Early childhood antibiotics use and autism spectrum disorders: a population-based cohort study

Abstract

Background

Changes in microbiota composition as a result of antibiotics use in early life has been proposed as a possible contributor in the aetiology of autism spectrum disorders (ASD). We aimed to examine the association between early life antibiotic exposure and risk of ASD.

Methods

This was a population-based cohort study which included all live births in Manitoba, Canada, between 1 April 1998 and 31 March 2016. We used administrative health data from the Manitoba Population Research Data Repository. Exposure was defined as having filled one or more antibiotic prescription during the first year of life. The main outcome was ASD diagnosis. Cox proportional hazards regression models were used to estimate the risk of developing ASD in the overall population and in a sibling cohort.

Results

Of all subjects in the cohort (n = 214 834), 94 024 (43.8%) filled an antibiotic prescription during the first year of life. During follow-up, 2965 children received an ASD diagnosis. Compared with children who did not use antibiotics during the first year of life, those who received antibiotics had a reduced risk of ASD [adjusted hazardz ratio (HR) 0.91, 95% confidence interval (CI) 0.84–0.99). Number of treatment courses and cumulative duration of antibiotic exposure were not associated with ASD. In the sibling-controlled analysis, early life antibiotic exposure was not associated with ASD (adjusted HR 1.03, 95% CI 0.86–1.23).

Conclusions

Our findings suggested no clinically significant association between early life antibiotics exposure and risk of autism spectrum disorders, and should provide reassurance to concerned prescribers and parents.

Neurodevelopmental disorders, microbiota, antibiotics, autism spectrum disorder

Key Messages

• Microbiota changes due to antibiotics use was proposed to play a role in the development of autism spectrum disorders.

• In this population-based cohort study, we found that antibiotics exposure during the first year of life was not associated with increased risk of autism spectrum disorders.

• Study findings should provide reassurance to concerned prescribers and parents.

Introduction

Autism spectrum disorders (ASD) are among the most commonly diagnosed neurodevelopmental disorders, with 52 million cases worldwide.^1–3 They represent a group of disorders characterized by impairment in social communication and interaction, with repetitive patterns of behaviour.⁴ The global prevalence of ASD has been increasing over the years, yet the exact aetiology remains unclear.^2,5–7 A complex genetics model is proposed to explain the aetiology of ASD, where an interaction of several gene variants along with environmental factors is required for the disorder to manifest.^8,9 A substantial amount of literature has examined prenatal and postnatal environmental factors as predictors for ASD;^10–16 however, causal relationships remain inconclusive. As such, it remains unclear how these environmental factors may play a role in the development of ASD, if any.

Recent research has pointed to a potential role of human microbiota in the aetiology of neurodevelopmental disorders. Microbiota, a diverse collection of microorganisms in the human body, has a major role in the communication between the gastrointestinal tract and the central nervous system, a relation referred to as the microbiota-gut-brain axis.^17–19 It has been shown that children with ASD presenting with frequent gastrointestinal symptoms have worse ASD symptom scores, which can be attributed to changes in gastrointestinal microbiota.²⁰ Moreover, many studies examining gastrointestinal microbiota in children with ASD reported significant differences in microbiota composition compared with control groups.^21–24 In a recent exploratory study, extended-duration microbiota transfer therapy to subjects with ASD resulted in improving gastrointestinal and behavioural symptoms of ASD.²⁵

Consistent with this theory, it has been hypothesized that early life changes to gut microbiota composition, potentially induced by antibiotics exposure, may impair the gut-brain axis and, as a result, increase the risk of ASD.^{17,18,26–29} In this study, we aimed to examine the association between antibiotic use during the first year of life and the risk of ASD.

Methods

Design and subjects

We conducted a population-based cohort study using the Manitoba Population Research Data Repository, which provides a comprehensive collection of administrative, registry, survey and other data on all Manitoba residents and is housed at the Manitoba Centre for Health Policy (MCHP). The health system in the province of Manitoba is universal and publicly funded, hence any encounter with the health system or drug dispensation is captured in the Repository. All patient records in the Repository are de-identified, and linkage among different databases was achieved through scrambled Personal Health Identification Numbers (PHIN).

