Autism Risk and Perinatal Antibiotic Use

We conducted a retrospective cohort study of everyone who delivered a live, singleton infant in the province of BC, Canada (population of ∼5 million), between April 1, 2000, and December 31, 2014. Children were followed until a clinical diagnosis of ASD, death, the study end date of December 31, 2016, or until they moved out of the province and were lost to follow-up. Population Data BC linked data from the BC Perinatal Data Registry,¹⁶  the BC Consolidation file,¹⁷  Vital statistics,¹⁸  the BC Autism Assessment Network,¹⁹  and the BC Ministry of Education.²⁰  Ethics approval was obtained from the BC Children’s and Women’s Research ethics board. Approval by the ethics board and the BC data stewards for use of deidentified administrative data files includes a waiver of informed consent from participants.

The BC Perinatal Data Registry is a provincewide registry that includes information on antenatal, intrapartum, and postpartum birth parent and infant care and outcomes for nearly 100% of deliveries in BC, including home births. These data were used to build our cohorts. We examined the relationship between antibiotic administration during labor and delivery and ASD among all singleton, term deliveries, regardless of mode of delivery. We excluded children who did not have 2 full years of follow-up and children who were assessed for ASD before aged 2 years because assessments are more reliable in children aged 2 years and above.²¹ 

To examine the relationship between administration of antibiotics during labor and delivery among a group of pregnant individuals being treated for the same underlying indication, we restricted our main cohort to a subcohort of only those who tested positive for GBS. In BC, GBS is diagnosed by bacterial culture and is considered a definite infection. Moreover, since intrapartum antibiotic prophylaxis for GBS infection is not recommended for planned cesarean deliveries in the absence of labor and rupture of membranes, cesarean deliveries were only included in the GBS-positive cohort if they were coupled with a spontaneous or induced labor code.

The BC Perinatal Data Registry includes data fields that are specific to antibiotic administration during the labor and delivery period. All systemic antibiotics were considered.

In BC, ASD assessment is publicly funded through the BC Autism Assessment Network (BCAAN), and conducted by pediatricians, psychiatrists, or psychologists who have completed intensive training in diagnosing ASD.¹⁹  Since 2004, all children have been evaluated in a standardized manner at BCAAN using the following 2 assessment instruments: Autism Diagnostic Observation Schedule and the Autism Diagnostic Interview-Revised.^22,23  Private practitioners who have undergone specialty training in ASD assessment also diagnose some children in BC, and are required to use the Autism Diagnostic Observation Schedule or Autism Diagnostic Observation Schedule-2 and ADI-R as well. We identified these cases through linkage to the data held by the Ministry of Education. While our cohort includes children born on or after January 1, 2000, very few children would have been diagnosed prior to standardization in 2004. This is because the median age of ASD assessment at BCAAN is 5.35 years; and because we excluded children in our cohort that were diagnosed under the age of two. The few children diagnosed between 2002 and 2004 were captured in our linked Ministry of Education data.

We adjusted for sociodemographic variables including birth parent age at delivery, neighborhood income quintile, location of residence (urban, rural, or semirural residence), and coparent age. Additionally, we adjusted for pregnancy conditions and risk factors that could have influenced fetal health, including parity, smoking during pregnancy, preexisting diabetes, gestational diabetes, pregnancy-induced hypertension and other types of hypertensions (which include preexisting hypertension, hypertensive renal disease, high blood pressure [distinct from preexisting hypertension], and proteinuria), prepregnancy BMI, and prenatal antibiotic exposure. Further, we controlled for labor, delivery, and antenatal care characteristics such as year and mode of delivery. Finally, we examined neonatal characteristics including sex, birth weight, neonatal antibiotics administered during hospital admission, gestational age, and whether the infant was small or large for gestational age. Infants were considered small for gestational age if they were below the 10^th percentile and large for gestational age if they were above the 90^th percentile of birth weight for their gestational age and sex.²⁴  For BMI, because of a high degree of missing data, multiple imputation was used. For all other variables, multiple imputation was not used, and data were assumed missing at random.

Covariates were compared according to antibiotic use, using standardized mean differences, first in the main cohort and then within the subcohort of pregnant individuals who tested positive for GBS. We considered a covariate to have a clinically meaningful difference between the 2 exposure groups if the standardized mean difference was >0.1.²⁵ 

Using Cox proportional hazards models, we derived crude and adjusted hazards ratios (HRs) and their 95% confidence intervals (CIs). We censored children who died or who were no longer registered in the province’s universal health insurance because this likely reflects a move out of the province. Clustering between multiple deliveries to the same birth parent over our study period was accounted for using robust SE estimates. These HRs may be interpreted as the relative risk of ASD because ASD is considered a rare disorder with a prevalence of <10%.

