Association of Epidural Analgesia During Labor and Delivery With Autism Spectrum Disorder in Offspring

Key Points

Question  Does use of epidural analgesia during labor and delivery increase the risk for autism spectrum disorder in offspring?

Findings  In this population-based retrospective cohort study that included 388 254 term singleton children born via vaginal delivery, the use of epidural analgesia during labor and delivery was significantly associated with a small increase in risk of autism spectrum disorder in offspring (1.53% in exposed children vs 1.26% in unexposed children; adjusted hazard ratio, 1.09).

Meaning  Maternal epidural analgesia use during labor and delivery was associated with a small increase in the risk of autism spectrum disorder in offspring that met the threshold for statistical significance; however, given the likelihood of residual confounding that may account for the results, these findings do not provide strong supporting evidence for this association.

Importance  Evidence from studies investigating the association of epidural analgesia use during labor and delivery with risk of autism spectrum disorder (ASD) in offspring is conflicting.

Objective  To assess the association of maternal use of epidural analgesia during labor and delivery with ASD in offspring using a large population-based data set with clinical data on ASD case status.

Design, Setting, and Participants  This population-based retrospective cohort study included term singleton children born in British Columbia, Canada, between April 1, 2000, and December 31, 2014. Stillbirths and cesarean deliveries were excluded. Clinical ASD diagnostic data were obtained from the British Columbia Autism Assessment Network and the British Columbia Ministry of Education. All children were followed up until clinical diagnosis of ASD, death, or the study end date of December 31, 2016.

Exposures  Use of epidural analgesia during labor and delivery.

Main Outcomes and Measures  A clinical diagnosis of ASD made by pediatricians, psychiatrists, and psychologists with specialty training to assess ASD. Cox proportional hazards models were used to estimate the hazard ratio of epidural analgesia use and ASD. Models were adjusted for maternal sociodemographics; maternal conditions during pregnancy; labor, delivery, and antenatal care characteristics; infant sex; gestational age; and status of small or large for gestational age. A conditional logistic regression model matching women with 2 births or more and discordance in ASD status of the offspring also was performed.

Results  Of the 388 254 children included in the cohort (49.8% female; mean gestational age, 39.2 [SD, 1.2] weeks; mean follow-up, 9.05 [SD, 4.3] years), 5192 were diagnosed with ASD (1.34%) and 111 480 (28.7%) were exposed to epidural analgesia. A diagnosis of ASD was made for 1710 children (1.53%) among the 111 480 deliveries exposed to epidural analgesia (94 157 women) vs a diagnosis of ASD in 3482 children (1.26%) among the 276 774 deliveries not exposed to epidural analgesia (192 510 women) (absolute risk difference, 0.28% [95% CI, 0.19%-0.36%]). The unadjusted hazard ratio was 1.32 (95% CI, 1.24-1.40) and the fully adjusted hazard ratio was 1.09 (95% CI, 1.00-1.15). There was no statistically significant association of epidural analgesia use during labor and delivery with ASD in the within-woman matched conditional logistic regression (839/1659 [50.6%] in the exposed group vs 1905/4587 [41.5%] in the unexposed group; fully adjusted hazard ratio, 1.07 [95% CI, 0.87-1.30]).

Conclusions and Relevance  In this population-based study, maternal epidural analgesia use during labor and delivery was associated with a small increase in the risk of autism spectrum disorder in offspring that met the threshold for statistical significance. However, given the likelihood of residual confounding that may account for the results, these findings do not provide strong supporting evidence for this association.

Epidural analgesia is commonly used for pain relief during labor and delivery, with rates of use reported to be as high as 57% in Canada and higher than 70% in the US in 2015 (with significant variations across provinces and states).^1,2 Epidural analgesia has a good safety profile and appears to relieve labor pain more effectively than opioid medications. However, few studies have examined long-term outcomes associated with epidural analgesia.³ Two studies reported conflicting results on the association of epidural analgesia during labor and delivery with risk for autism spectrum disorder (ASD) in children.^4,5 Given that millions of newborns are exposed to epidural analgesia annually, determining associations between this exposure and ASD is important.

