Background
No prospective cohort study of high-risk children has used rigorous exposure assessment and optimal diagnostic procedures to examine the perinatal antecedents of autism spectrum disorder separately among those with and without cognitive impairment.
Objective
We sought to identify perinatal factors associated with increased risk for autism spectrum disorder with and without intellectual disability (intelligence quotient <70) in children born extremely preterm.
Study Design
This prospective multicenter (14 institutions in 5 states) birth cohort study included children born at 23-27 weeks’ gestation in 2002 through 2004 who were evaluated for autism spectrum disorder and intellectual disability at age 10 years. Pregnancy information was obtained from medical records and by structured maternal interview. Cervical-vaginal “infection” refers to maternal report of bacterial infection (n = 4), bacterial vaginosis (n = 30), yeast infection (n = 62), mixed infection (n = 4), or other/unspecified infection (n = 43; eg, chlamydia, trichomonas, or herpes). We do not know the extent to which infection per se was confirmed by microbial colonization. We use the terms “fetal growth restriction” and “small for gestational age” interchangeably in light of the ongoing challenge to discern pathologically from constitutionally small newborns. Severe fetal growth restriction was defined as a birthweight Z-score for gestational age at delivery <–2 (ie, ≥2 SD below the median birthweight in a referent sample that excluded pregnancies delivered for preeclampsia or fetal indications). Participants were classified into 4 groups based on whether or not they met rigorous diagnostic criteria for autism spectrum disorder and intellectual disability (autism spectrum disorder+/intellectual disability–, autism spectrum disorder+/intellectual disability+, autism spectrum disorder–/intellectual disability+, and autism spectrum disorder–/intellectual disability–). Temporally ordered multinomial logistic regression models were used to examine the information conveyed by perinatal factors about increased risk for autism spectrum disorder and/or intellectual disability (autism spectrum disorder+/intellectual disability–, autism spectrum disorder+/intellectual disability+, and autism spectrum disorder–/intellectual disability+).
Results
In all, 889 of 966 (92%) children recruited were assessed at age 10 years, of whom 857 (96%) were assessed for autism spectrum disorder; of these, 840 (98%) children were assessed for intellectual disability. Autism spectrum disorder+/intellectual disability– was diagnosed in 3.2% (27/840), autism spectrum disorder+/intellectual disability+ in 3.8% (32/840), and autism spectrum disorder–/intellectual disability+ in 8.5% (71/840). Maternal report of presumed cervical-vaginal infection during pregnancy was associated with increased risk of autism spectrum disorder+/intellectual disability+ (odds ratio, 2.7; 95% confidence interval, 1.2–6.4). The lowest gestational age category (23-24 weeks) was associated with increased risk of autism spectrum disorder+/intellectual disability+ (odds ratio, 2.9; 95% confidence interval, 1.3–6.6) and autism spectrum disorder+/intellectual disability– (odds ratio, 4.4; 95% confidence interval, 1.7–11). Severe fetal growth restriction was strongly associated with increased risk for autism spectrum disorder+/intellectual disability– (odds ratio, 9.9; 95% confidence interval, 3.3–30), whereas peripartum maternal fever was uniquely associated with increased risk of autism spectrum disorder–/intellectual disability+ (odds ratio, 2.9; 95% confidence interval, 1.2–6.7).
Conclusion
Our study confirms that low gestational age is associated with increased risk for autism spectrum disorder irrespective of intellectual ability, whereas severe fetal growth restriction is strongly associated with autism spectrum disorder without intellectual disability. Maternal report of cervical-vaginal infection is associated with increased risk of autism spectrum disorder with intellectual disability, and peripartum maternal fever is associated with increased risk for intellectual disability without autism spectrum disorder.
Introduction
Meta-analyses and comprehensive reviews describe inconsistencies in research findings on the perinatal antecedents of autism spectrum disorder (ASD) as likely reflecting study differences in ascertainment and diagnostic procedures, sample size, exposure assessment, and treatment of potential confounders. Nevertheless, mounting evidence suggests that a constellation of perinatal factors contribute to increased risk of ASD, including preterm birth, fetal growth restriction (or lower than expected birthweight for gestational age ), and their correlates (eg, placental insufficiency and preeclampsia).
Risk factors for ASD apparently differ between children who have and who do not have co-occurring intellectual impairment (intellectual disability [ID]). Yet, no prospective cohort study of high-risk children has used rigorous exposure assessment and optimal diagnostic procedures to examine the perinatal antecedents of ASD, taking into account co-occurring cognitive impairment. The large Extremely Low Gestational Age Newborn (ELGAN) Study cohort of infants born <28th week of gestation afforded us the opportunity to examine prospectively the antenatal and neonatal antecedents of ASD diagnosed at age 10 years, separately in children with and without co-occurring ID (intelligence quotient [IQ] <70).