As a proxy for dose, since intrapartum antibiotic prophylaxis for GBS infection is given at the onset of labor and again every 4 hours until delivery, we studied subcohorts of pregnant individuals who had their first stage of labor <4 hours, between 4 and 7 hours, between 8 and 11 hours, and 12 hours or longer, to assess for a potential dose–response relationship between antibiotic exposure and the risk of ASD. In addition, we examined a subcohort of pregnant individuals who were exposed to antibiotics, both prenatally and during labor and delivery, and another subcohort of children exposed to labor and delivery antibiotics, as well as neonatal antibiotics. Finally, all analyses were stratified by sex and mode of delivery to examine potential effect modification.

We identified 599 280 singleton births among 386 066 birth parents between April 1, 2000, and December 31, 2014 (Fig 1). After the exclusion criteria were applied, the main cohort included 569 953 deliveries to 366 152 birth parents. Of these deliveries, 231 629 infants (40.6%) were exposed to antibiotics during labor and delivery. Children were followed up for a mean of 8.7 years (SD, 4.3 years).

There were important differences among deliveries according to antibiotic exposure (Table 1) Birth parents exposed to antibiotics during labor and delivery were older, more likely to live in urban settings, had a higher prevalence of hypertension, and were more commonly multiparous, diabetic, and higher in prepregnancy BMI. Further, exposed deliveries had a higher proportion of cesarean deliveries, epidural use, and labor induction, and a longer first stage of labor. Infants who were exposed during labor and delivery had larger birth weights, older gestational ages, higher likelihood of being admitted to the NICU and receiving neonatal antibiotics, and shorter study follow-up times. Among the GBS-positive cohort, those exposed to antibiotics were more likely to have been delivered by cesarean delivery, given an epidural, had their labor induced or augmented, had a longer first stage of labor, and were more likely exposed to neonatal antibiotics (Table 2).

By the end of follow-up on December 31, 2016, there were 8729 children (1.53%) who had been diagnosed with ASD. Of note, 72% of these diagnoses were made by clinicians in the BCAAN and the remaining 28% were diagnosed by private practitioners in the province. Of the children who were exposed to antibiotics during labor and delivery, 1.66% received an ASD diagnosis. In the group of children who were unexposed during labor and delivery, 1.44% received an ASD diagnosis. The unadjusted risk difference was 0.22%.

The association between ASD and exposure to antibiotics during labor and delivery had an unadjusted HR of 1.29 (95% CI 1.24–1.35; Table 3; Fig 2). After adjusting for birth-parent sociodemographic characteristics, pregnancy conditions and risk factors, labor, delivery and antenatal care characteristics, and neonatal characteristics, the association attenuated to the null with an HR of 0.99 (95% CI 0.94–1.04).

Upon categorizing deliveries on the basis of length of first stage of labor, prenatal with labor and delivery antibiotic status, and labor and delivery with neonatal antibiotic status, results from the main cohort remained consistent in terms of direction and effect size (Table 3; Fig 2). For all lengths of labor and antibiotic exposure subcohorts that we analyzed, the unadjusted HR suggested an increased risk associated with antibiotic exposure, but there was no statistically significant association between antibiotic exposure and ASD in the fully adjusted models.

Of children who were exposed to antibiotic prophylaxis for GBS during labor and delivery, 1.41% received an ASD diagnosis, whereas 1.46% of unexposed children received an ASD diagnosis (Table 4; Fig 3). The unadjusted risk difference was 0.06%. Results differed from the main cohort in terms of effect size and direction. Every model, even unadjusted, reported no statistically significant association between antibiotic exposure and ASD (HR 1.07 [95% CI 0.90–1.27]). Sensitivity analyses based on length of first stage of labor, as well as exposure to both prenatal and labor and delivery antibiotics and both labor and delivery and neonatal antibiotics, were also conducted for this GBS-positive cohort. Results showed no substantial difference in the observed associations (Table 4; Fig 3).

After stratifying the analysis by sex, the male cohort included 292 563 children, whereas the female cohort consisted of 277 379 children. Among males, 2.64% of children exposed to antibiotics during labor and delivery and 2.32% of unexposed children received an ASD diagnosis (unadjusted risk difference of 0.32%). Among females, 0.59% of exposed children and 0.54% of unexposed children received an ASD diagnosis (unadjusted risk difference of 0.06%). Results were consistent with the main cohort regarding direction of association and effect size (Supplemental Tables 5, 6, 7, and 8). The analyses were additionally stratified by mode of delivery, with no substantial variation in the associations observed after adjusting for covariates in the different models (Supplemental Tables 9 and 10).