The prevalence of ASD has increased from 1 in 150 in 2000 to 1 in 54 children in 2016 in the US.⁶ Both genetic and environmental factors and interactions between genes and the environment appear to contribute to ASD risk,^7,8 and research is increasingly focused on identifying environmental exposures that increase risk. Although meta-analyses have found an association of gestational diabetes,⁹ hypertensive disorders of pregnancy,¹⁰ and a group of conditions reflecting compromised neonatal health with ASD,¹¹ the reported point estimates are inconsistent.¹²

Given the challenges inherent in this observational research, the methodological requirements to improve the understanding of potential associations include: (1) the use of high-quality data on ASD case status and exposure to epidural analgesia; (2) access to data on confounders between the exposure groups; and (3) the ability to control for unmeasured confounding. To this end, this population-based study from British Columbia, Canada, examined the association of epidural analgesia use during labor and delivery with ASD in offspring using detailed antenatal care and labor and delivery data linked with ASD diagnoses made by practitioners trained to assess ASD.

We conducted a retrospective cohort study of all women who delivered a live singleton term infant in the province of British Columbia, Canada (population of approximately 4.6 million) between April 1, 2000, and December 31, 2014. All children were followed up until a clinical diagnosis of ASD, death, the study end date of December 31, 2016, or until they moved outside British Columbia and were lost to follow-up. Population Data BC accessed data from the British Columbia Perinatal Data Registry,¹³ the British Columbia Central Demographics File (formerly known as the Consolidation File),¹⁴ and the British Columbia Vital Statistics Agency¹⁵ and linked the data with the British Columbia Autism Assessment Network.¹⁶ We obtained ethics approval from the UBC Children’s and Women’s Research Ethics Board. Approval by the ethics board and the British Columbia data stewards for use of deidentified administrative data files includes a waiver of informed consent from participants.

The British Columbia Perinatal Data Registry is a province-wide registry that includes information on antenatal, intrapartum, and postpartum maternal and infant care and outcomes for nearly 100% of deliveries regardless of location of delivery (hospital or elsewhere). These data provide details on antenatal care and the labor and delivery episode, including the use of epidural analgesia. We included term singleton children born via vaginal delivery (Figure). We excluded children who were assessed for ASD before aged 2 years because of concern that assessment of children at younger ages may reflect data entry errors. Children who did not have 2 full years of follow-up either due to death or because they moved outside the province before their second birthday also were excluded.

The British Columbia Perinatal Data Registry includes several data fields that are specific to pain relief administered during the labor and delivery episode. This includes the use of epidural analgesia and use of other forms of analgesia and anesthesia during labor and delivery.

In British Columbia, primary care practitioners and pediatricians conduct general developmental surveillance and field concerns from parents. If there is reason to warrant an assessment for ASD, and the family opts to proceed with a publicly funded ASD assessment, the child is referred to the British Columbia Autism Assessment Network for a clinical evaluation made by pediatricians, psychiatrists, and psychologists who have completed additional structured training for the diagnosis of ASD. Diagnostic assessment for ASD has been standardized within the British Columbia Autism Assessment Network since 2004 and must be informed by 2 instruments, the Autism Diagnostic Observation Schedule (ADOS) or the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2) and the Autism Diagnostic Interview–Revised (ADI-R).

The assessment also draws on clinical history, evaluation of developmental status, and reports from schools and other relevant clinicians. Private practitioners who undergo additional training also diagnose some children in British Columbia and also are required to use the ADOS or ADOS-2 and the ADI-R for children to receive funding through the Ministry of Child and Family Development. We identified cases diagnosed by private practitioners through linkage to data held by the British Columbia Ministry of Education.

We had access to the following sociodemographic variables: maternal age, co-parent age, neighborhood income quintile, and community size. We also had data on maternal conditions that could have influenced fetal health, including preexisting diabetes, gestational diabetes, parity, whether the woman smoked during the pregnancy, prepregnancy body mass index (BMI; calculated as weight in kilograms divided by height in meters squared), and pregnancy-induced hypertension and other types of hypertension. We examined labor and delivery and antenatal care characteristics, including the year of delivery, whether labor was induced or augmented, whether antibiotics were administered, and length of the first stage of labor.

For neonatal characteristics, we examined gestational age, whether there was a congenital anomaly, sex, and whether the infant was small or large for gestational age. Infants were considered small for gestational age if they were below the 10th percentile and large for gestational age if they were above the 90th percentile of birth weight for their gestational age and sex based on the population-based charts of Kierans et al,¹⁷ which were created in British Columbia.