The cohort included all births identified in the Manitoba Health Insurance Registry between 1 April 1998 and 31 March 2016. The birthdate was assigned as the index date for cohort entry. A minimum of 18 months of valid Manitoba health registration was required for children to be included in the cohort, and we excluded subjects whose mothers had less than 2 years of valid Manitoba health registration preceding the index date. Children were followed until the earliest of a diagnosis of ASD, migration out of province, age of 18 years, death or end of study period (31 March 2016).

Other data sources of the study included the Drug Program Information Network (DPIN), In-hospital Pharmaceuticals, Hospital Abstracts, physician claims from the Medical Services database, the Manitoba Education and Training Special Needs Funding data, the Hospital Newborn to Mother LinkRegistry, BabyFirst - Families First Screen and the Social Allowances Management Information Network (SAMIN) (see Supplementary Table 1, available as Supplementary data at IJE online, for description of data sources). The study was approved by the University of Manitoba Health Research Ethics Board and the Health Information Privacy Committee of Manitoba Health, Seniors and Active Living.

Exposure

The primary measure of exposure was defined as having filled one or more antibiotic prescriptions during the first year of life as recorded in the DPIN, which captures all prescription drug dispensation outside the hospital setting. We further analysed the exposure based on the number of antibiotic courses received, cumulative duration within the first year of life and the class of antibiotic (see Supplementary Table 2, available as Supplementary data at IJE online for antibiotics classification and ATC codes).

Outcome

The primary outcome was ASD diagnosis identified between the age of 18 months and the end of follow-up period, using claims from the hospital discharge abstracts, medical services (physician claims) and the Manitoba Education and Training Special Needs Funding data. In Canada, the 9th and 10th revisions of the International Classification of Disease (ICD) coding system are currently used to report health services in physician claims (outpatient) and hospital discharge abstracts (inpatient admissions), respectively.

We defined ASD according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) which includes childhood autism, atypical autism, Asperger’s disorder, childhood disintegrative disorder, other pervasive developmental disorders and pervasive developmental disorders not otherwise specified.⁴ As established by previous research,^13,30–32 ASD was defined as one or more hospitalization with an ASD code (ICD-9 299.0, 299.1, 299.8 or 299.9, or ICD-10 F84.0, F84.1, F84.3, F84.5, F84.8 or F84.9), one or more physician visit with ICD-9 code of 299 or presence of an ‘ASD’ identifier in the Manitoba Education and Training Special Needs Funding data. This definition was based on a previous validated algorithm with a positive predictive value of 88%,³³ with the addition of educational data as a source to identify ASD.

Covariates

Sociodemographic characteristics, potential confounders and previously reported ASD predictors were included as covariates.^{10–16,28,34–36} Maternal covariates included region of residence (urban or rural) andsocioeconomic status (SES), prenatal smoking, prenatal alcohol or drug use, prenatal infections, medical conditions of interest and prenatal use of antidepressants (see Supplementary Tables 3 and 4, available as Supplementary data at IJE online for list of medications included as covariates and diagnostic criteria for maternal and childhood medical conditions).Since previous literature reported an increased risk of ASD at maternal ages above 30 years,¹⁰ we included mothers’ age at delivery with three categories:less than 30, 30 to 39, and 40 years or greater. Mothers’ physician visits in the year preceding pregnancy were also included, to control for differences in health care access. Child covariates included sex, size for gestational age, mode of delivery, birth complications, breastfeeding initiation, multiple births, birth order (firstborn or subsequent) and medical conditions of interest. Season of birth and year of birth werealso included to adjust for detection bias. In addition, we included childhood infections to account for potential bias due to confounding by indication.

Statistical analysis

A multivariable Cox proportional hazards regression model was used to examine the association between antibiotic exposure and ASD diagnosis, adjusting for significant confounders and covariates. Subjects were right-censored at migration out of province, death, age of 18 years, or end of study period if they did not receive an ASD diagnosis. We excluded subjects with missing data on any of the relevant covariates from the analysis. Hazard ratios were estimated to compare risk of ASD among those exposed to antibiotics and those who were not exposed. Interactions of covariates with antibiotic exposure were explored. Correlation matrix and chi-square tests were conducted to examine multicollinearity among covariates. Proportional hazards assumptions were tested by examining the correlation between follow-up time and Schoenfeld residuals of exposure.

