different conditions within the "other" pregnancycategory and had up to 3 conditions listed each. The most common conditions(hypertension, early labor, and fetal/placental problems) were found in similarproportions in both the case and control subject groups. Case mothers hadhigher frequencies of anesthesia use during labor (327 [88.6%] of 369 caseswith data recorded since 1986 compared with 616 [85.2%] of 723 control subjectswith data recorded since 1986), specifically an epidural caudal anesthesia(OR, 1.68; 95% CI, 1.12-2.51).Case mothers were more likely to experienceno labor (OR, 2.04; 95% CI, 1.50-2.76), be induced (OR, 1.43; 95% CI, 1.12-1.83),or have a labor duration of less than 1 hour (89 cases [19.1%]; 155 controlsubjects [11.8%]) (OR, 1.8; 95% CI, 1.3-2.4). More case mothers experienceda labor complication (290 cases [62.4%]; 680 control subjects [51.8%]). Casemothers had greater frequencies of postpartum hemorrhage (OR, 2.33; 95% CI,1.11-4.89) and "other" labor complications (OR, 1.55; 95% CI, 1.24-1.92).There were 120 differently coded "other" labor complications across case andcontrol subject mothers. Case mothers were more likely to have an electivecesarean section (OR, 2.05; 95% CI, 1.49-2.82) or an emergency cesarean section(OR, 1.57; 95% CI, 1.11-2.22).

Control subject mothers had higher parity than the case mothers (P = .01) (Table 4).Among the cases, there were 13 twin pairs, 1 of which was concordant (oneautism diagnosis and one PDD-NOS diagnosis). Of the 12 discordant pairs, 10were diagnosed with autism and 2 with PDD-NOS. Seven twin pairs were the samesex, but information on zygosity was not known from the birth records. Therate of twinning in this sample has been previously described⁴⁸ anddoes not implicate twinning as a risk factor for autism.

Cases were more likely to have experienced fetal distress during labor(OR, 1.64; 95% CI, 1.15-2.34). Apgar scores calculated at 1 minute showedthat significantly more cases achieved a score of 6 or less (54 [19.5%] of277 cases with data recorded since 1991; 66 [12.9%] of 512 control subjectswith data recorded since 1991)(OR, 1.6; 95% CI, 1.1-2.4), and cases were morelikely to have taken more than 1 minute before the onset of spontaneous respiration(OR, 1.4; 95% CI, 1.0-1.9). Apgar scores at 5 minutes showed that fewer casesachieved a score of at least 8 (437 cases [94.2%]; 1258 control subjects [96.1%]),but differences were not significant (OR, 0.7; 95% CI, 0.4-1.1). No differencesin gestational age (including the proportion of premature infants), weightfor gestational age, head circumference, or length were observed between casesand control subjects. Female cases were shorter in length (t = 2.47; P = .01) and had a longer gestationalage (t = −2.73; P =.007) compared with male cases. Female control subjects weighed less (t = 1.98; P = .048) and had alonger gestational age (t = −2.26; P = .02) than male control subjects.

Compared with their siblings, cases were more likely to have been induced(OR, 1.40; 95% CI, 1.03-1.90), experienced fetal distress (OR, 1.64; 95% CI,1.15-2.34), had an Apgar score at 1 minute of 6 or less (OR, 1.64; 95% CI,1.02-2.65), and needed longer than 1 minute to breathe spontaneously (OR,1.81; 95% CI, 1.21-2.69) (Table 5).The cases had a higher rate of threatened abortion than their siblings, butthis was not statistically significant (OR, 1.60; 95% CI, 0.96-2.68). Whenadjusted for sex, these differences remained. The cases did not differ fromtheir siblings on any of the continuous variables tested.