Maternal sociodemographic characteristics, pregnancy-related conditions and risk factors, labor and delivery characteristics, neonatal characteristics, and covariates were compared using the mean standardized difference according to epidural analgesia use. A difference between covariates was considered important if the mean standardized difference was greater than 0.1.¹⁸

The association of maternal use of epidural analgesia during labor and delivery with ASD was modeled using Cox proportional hazards models. Children were censored at the time of death or withdrawal from British Columbia’s universal health insurance because this likely reflected a move out of the province. Clustering between multiple deliveries to the same mother during the study period was accounted for using robust standard error estimates. To understand which covariates were important confounders in the relationship between epidural analgesia and ASD, a series of regression analyses were performed, sequentially increasing the degree of control for potential confounders.

All continuous variables were entered into the model continuously. Crude relationships are presented first. Model 1 adjusted for maternal sociodemographic characteristics, which included age, co-parent age, community size, neighborhood income quintile, and year of birth. Model 2 additionally controlled for the pregnancy conditions and risk factors of gestational diabetes, preexisting diabetes, parity, smoking during pregnancy, prepregnancy BMI, pregnancy-induced hypertension, and other types of hypertension (preexisting hypertension, hypertensive kidney disease, high blood pressure, and proteinuria). Model 3 additionally controlled for induction of labor, gestational age, sex, small or large for gestational age at birth, and whether the infant had a congenital anomaly. Augmentation of labor and administration of antibiotics were not included because the timing of these interventions may not precede the epidural analgesia. For BMI, due to a high degree of missingness, multiple imputation was used. For all other variables, multiple imputation was not used, and the data were assumed missing at random.

To further evaluate the effect of potential confounders, several additional sensitivity analyses were examined using Cox proportional hazards models, including uncomplicated deliveries, wherein we excluded any infants with congenital anomalies, those who were small or large for gestational age at birth, any infants with Apgar scores of less than 7 at 5 minutes, and those admitted to the neonatal intensive care unit (NICU). Given the increasing use of epidural analgesia during the study period, and because the median age at assessment for children seen at the British Columbia Autism Assessment Network was 5.35 years, an analysis including only children with at least 6 years of follow-up also was conducted. For the main and sensitivity analyses, the proportional hazards assumptions were assessed after fitting all the models using Schoenfeld residuals for nonzero slope. We found no evidence that this assumption was violated.

Further analyses were conducted in a sibling cohort nested within the larger cohort. Unconditional logistic regression models restricted to women who had more than 1 delivery during the study period were performed using robust standard errors to account for clustering related to the inclusion of more than 1 pregnancy in the same woman. This was done to determine whether the cohort of siblings was different enough from the total cohort to change the direction and effect size of the associations between epidural analgesia and ASD. Next, we used conditional logistic regression to estimate the association of epidural analgesia with risk of ASD within the same woman (ie, a matched design where a woman is matched to herself in successive deliveries). This approach controlled for characteristics that do not change or remain highly stable over time within a woman such as genetics and other predisposing factors for ASD. The conditional logistic regression selects discordant pregnancies; thus, only women who were discordant for the ASD status of at least 2 of their children (meaning ≥1 child was diagnosed with ASD and ≥1 other child was not diagnosed with ASD) contributed to this model.

All P values are 2-sided. Statistical significance was defined as P < .05 for all analyses. Because of the potential for type I error due to multiple comparisons, findings for the secondary analyses should be interpreted as exploratory. All analyses were performed using Stata version 16.0 (StataCorp).

The initial cohort included 599 527 singleton births in British Columbia among 386 192 women. After exclusions were made as outlined in the Figure, the main cohort included 388 254 vaginal deliveries to 258 472 women. Of these deliveries, 111 480 infants (28.71%) were born to women who used epidural analgesia during labor and delivery. The children were followed up for a mean of 9.05 years (SD, 4.3 years). There was a high degree of missingness for prepregnancy BMI (109 735 of 388 254 deliveries [28.26%]).

There were important differences between deliveries by exposure status to epidural analgesia for community size, parity, rates of other types of hypertension and pregnancy-induced hypertension, year of birth, induction and augmentation of labor, receipt of HIV and group B streptococcal tests, administration of antibiotics during labor, and NICU admission (Table 1). The most pronounced between-group difference was for the length of the first stage of labor with women who used epidural analgesia having a longer first stage of labor (8.5 hours) vs those who did not (4.4 hours) (between-group difference, 4.1 hours [mean standardized difference, 0.82]).