The main analysis included history of antibiotic exposure in the first year of life as the predictor and was stratified based on region and sex as potential effect modifiers. We conducted secondary analyses to examine the association based on the number of antibiotic courses, antibiotic classes and cumulative duration within the first year of life.

Multiple sensitivity analyses were conducted. First, we restricted the cohort to children with diagnosed infection in the first year of life, to account for potential confounding by indication. Second, to address potential misclassification of the outcome, we applied a stricter diagnostic algorithm to identify ASD by requiring one hospitalization, two physician claims within 3 years or one physician claim plus educational special needs funding for ASD within 3 years. Additional sensitivity analyses were also done by modifying the minimum ages for ASD diagnosis to 1 and 2 years, altering the antibiotics exposure window to 6 and 18 months of life, including in-hospital antibiotic use in identifying exposure and examining the risk of ASD co-occurring with attention-deficit hyperactivity disorder (ADHD) or intellectual disability (ID).

To account for the likelihood of unmeasured confounding due to familial and genetic factors, we conducted a sibling-controlled analysis by limiting the cohort to subjects who have at least one maternal sibling discordant in antibiotic exposure. Using multivariable Cox proportional hazards regression models, the sibling cohort were stratified by their mothers to examine the association between antibiotic exposure and ASD diagnoses, adjusted for the same covariates used in the main model except for health care access and maternal medical conditions, which we assumed to be static across pregnancies.The statistical software SAS^® 9.4 (SAS Institute; Cary, NC) was used for all data analyses.

Results

Description of study population

There was a total of 261 655 births in Manitoba between 1998 and 2016. Of those, 214 834 children met the criteria for cohort inclusion (Figure 1). The cohort was well balanced in sex (51.3% males) and geographical area (54.4% were living in urban regions). Other baseline characteristics are presented in Table 1. Of all subjects, 43.8% received at least one antibiotic course during the first year of life. Among those, 75.8% received one or two courses, 56.1% received antibiotics for less than or equal to 2 weeks and 58.6% received a penicillin antibiotic (see Supplementary Table 5, available as Supplementary data at IJE online for description of antibiotics use). Comparing baseline characteristics between the two exposure groups showed a lack of balance in several variables such as sex, socioeconomic status, prenatal smoking and childhood medical conditions. Prenatal smoking, alcohol and drug use were excluded from the main model due to the large percentage of missing data (Table 1), but were subsequently examined in a sensitivity analysis. Subjects were followed for a total of 1 943 612 person-years with a median of 8.6 person-years [interquartile range (IQR) 4.8–13.2]. During follow-up, 2965 children received an ASD diagnosis at a mean age of 5.48 years [standard deviation (SD) 3.24]. The crude incidences for ASD diagnosis were 1.46 per 1000 person-years and 1.59 per 1000 person-years in children exposed and unexposed to antibiotics in early life, respectively.

Table 1.