Materials and Methods
Participants
The ELGAN Study is a multicenter observational study designed to identify characteristics and exposures associated with increased risk of structural and functional neurologic disorders in extremely preterm infants. During the years 2002 through 2004, women delivering <28 weeks’ gestation at 1 of 14 participating institutions were asked to enroll in the study; 1249 mothers of 1506 infants consented to participate, and 1198 children survived to 10 years (see Supplemental Figure 1 for a flow diagram of study participants). Of 966 children who were actively recruited for follow-up at age 10 years (because of the availability of blood samples from their first postnatal month), informed consent was obtained for the participation of 889 (92%). The institutional review boards of participating institutions approved the study procedures.
Demographic, pregnancy, delivery, and newborn variables
Methods of data collection for demographic, pregnancy, delivery, and newborn variables are described elsewhere, and in detail in the Supplement. In brief, gestational age estimates were based on a hierarchy of the best information available as described in the Supplement (92% were based on fetal ultrasound; most <14 weeks). Cervical-vaginal “infection” refers to maternal report of bacterial infection (n = 4), bacterial vaginosis (n = 30), yeast infection (n = 62), mixed infection (n = 4), or other/unspecified infection (n = 43; eg, chlamydia, trichomonas, or herpes). Previous research indicates that such information gained through self-report can be more accurate than that obtained from medical records or birth certificates, but we do not know the extent to which infection per se was confirmed by microbial colonization. The terms “fetal growth restriction” and “small for gestational age: meet our needs equally, and we use them interchangeably since accurate differentiation of pathologically from constitutionally small newborns is an ongoing challenge. Severe fetal growth restriction was defined by a birthweight Z-score <–2. Birthweight Z-score was calculated as the number of SD each infant’s birthweight was above or below the median birthweight in referent samples that excluded pregnancies delivered for preeclampsia or fetal indications. Physiology, laboratory, and therapy data for the first 12 postnatal hours were collected to calculate a Score for Neonatal Acute Physiology-II. Additional data were collected on placenta microbiology and histology, mode of ventilation and respiratory care, bacteremia, patent ductus arteriosis, illnesses and medications used in the first 28 days postpartum, necrotizing enterocolitis, and retinopathy.
Assessment at 10 years of age
The assessment procedures, and all relevant test scores for ASD and ID, are reported in a prior publication. Briefly, diagnostic assessment of ASD was conducted with 3 well-validated measures, administered sequentially. First was the Social Communication Questionnaire (SCQ) with a screen-in score ≥11 to increase sensitivity relative to the standard criterion score of ≥15. Children who met the SCQ criterion were then assessed with the Autism Diagnostic Interview–Revised (ADI-R). All children who met ADI-R criteria for autism or ASD, or who had a prior clinical diagnosis of ASD and/or exhibited symptoms of ASD during cognitive testing according to the site psychologist were then assessed with the Autism Diagnostic Observation Schedule, Second Version (ADOS-2) –the criterion measure of ASD in this study.
All ADOS-2 administrations were independently scored by a second rater with autism diagnostic and ADOS-2 expertise (R.M.J.) who did not have knowledge of the child’s SCQ and ADI-R results or clinical history. In cases of scoring disagreements, consensus was reached between raters. Item-by-item interrater agreement for the 14 ADOS-2 diagnostic algorithm scores was on average .93 (SD = .12). Of 90 ADOS-2 assessments, interrater disagreement and consensus scoring resulted in 4 changes of classification, 3 from non-ASD to ASD and 1 from ASD to non-ASD (Cohen kappa = .90).
Intellectual ability (IQ) was assessed with the School-Age Differential Ability Scales-II. Children with IQ [(verbal + nonverbal reasoning scores)/2] <70 were classified as having ID. Because ASD cannot be validly diagnosed in children with significant visual and/or motor impairment accompanied by severe ID, children with these conditions were excluded from diagnostic consideration of ASD ( Supplemental Figure 1 ). Severe gross motor dysfunction was defined as level 5 (ie, no self-mobility) on the Gross Motor Function Classification System. A child was considered to have severe visual impairment if the parent reported uncorrectable functional blindness in both eyes. No participant had a significant, uncorrected hearing impairment.
Data analyses
We evaluated the null hypothesis that ASD+/ID–, ASD+/ID+, and ASD–/ID+ are not associated with any maternal, pregnancy, delivery, or postnatal characteristic or exposure. We began by classifying children into 4 groups based on whether or not they met diagnostic criteria for ASD and ID at age 10 years. We then described the percent of children in each group whose mother had selected demographic and pregnancy characteristics or who themselves had perinatal and neonatal characteristics and exposures and who were diagnosed with ASD and/or ID at age 10 years.