When compared with control subjects, the diagnostic groupings differedin their pattern of obstetric experience (Table 6). The autism group had the greatest number of complications.The PDD-NOS group had similar types of complications to the autism group,but fewer variables reached statistical significance. The Asperger syndromegroup had the fewest obstetric differences and was distinguished by a lackof significantly increased pregnancy and labor complications requiring forcepsor vacuum extraction and by having private health insurance. When comparedwith the autism group on the same variables, the PDD-NOS group only differedby having a greater risk of cephalopelvic disproportion (OR, 3.0; 95% CI,1.3-6.8). Compared with the autism group, the Asperger cases were more likelyto have a vacuum or forceps delivery (OR, 2.5; 95% CI, 1.3-4.7) and have privatehealth insurance (OR, 2.4; 95% CI, 1.3-4.2).

Year of birth and birth order were entered into the regression analysisalong with all other variables. Numerous interactions were also included,but none were significant. After the deletion of 3 individuals with missingvalues, 1775 were available for the analysis (464 cases and 1311 control subjects). Table 7 shows regression coefficients,Wald statistics, ORs, and 95% CIs for ORs for each of the 6 predictors. Themodel predictors using autism as an outcome were year of birth (cases weremore likely to be recently born), birth order (cases were more likely to befirstborn), maternal age (cases were more likely to have an older mother),threatened abortion (more likely to be present in cases), fetal distress (morelikely to be present in cases), and elective cesarean section (more likelyto be present in cases).

To our knowledge, this study is the largest reported population-basedsample to date of people born and diagnosed with an autism spectrum disorderwithin a single geographical area, using prospectively collected obstetricdata and comparing siblings and autism subgroups in a single research design.The findings indicate that individuals diagnosed within the autism spectrumare more likely to have experienced obstetric difficulties during pregnancy,labor, delivery, and the neonatal period compared with people without an autismdiagnosis. This is in agreement with the general interpretations in previousstudies.^{31 - 32 ,34 - 36 ,39 ,49 - 50}

The study sample is representative of the Western Australian populationat a point in time but was restricted to cases who had received a formal diagnosisin the autism spectrum. Therefore, a proportion of cases are likely to havebeen missed from the case ascertainment process, in particular milder and/orolder cases who were not referred for autism services. The study also doesnot include cases who were born between 1980 and 1995 but subsequently diagnosedafter 1999.

Advanced maternal age in case mothers was one of the strongest findingsof the study. Increased maternal age has previously been found as a risk factorin autism, regardless of the range of birth years examined.^{51 - 54} Theeffect in this study was incremental such that younger mothers had the lowestrisk and older mothers had the highest risk. In the Croen et al⁵⁵ studythat compared maternal age of 4356 autism cases born in California with nearly3.5 million control subject mothers, the risk of mothers 35 years or olderhaving a child with autism was increased 3-fold compared with mothers youngerthan 20 years.⁵⁵ This effect was also foundin the present data and in the recent Swedish cohort of 408 cases.³⁹ Older mothers have an increased risk of complicationsduring labor and delivery,^{56 - 57} possiblyattributable to the dysfunction of the uterine muscles and blood supply thatoccurs with age,⁵⁷ which may be compoundedby older mothers delivering a larger number of firstborn cases. However, inthe present study, increased maternal age emerged as an independent risk factorafter adjustment for the other variables.

Paternal age was also greater in fathers of cases compared with fathersof control subjects. Paternal age is not usually addressed or has been foundto be unimportant⁵³ in previous studies. Increasedpaternal age has been documented in patients with schizophrenia.^{58 - 59} Ithas a recognized effect in autosomal dominant mutations such as Apert syndromeand achondroplasia,^{60 - 61} in Downsyndrome,⁶² and also in the inheritance of MeCP2 mutations in girls with Rett syndrome.⁶³ Increasedpaternal age did not emerge as a significant predictor in the regression analysisand was statistically weaker in the univariate analyses, suggesting that itseffect is most likely to be related to increased maternal age.