By the end of follow-up on December 31, 2016, there were 5192 children (1.34%) who had been diagnosed with ASD. Of all children diagnosed with ASD, 72% were diagnosed by clinicians in the British Columbia Autism Assessment Network. The other 28% were diagnosed by private practitioners in the province. The proportion of children diagnosed with ASD was 1.53% in the group that were exposed to epidural analgesia and 1.26% in the group that were not exposed to epidural analgesia (absolute risk difference, 0.28% [95% CI, 0.19%-0.36%]; Table 2).

In the unadjusted analysis, the hazard ratio (HR) was 1.32 (95% CI, 1.24-1.40) for the association of ASD with exposure to epidural analgesia. Epidural analgesia continued to be statistically significantly associated with increased risk for ASD in model 1 (adjusting for year of birth, maternal and co-parent age, neighborhood income quintile, and community size) (adjusted HR, 1.30 [95% CI, 1.22-1.38]) but attenuated following further adjustment in model 2 for gestational diabetes, preexisting diabetes, pregnancy-induced hypertension, other hypertension, parity, smoking during pregnancy, and BMI (adjusted HR, 1.12 [95% CI, 1.05-1.20]). In model 3, after additional adjustment for induction of labor, gestational age, sex, status of small or large for gestational age at birth, and congenital anomaly, there was further attenuation (adjusted HR, 1.09 [95% CI, 1.00-1.15]).

Table 3 presents the results from the analyses performed for the sibling cohort and the conditional logistic regressions that match a woman to herself across her deliveries. The unmatched results from the sibling cohort were consistent in terms of direction and effect size with the HRs presented in Table 2 (absolute risk difference, 0.36% [95% CI, 0.25%-0.48%] and adjusted HR for model 3, 1.10 [95% CI, 0.99-1.20]). The within-woman sibling matched analysis further attenuated this relationship (absolute risk difference, 9.04% [95% CI, 6.25%-11.84%] and adjusted HR for model 3, 1.07 [95% CI, 0.87-1.30]).

The sensitivity analyses were consistent with the results from the overall cohort (eTable in the Supplement). A similarly small but statistically significant association was observed when restricting the analysis to uncomplicated deliveries (n = 322 947) by eliminating any infants with a congenital anomaly, those who were small or large for gestational age, those who had an Apgar score of less than 7 at 5 minutes, and those who were admitted to the NICU; of the 91 076 infants in the group exposed to epidural analgesia, there were 1307 (1.44%) diagnosed with ASD and of the 231 871 infants in the unexposed group, there were 2717 (1.17%) diagnosed with ASD (risk difference, 0.26% [95% CI, 0.17%-0.35%] and adjusted HR for model 3, 1.11 [95% CI, 1.03-1.19]). Among children with at least 6 years of follow-up (n = 275 069), there were 1381 (1.85%) diagnosed with ASD in the group exposed to epidural analgesia (n = 74 679) and 3069 (1.53%) diagnosed with ASD in the unexposed group (n = 200 390); there was no statistically significant association between exposure to epidural analgesia and ASD (absolute risk difference, 0.32% [95% CI, 0.21%-0.43%] and adjusted HR for model 3, 1.03 [95% CI, 0.96-1.11]).

In this population-based cohort in British Columbia, use of epidural analgesia during labor and delivery was associated with a small, statistically significant increase in risk of ASD in children after adjusting for sociodemographic, maternal conditions, and risk factors during pregnancy and after adjusting for labor and delivery and neonatal characteristics. However, the interpretation of these findings is limited by the possibility of residual confounding, and the study does not provide strong supporting evidence for this association. Moreover, there was no statistically significant association between epidural analgesia and ASD in the within-mother matched analysis or the sibling cohort.

The statistically significant increased risk for ASD among those who were exposed to epidural analgesia during labor and delivery reported in this study is consistent with a recently published study by Qiu et al⁴ with a cohort of 147 895 live births in California. However, that study⁴ reported that epidural analgesia was associated with an HR of 1.37, whereas the current study found an HR of only 1.09 and an absolute risk difference of 0.28%. Qiu et al⁴ also reported a dose-response relationship between epidural analgesia use and ASD.⁴ A population-based study by Wall-Wieler et al⁵ reported no statistically significant association between epidural analgesia and ASD after controlling for confounders in a cohort of 123 175 births in the Canadian province of Manitoba. However, that study⁵ controlled for fetal distress, augmentation of labor, and labor dystocia, which are all possible mediators in the relationship between epidural analgesia use and ASD, and this limits the interpretation of the nonsignificant association between epidural analgesia and ASD.