Because antepartum phenomena can influence postnatal phenomena, we tested our null hypothesis with temporally oriented models. Primary exposures included inflammation-related phenomena during pregnancy (eg, maternal report of cervical-vaginal infection) and at delivery (eg, intrapartum maternal fever), indicators of fetal growth restriction and its correlates (eg, birthweight Z-score for gestational age <–2 and preeclampsia), and lowest gestational age category (ie, 23-24 weeks). We considered variables as confounders if identified in the literature or if in our data they were associated with both the exposure and the outcome with probabilities ≤.25. To construct the time-oriented models, we used a step-down procedure seeking a parsimonious solution without effect modification terms. First, we examined pregnancy information in a multinomial logistic regression model of risk for ASD and ID. Then we added factors measured around the time of delivery, adjusting for those variables with statistically significant associations in the pregnancy-stage model. Finally, we added neonatal factors, adjusting for all variables selected in by the earlier models.
We present magnitudes of association as odds ratios (OR) with 95% confidence intervals (CI). Associations were statistically significant when the 95% CI did not include the null estimate (ie, OR, 1.0). The primary outcomes ASD+/ID– and ASD+/ID+ affected 3-4% of our sample of ELGANs, giving us 80% power to detect associations with a minimal detectable OR of 3.2, assuming an exposure prevalence of 0.3. We also describe the prevalence and antecedents of ID without ASD as a secondary outcome.
Materials and Methods
Participants
The ELGAN Study is a multicenter observational study designed to identify characteristics and exposures associated with increased risk of structural and functional neurologic disorders in extremely preterm infants. During the years 2002 through 2004, women delivering <28 weeks’ gestation at 1 of 14 participating institutions were asked to enroll in the study; 1249 mothers of 1506 infants consented to participate, and 1198 children survived to 10 years (see Supplemental Figure 1 for a flow diagram of study participants). Of 966 children who were actively recruited for follow-up at age 10 years (because of the availability of blood samples from their first postnatal month), informed consent was obtained for the participation of 889 (92%). The institutional review boards of participating institutions approved the study procedures.
Demographic, pregnancy, delivery, and newborn variables
Methods of data collection for demographic, pregnancy, delivery, and newborn variables are described elsewhere, and in detail in the Supplement. In brief, gestational age estimates were based on a hierarchy of the best information available as described in the Supplement (92% were based on fetal ultrasound; most <14 weeks). Cervical-vaginal “infection” refers to maternal report of bacterial infection (n = 4), bacterial vaginosis (n = 30), yeast infection (n = 62), mixed infection (n = 4), or other/unspecified infection (n = 43; eg, chlamydia, trichomonas, or herpes). Previous research indicates that such information gained through self-report can be more accurate than that obtained from medical records or birth certificates, but we do not know the extent to which infection per se was confirmed by microbial colonization. The terms “fetal growth restriction” and “small for gestational age: meet our needs equally, and we use them interchangeably since accurate differentiation of pathologically from constitutionally small newborns is an ongoing challenge. Severe fetal growth restriction was defined by a birthweight Z-score <–2. Birthweight Z-score was calculated as the number of SD each infant’s birthweight was above or below the median birthweight in referent samples that excluded pregnancies delivered for preeclampsia or fetal indications. Physiology, laboratory, and therapy data for the first 12 postnatal hours were collected to calculate a Score for Neonatal Acute Physiology-II. Additional data were collected on placenta microbiology and histology, mode of ventilation and respiratory care, bacteremia, patent ductus arteriosis, illnesses and medications used in the first 28 days postpartum, necrotizing enterocolitis, and retinopathy.
Assessment at 10 years of age
The assessment procedures, and all relevant test scores for ASD and ID, are reported in a prior publication. Briefly, diagnostic assessment of ASD was conducted with 3 well-validated measures, administered sequentially. First was the Social Communication Questionnaire (SCQ) with a screen-in score ≥11 to increase sensitivity relative to the standard criterion score of ≥15. Children who met the SCQ criterion were then assessed with the Autism Diagnostic Interview–Revised (ADI-R). All children who met ADI-R criteria for autism or ASD, or who had a prior clinical diagnosis of ASD and/or exhibited symptoms of ASD during cognitive testing according to the site psychologist were then assessed with the Autism Diagnostic Observation Schedule, Second Version (ADOS-2) –the criterion measure of ASD in this study.
All ADOS-2 administrations were independently scored by a second rater with autism diagnostic and ADOS-2 expertise (R.M.J.) who did not have knowledge of the child’s SCQ and ADI-R results or clinical history. In cases of scoring disagreements, consensus was reached between raters. Item-by-item interrater agreement for the 14 ADOS-2 diagnostic algorithm scores was on average .93 (SD = .12). Of 90 ADOS-2 assessments, interrater disagreement and consensus scoring resulted in 4 changes of classification, 3 from non-ASD to ASD and 1 from ASD to non-ASD (Cohen kappa = .90).