Characteristics of study cohort: overall and by antibiotics exposure status^a

	All subjects n (%) (n = 214 834)	Antibiotic use during the first year of life
		No n (%) (n  = 120 810)	Yes n (%) (n  = 94 024)
Male	110 107(51.3)	59 240 (49.0)	50 867 (54.1)
Urban region	116 865(54.4)	64 806 (53.6)	52 059 (55.4)
Socioeconomic status (SES)b:
High	21 212 (9.9)	13 286 (11.0)	7926 (8.4)
Middle	77 708 (36.2)	45 578 (37.7)	32 130 (34.2)
Low-mid	67 059 (31.2)	35 857 (29.7)	31 202 (33.2)
Low	48 855 (22.7)	26 089 (21.6)	22 766 (24.2)
Receipt of income assistancec	37 158 (17.3)	14 058 (11.6)	23 100 (24.6)
Maternal age at delivery (years):
<30	128 229(59.7)	68 793 (56.9)	59 436 (63.2)
30–39	81 963 (38.2)	49 101 (40.6)	32 862 (35.0)
>= 40	4642 (2.2)	2916 (2.4)	1726 (1.8)
Breastfeeding initiationd	172 952 (80.8)	99 269 (82.5)	73 683 (78.7)
Multiple birthe	5383 (2.5)	3202 (2.7)	2181 (2.3)
Caesarian section	43 782 (20.4)	24 751 (20.5)	19 031 (20.2)
Birth complications	21 022 (9.8)	11 493 (9.5)	9529 (10.1)
Firstborn child	80 758 (37.6)	49 211 (40.7)	31 547 (33.6)
Small for gestational agef	16 507 (7.7)	9501 (7.9)	7006 (7.5)
Prenatal alcohol/drug useg	12 959 (12.3)	6407 (11.4)	6552 (13.4)
Prenatal smokingh	20 844 (19.4)	9199 (16.0)	11 645 (23.4)
Childhood medical conditions:
Infections:
None	67 430 (31.4)	59 635 (49.4)	7795 (8.3)
Mild-moderatei	134 794 (62.7)	56 960 (47.2)	77 834 (82.8)
Severej	12 610 (5.9)	4215 (3.5)	8395 (8.9)
Epilepsy	1056 (0.5)	424 (0.4)	632 (0.7)
Neonatal jaundice	20 517 (9.6)	10 956 (9.1)	9561 (10.2)
Other developmental disabilities	859 (0.4)	355 (0.3)	504 (0.5)
Asthma	26 641 (12.4)	5276 (4.4)	21 365 (22.7)
Maternal medical conditions:
Mood and anxiety disorders	15 885 (7.4)	7528 (6.2)	8357 (8.9)
Schizophrenia	196 (0.1)	111 (0.1)	85 (0.1)
Diabetes	6085 (2.8)	3372 (2.8)	2713 (2.9)
Prenatal infections	67 559 (31.5)	31 506 (26.1)	36 053 (38.3)
Prenatal antidepressants exposure	5759 (2.7)	2900 (2.4)	2859 (3.0)
Year of birth:
1998–2001	47 107 (21.9)	20 593 (17.1)	26 514 (28.2)
2002–05	48 596 (22.6)	25 260 (20.9)	23 336 (24.8)
2006–09	53 052 (24.7)	31 640 (26.2)	21 412 (22.8)
2010–14	66 079 (30.8)	43 317 (35.9)	22 762 (24.2)
Season of birth:
Winter	49 039 (22.8)	27 677 (22.9)	21 362 (22.7)
Spring	56 517 (26.3)	30 493 (25.2)	26 024 (27.7)
Summer	58 987 (27.5)	33 331 (27.6)	25 656 (27.3)
Fall	50 291 (23.4)	29 309 (24.3)	20 982 (22.3)

a

Numbers (percentage). Percentages are calculated based on non-missing data.

b

Socioeconomic factor index, a neighbourhood level measure based on Canada census, was categorized with cut-off points within one standard deviation from the mean into high, middle, low-middle and low SES.

c

Defined as receiving income assistance for at least 2 months within 1 year before to 18 months after index date.

d

Missing data for 886 (0.4%) subjects.

e

Defined as the number of births following a multiple gestation pregnancy.

f

Defined as having birthweight below the 10 percentile for the gestational age and sex. Missing data for 502 (0.2%) subjects.

g

Missing data for 109 491 (51.0%) subjects.

h

Missing data for 107 550 (50.1%) subjects.

i

Defined as having an infection code in physician claims only.

j

Defined as having a hospitalization with an infection code.

Figure 1

Study populations: overall and sibling cohort: 261 655 births were identified in Manitoba Health Insurance Registry during study period. Among those, 214 834 met the inclusion criteria for the main cohort; 80 225 subjects were identified with siblings discordant to antibiotic exposure and were included in the siblings’ cohort.

Cox regression model for overall cohort

In the main analysis, antibiotic exposure was not found to be associated with ASD diagnosis (HR 0.93, 95% CI 0.87–1.00). After adjusting for covariates, the risk estimates varied very little. Although statistical significance was found (HR 0.91, 95% CI 0.84–0.99), we were unable to conclusively reject the null, given the confidence interval limits’ proximity to 1.00. Stratifying the analysis by sex and region also resulted in a statistically significant association in males (adjusted HR 0.91, 95% CI 0.83–1.00), and with those residing in urban regions (adjusted HR 0.85, 95% CI 0.77–0.94). In secondary analyses, number of antibiotic courses and cumulative duration were not associated with ASD (Table 2).

Table 2.