Alternatively, the association reported by Qiu et al,⁴ and in this study, may reflect residual confounding related to differences in the underlying maternal, perinatal, and genetic factors. The data reported in the current study illustrate that women who chose epidural analgesia during labor and delivery were very different from women who did not. Their labor experiences also were different—they were longer and there was increased use of induction, augmentation, and antibiotics. When using a woman as her own control to account for unmeasured factors that are more difficult to quantify but that tend to be relatively stable within women over time, the relationship between use of epidural analgesia and ASD was attenuated and no longer statistically significant.

This study is strengthened by linkage with the British Columbia Autism Assessment Network, which increases the quality of the ASD data. In the absence of access to documented clinical diagnosis of ASD, many epidemiological studies^9,13 have used diagnostic codes from physician billing records, as was the case for the studies by Qiu et al⁴ and Wall-Wieler et al.⁵ Recent research has suggested that even though using diagnostic codes can produce a good positive predictive value, the negative predictive value and C-statistics show that diagnostic codes cannot be reliably used to identify cases of ASD.¹⁹ Data on epidural analgesia use came from the British Columbia Perinatal Data Registry, a registry that has been shown to be a valid and reliable source of data on interventions provided during labor and delivery.²⁰

This study has several limitations. First, given the observational design, there is certain to be some residual confounding. Given that the lower confidence limit for the HR was 1.00, and that the absolute risk difference was very small, even a small amount of residual confounding could account for the statistically significant findings. There were important differences between deliveries in which epidural analgesia was used and not used, particularly with respect to variables that suggest a worse prognosis for women exposed to epidural analgesia. This increases the probability that residual confounding might explain the association between epidural analgesia use and ASD. For example, women with labor and delivery experiences in which epidural analgesia was used had a higher prevalence of hypertension (both pregnancy-induced and other forms), an increased length of the first-stage of labor, more induction and augmentation of labor, increased administration of antibiotics during labor, and a higher likelihood of the infant being admitted to the NICU than deliveries in which epidural analgesia was not used. The study should therefore not be interpreted as providing strong evidence of an association of maternal epidural analgesia use during labor and delivery with risk of ASD. Second, there was a lack of detailed information on the dose of epidural analgesia to directly address the dose-response relationship.

In this population-based study, maternal epidural analgesia use during labor and delivery was associated with a small increase in the risk of autism spectrum disorder in offspring that met the threshold for statistical significance. However, given the likelihood of residual confounding that may account for the results, these findings do not provide strong supporting evidence for this association.

Corresponding Author: Gillian E. Hanley, PhD, Vancouver General Hospital Research Pavilion, 828 W 10th Ave, Room 590, Vancouver, BC V5Z 1M9, Canada (gillian.hanley@vch.ca).

Accepted for Publication: August 17, 2021.

Author Contributions: Dr Hanley had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Hanley, N. Lanphear, Zwaigenbaum, Oberlander.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Hanley, N. Lanphear, Oberlander.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Hanley, Bickford.

Obtained funding: Hanley, N. Lanphear, Zwaigenbaum, Oberlander.

Administrative, technical, or material support: Hanley, Ip, N. Lanphear, B. Lanphear, Weikum, Oberlander.

Supervision: Hanley, N. Lanphear.

Conflict of Interest Disclosures: Dr Ip reported receiving funding from the Sunny Hill Foundation for Children. Dr Weikum reported receiving a portion of her salary from the Sunny Hill Foundation for Children. No other disclosures were reported.

Funding/Support: This study was funded by an operating grant from the Canadian Institutes of Health Research and was funded by the Sunny Hill Foundation for Children (part of the BC Children’s Hospital Foundation). Dr Hanley’s work was funded by a new investigator award from the Canadian Institutes of Health Research and a scholar award from the Michael Smith Foundation for Health Research.

Role of the Funder/Sponsor: The funders/sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: All inferences, opinions, and conclusions drawn are those of the authors and do not reflect the opinions or policies of the data stewards (the British Columbia Ministry of Health, the British Columbia Perinatal Data Registry, the British Columbia Central Demographics File, the British Columbia Vital Statistics Agency, and the British Columbia Autism Assessment Network).