Intellectual ability (IQ) was assessed with the School-Age Differential Ability Scales-II. Children with IQ [(verbal + nonverbal reasoning scores)/2] <70 were classified as having ID. Because ASD cannot be validly diagnosed in children with significant visual and/or motor impairment accompanied by severe ID, children with these conditions were excluded from diagnostic consideration of ASD ( Supplemental Figure 1 ). Severe gross motor dysfunction was defined as level 5 (ie, no self-mobility) on the Gross Motor Function Classification System. A child was considered to have severe visual impairment if the parent reported uncorrectable functional blindness in both eyes. No participant had a significant, uncorrected hearing impairment.
Data analyses
We evaluated the null hypothesis that ASD+/ID–, ASD+/ID+, and ASD–/ID+ are not associated with any maternal, pregnancy, delivery, or postnatal characteristic or exposure. We began by classifying children into 4 groups based on whether or not they met diagnostic criteria for ASD and ID at age 10 years. We then described the percent of children in each group whose mother had selected demographic and pregnancy characteristics or who themselves had perinatal and neonatal characteristics and exposures and who were diagnosed with ASD and/or ID at age 10 years.
Because antepartum phenomena can influence postnatal phenomena, we tested our null hypothesis with temporally oriented models. Primary exposures included inflammation-related phenomena during pregnancy (eg, maternal report of cervical-vaginal infection) and at delivery (eg, intrapartum maternal fever), indicators of fetal growth restriction and its correlates (eg, birthweight Z-score for gestational age <–2 and preeclampsia), and lowest gestational age category (ie, 23-24 weeks). We considered variables as confounders if identified in the literature or if in our data they were associated with both the exposure and the outcome with probabilities ≤.25. To construct the time-oriented models, we used a step-down procedure seeking a parsimonious solution without effect modification terms. First, we examined pregnancy information in a multinomial logistic regression model of risk for ASD and ID. Then we added factors measured around the time of delivery, adjusting for those variables with statistically significant associations in the pregnancy-stage model. Finally, we added neonatal factors, adjusting for all variables selected in by the earlier models.
We present magnitudes of association as odds ratios (OR) with 95% confidence intervals (CI). Associations were statistically significant when the 95% CI did not include the null estimate (ie, OR, 1.0). The primary outcomes ASD+/ID– and ASD+/ID+ affected 3-4% of our sample of ELGANs, giving us 80% power to detect associations with a minimal detectable OR of 3.2, assuming an exposure prevalence of 0.3. We also describe the prevalence and antecedents of ID without ASD as a secondary outcome.
Results
Of the 996 children recruited to participate at age 10 years, 889 (92%) were enrolled, of whom 840 (95%) were assessed both for ASD and for ID ( Supplemental Figure 1 ). Of the 840 children in our final sample, 7.0% (n = 59) met study criteria for ASD and 12.3% (n = 103) had ID (IQ <70). ASD+/ID– was diagnosed in 3.2% (27/840), ASD+/ID+ in 3.8% (32/840), and ASD–/ID+ in 8.5% (71/840) of the final sample.
Descriptive univariate analyses
The results of univariate analyses are displayed visually using the same format in Figures 1 to 4 , and Supplemental Figure 2 to 5 , in which a legend at the top of each figure names the 4 symbols used to describe each of the 4 mutually exclusive ASD/ID outcome groups. The horizontal axis labeled at the bottom of each figure indicates row percent (ie, each row sums to 100%), and the characteristic that each row plotted percent describes is labeled on the left vertical axis. The top row of each figure displays 4 symbols to indicate the cohort prevalence of each of the 4 ASD/ID groups; dotted vertical lines proceeding downward from each symbol through the plot are provided to enable easy visual comparison between cohort prevalence and each plotted percent (ie, prevalence among study groups formed according to pregnancy, birth, and neonatal characteristics). The total number of children who had or were exposed to each characteristic is shown on the right vertical axis. We describe maternal demographic characteristics in Figure 1 ; maternal exposures, illnesses, and medications in Figure 2 ; pregnancy characteristics and complications in Figure 3 ; and newborn characteristics in Figure 4 . Supplemental Figures 2 to 5 illustrate placenta characteristics, early postnatal characteristics, newborn medications and therapies, and newborn diagnoses and dysfunctions. The accompanying legends provide brief descriptions of the distribution of ASD/ID study groups shown in each figure to represent the entire study population.
Analytic multivariable regression analyses
Time-ordered multinomial logistic regression models ( Table ) were used to examine the extent to which pregnancy, delivery, and neonatal factors are associated with increased risk for ASD+/ID–, ASD+/ID+, or ASD–/ID+, adjusting for potential confounders.