Threatened abortions were significantly more common in the case pregnanciescompared with control subjects and attained the highest OR value. Bleedingduring pregnancy has previously been reported in case mothers^{32 ,34 ,36 ,39 - 40 ,64} andalso an increased number of prior miscarriages.^{65 - 66} Ahigher prevalence of previous miscarriages and stillbirths among case motherswas not observed in the present dataset.

Elective cesarean sections were more common among case births and theirsiblings compared with control subjects and emerged as a risk factor afteradjusting for maternal age and year of birth. The prevalence of cesarean sectionshas increased in recent decades, the reasons for which are both socially andmedically complex.^{67 - 69} Themost common antecedents for a cesarean section are previous cesarean section,failure to progress in labor, fetal distress, multiple pregnancies, breechpresentation, and unstable lie.^{70 - 71} Recently,there has been an increasing trend for women to request a cesarean delivery,and this is becoming the most popular reason for having the procedure.^{72 - 73} Unfortunately, the reasons for performingcesarean section deliveries are not recorded in the MCHRDB, and thus, inferencescannot be made from the available data.

Epidural caudal anesthesia was more commonly administered to case mothersduring labor and may have been used for cesarean section or maternal painrelief.^{74 - 75} Physiological responsesto pain include increased blood pressure and decreased uterine blood flowand may negatively affect the fetus by altering heart rate and increasingoxygen consumption.^{75 - 76} Epiduralanesthesia is considered the most effective method of pain relief during labor,^{71 ,77} but it is a cause of intrapartumfever and may have adverse effects on the fetus if the fetal core temperaturebecomes elevated above the maternal core temperature.⁷⁶ Adverseeffects such as low Apgar scores may be observed at birth and may also beassociated with administration of epidural caudal anesthesia and the decisionto perform an elective caesarean section in some cases.

In laboratory experiments, cesarean section deliveries can produce ratswith significantly increased binding of dopamine D1 receptors in the nucleusaccumbens compared with rats delivered vaginally.⁷⁸ Theeffect is enhanced by anesthesia and may be related to the presence of hypoxia.The increase in D1-receptor binding is only evident in the rat brain duringadulthood, possibly owing to an interaction with events during development.This may be more severe in rats that are genetically susceptible because thereare different responses among rat strains.⁷⁸ Insome individuals with autism, problem behaviors may be managed effectivelywith dopamine inhibitors.⁷⁹ These individualsmay represent a subgroup whose etiology involves early damage to the ventralbasal ganglia, which may be associated with cerebral changes that are initiatedat birth.

The presence of fetal distress, no labor, caesarean section delivery,and poor condition at birth (as measured by a low Apgar score at 1 minuteand taking greater than 1 minute to spontaneously respirate) was characteristicof the case group. This suggests an association with anoxia and/or placentaldeficiency, evidence of which was found in the Hultman et al³⁹ cohort.Anoxia is often caused by placental abruption that typically results in deliverywithin 1 hour, meconium staining, abnormal fetal heart patterns, and umbilicalcord blood acidity.⁸⁰ There are no data onthe long-term effects of abnormal placentas, reflecting the lack of knowledgeabout its role in development. This is an important direction for future research.The simultaneous presence of anoxia and hyperthermia before birth has beenshown to produce neurological consequences in animal offspring and could thereforebe a potential source of injury to neonates.^{81 - 83}

Although the case group spent less time in special care when comparedwith the statistical mean, both groups were similar using the median statistic.This indicates that some control subjects spent lengthy periods in specialcare, and this has affected the comparison of the 2 groups. Unfortunately,the reasons for requiring special care are not available from the MCHRDB,and thus, no judgments can be made on the qualitative differences in the 2groups' need for care.

Cases were more likely to be firstborn compared with control subjects.The finding is consistent with other reports of first births carrying a riskfor autism.^{31 ,36 ,54} Theearlier finding of autism cases more commonly being born fourth or later^{31 ,34 ,54} was not observed.Those born first or fourth or later naturally have more complications, butthe autism group experienced more complications at all birth orders. A possibleexplanation is that pregnancy complications are part of the phenotypic expressionof some of the group of autism genes rather than a contributory cause.