In this population-based study in BC, we observed no significant association between antibiotic use during labor and delivery and ASD after controlling for birth-parent sociodemographic and pregnancy-related conditions and risk factors, as well as labor, delivery, and neonatal characteristics. We conducted sensitivity analyses using length of first stage of labor as a proxy measure for dose to assess for a dose–response relationship. Additionally, we studied subcohorts of children who were exposed to both prenatal and labor and delivery antibiotics, and children who were exposed to both labor and delivery and neonatal antibiotics. Our results showed no substantial difference in the association between antibiotic exposure and ASD depending on length of the first stage of labor or combined antibiotic statuses.

Our findings are consistent with a previous population-based, case-control study which found no significant association between peripartum antibiotic exposure and ASD in the offspring.²⁶  However, those study results were weakened by a small sample size, wide CIs, and imprecise estimates. Our null findings could be because of the magnitude of microbial change being too small to impact neurodevelopment or that the duration of antibiotic exposure was too short to produce an impactful change in the vaginal microbiome. Further research is needed to investigate longer durations of antibiotic use throughout pregnancy and its association with ASD.

Recent research has suggested a role of the childhood microbiome–gut–brain axis in the pathogenesis of ASD; thus, highlighting exposures that alter the microbiome as risk factors of ASD development.^6,7  Several studies have shown that intrapartum antibiotic prophylaxis reduces the microbial diversity of infants.^27–29  Further, studies reported that infants who were exposed to antibiotics during the intrapartum or peripartum period had significant changes in both the quantities and diversity of microbial composition that continued until at least 1 year of age.^30,31  Despite these findings, research following intrapartum antibiotic prophylaxis-exposed infants through to examine clinical neurodevelopmental outcomes is limited.³²  Our study helped to address this knowledge gap by assessing the association between antibiotic use during labor and delivery with the risk of ASD in offspring. Our findings provide reassurance that this common obstetrical practice does not increase the risk of ASD.

This study has multiple strengths. First, the study population included all deliveries that occurred in BC during the designated study period, thus eliminating selection bias. Second, we were able to assess the association between antibiotic use and ASD in a group of deliveries that were given antibiotics for the same indication, birth-parent GBS. This provided a unique way to address confounding by indication, which has been a key limitation of previous research exploring associations with antibiotic exposure.^33,34  Further, through linkage with the BCAAN, we gained prospectively collected clinical gold standard diagnostic data for our ASD outcome. Many epidemiologic studies have used diagnostic codes from administrative data sets for outcome ascertainment. Recent research has found that this is not the most reliable method to identify ASD cases, and although it may produce a good positive predictive value, the negative predictive value is of concern.³⁵  Finally, antibiotic exposure data came from the BC Perinatal Data Registry. A validation study of the registry was previously conducted and determined that its clinical variables on interventions during labor and delivery have overall excellent validity.³⁶ 

This study has several limitations. Apart from GBS infection, we were unable to identify the other indications for antibiotic use during labor and delivery in the main cohort. Stratifying by multiple indications for antibiotic use, especially by those of differing severity, would provide a deeper understanding into the potential role of confounding by indication. We were also unable to observe the exact number of doses and specific durations of intrapartum antibiotic exposure. It is likely that a higher number of doses and a longer cumulative duration of exposure would have a more significant impact on the vaginal microbiome and the initial colonization of the infant’s microbiome; thus, future research with data on the number of doses and durations of antibiotic exposure is needed for a strong assessment of a dose–response relationship between perinatal antibiotic exposure and ASD. In addition, we were unable to identify the specific classes of antibiotics administered or the ethnicities of the birth parents included in our cohort. Our GBS-positive cohort had a smaller sample size, which widened CIs and resulted in less-precise risk estimates. Moreover, since this study has an observational design, some residual confounding likely remains. This is apparent when observing important differences between characteristics of deliveries in which antibiotics were administered and not administered. These differences were particularly evident among variables that suggest a worse prognosis for birth parents, such as diabetes, larger prepregnancy BMIs, higher prevalence of labor induction and augmentation, increased epidural usage, and a greater number of NICU admissions.

In this population-based study, antibiotic use during labor and delivery was not statistically significantly associated with an increased risk of ASD in the offspring. Our findings suggest that concern for ASD should not factor into the clinical decision on whether to administer antibiotics during labor and delivery for any indication. Future research should investigate the impact of longer periods of antibiotic exposure during pregnancy, and exposure by antibiotic class, on the risk of ASD while continuing to address confounding by indication.

ASD

autism spectrum disorder

BC

British Columbia

BCAAN

British Columbia Autism Assessment Network

CI

confidence interval

GBS

group B Streptococcus

HR

hazard ratio

IAP

intrapartum antibiotic prophylaxis