Overall, case siblings had fewer complications than their affected siblingsbut more than control subjects. They had similar experiences to the casesduring the pregnancy period but experienced less fetal distress during labor,were less likely to be induced than the cases, and were more likely to takemore than 1 minute for spontaneous respiration. These findings remained afteradjusting for sex. Several inferences can be made from this relationship.First, the obstetric complications may represent a compromised maternal environmentto which all offspring are exposed but autism develops in only some offspring,either because of certain environmental stimuli or a particular genotype thatis vulnerable to develop the disorder. Second, if the development of autismdepends on having a certain number of genes, then siblings who share somegenes may also show complications in utero and may also have milder autistictraits. Siblings and other close relatives of people with autism often showmild autistic traits on cognitive assessments.^{9 - 12} Thisis consistent with the hypothesis that shared genes cause similarities inutero and those with the autism genotype will develop the disorder and siblingswho share some of the genes show milder symptomatology.

Alternatively, individuals with autism may react differently to thesame environmental stimuli and may have less tolerance to the prenatal experiencecompared with their siblings. Given that siblings had similar obstetric experiencesto the cases, some other factor could be influential in late pregnancy forthe case births, but not for siblings, to cause autism. It could also be thatthe measurement of complications is too crude and that a more refined measurementmay reveal the subtle differences of the prenatal experience of the childwith autism.

When compared with the control group, the profile of obstetric complicationsdiffered by autism subgroup, whereby the Asperger syndrome group had fewercomplications than the PDD-NOS group, which in turn had fewer, but similar,complications than the autism group. This pattern suggests there is a positiverelationship between autism severity and adverse obstetric experience. TheAsperger syndrome group had a more favorable experience during the pregnancyand labor periods. Previous investigations of the obstetric experience forpeople with Asperger syndrome have used small sample sizes, ranging from 10to 23 cases.^{45 ,84 - 85} Allreport an increase in complications in the Asperger groups compared with thecontrol subjects but fewer than the autism groups, which is similar to thepresent dataset. Fewer complications may be related to the decreased probabilityof people with Asperger syndrome having comorbid medical conditions.⁸⁶ Clinically, people diagnosed with Asperger syndromegenerally have better-developed language skills and a higher level of intelligencethan those with autism.^{2 ,6 ,87 - 88} Autismcases who have a lower IQ have been shown to experience more obstetric complications^{33 ,40 ,64 ,89} andalso more severe symptoms than those with a high IQ.⁹⁰ Itis therefore likely that IQ is a factor that is associated with obstetricexperience.

This research has used a population-based cohort of people diagnosedwith autism spectrum disorders to investigate whether the etiology or natureof autism can be defined by the presence of particular biological characteristicsaround the time of birth. For some cases, differences in the environment and/orin their reaction to it occur in utero, as seen by an increased number ofrisk factors and obstetric complications that differentiate the experienceof autism cases from nonautistic control subjects. The strongest findingswere increased maternal age and a threatened abortion during pregnancy. Itis unlikely that single factors or events cause autistic disorders, althoughit is possible that early nongenetic influences may act on the causal pathwayfor some cases. The observed complications are generally nonspecific and cannotpredict autism development. This research supports the hypothesis that thedevelopment of autism spectrum disorders is dependent on the genotype, andthe presence of complications can be explained by a compromised prenatal experiencefor that genotype.

Corresponding author: Emma Glasson, School of Population Health,University of Western Australia, 35 Stirling Hwy, Crawley WA 6009, Australia(e-mail: emma.glasson@health.wa.gov.au).

Submitted for publication June 2, 2003; final revision received January21, 2004; accepted January 28, 2004.

Data from this study were presented at the Inaugural World Autism Congress:Unity Through Diversity; November 13, 2002; Melbourne, Australia.