Association between antibiotics use and risk of ASD in the overall cohort

Variable	Person-years	Number of events	HR (95% CI)
			Unadjusted	Adjusteda
Main analysis
History of antibiotics use	1 943 612	2965	0.93 (0.87–1.00)	0.91 (0.84–0.99)*
Stratified by sex:
Male	991 939	2401	0.86 (0.79–0.93)***	0.91 (0.83–1.00)*
Female	951 673	564	0.97 (0.82–1.14)	0.92 (0.76–1.12)
Stratified by region:
Rural	887 497	953	1.10 (0.97–1.25)	1.01 (0.88–1.17)
Urban	1 056 105	2012	0.89 (0.82–0.97)**	0.85 (0.77–0.94)**
Secondary analyses
Antibiotic class
None	1 013 186	1610	1.00 [Reference]	1.00 [Reference]
Penicillin	358 792	1131	0.93 (0.86–1.01)	0.92 (0.84–1.00)*
Macrolides and related antibiotics	254 769	366	0.92 (0.82–1.03)	0.87 (0.77–0.99)*
Other beta lactams	201 080	315	1.00 (0.88–1.12)	0.93 (0.82–1.05)
Others	115 785	156	0.89 (0.76–1.05)	0.92 (0.77–1.09)
Number of antibiotic courses
0	1 013 186	1610	1.00 [Reference]	1.00 [Reference]
1	453 484	657	0.94 (0.86–1.03)	0.92 (0.83–1.01)
2	229 957	329	0.94 (0.83–1.06)	0.89 (0.78–1.01)
3	120 114	179	0.98 (0.84–1.15)	0.93 (0.79–1.09)
>= 4	126 871	190	0.99 (0.86–1.16)	0.90 (0.76–1.06)
Cumulative antibiotic duration (days)
0	1 013 186	1610	1.00 [Reference]	1.00 [Reference]
1–7	173 235	243	0.89 (0.78–1.02)	0.91 (0.79–1.05)
8–14	320 640	464	0.92 (0.83–1.02)	0.91 (0.82–1.01)
15–21	177 468	269	0.98 (0.86–1.11)	0.95 (0.82–1.08)
>21	259 083	379	0.95 (0.85–1.06)	0.88 (0.78–1.00)

a

Adjusted for sex, region, health care access, SES, maternal age at delivery, maternal medical conditions (mood and anxiety disorders, schizophrenia, diabetes, prenatal infections), prenatal antidepressants use, size for gestational age, childhood medical conditions (epilepsy, infections, neonatal jaundice, asthma and diagnosis with other developmental disability disorder), birth complications, mode of delivery, multiple birth, breastfeeding initiation, year of birth, season of birth and birth order.

*

P-value <0.05;

**

<0.01;

***

<0.001.

Table 3.

Association between antibiotics use and risk of ASD in siblings’ cohort

Variable	Person-years	Number of events	HR (95% CI)
			Unadjusted	Adjusteda
Main analysis
History of antibiotics use	746 135	1012	1.03 (0.89–1.19)	1.03 (0.86–1.23)
Stratified by sex:
Male	380 952	817	0.99 (0.80–1.23)	0.98 (0.76–1.26)
Female	365 184	195	0.85 (0.54–1.33)	0.76 (0.41–1.39)
Stratified by region:
Rural	393 205	395	1.14 (0.89–1.45)	1.10 (0.80–1.50)
Urban	352 930	617	1.03 (0.85–1.24)	1.00 (0.79–1.28)
Secondary analyses
Antibiotic class
None	368 556	477	1.00 [Reference]	1.00 [Reference]
Penicillin	306 227	435	1.11 (0.95–1.29)	1.11 (0.93–1.34)
Macrolides and related antibiotics	94 429	134	1.13 (0.89–1.42)	1.13 (0.87–1.46)
Other beta lactams	78 132	130	1.31 (1.04–1.66)*	1.18 (0.91–1.53)
Others	37 229	56	1.14 (0.81–1.59)	1.22 (0.85–1.75)
Number of antibiotic courses
0	368 556	477	1.00 [Reference]	1.00 [Reference]
1	210 538	282	0.97 (0.81–1.16)	0.98 (0.79–1.21)
2	89 899	131	1.07 (0.82–1.39)	1.09 (0.80–1.47)
3	41 010	69	1.42 (0.97–2.08)	1.46 (0.94–2.26)
>= 4	36 132	53	0.99 (0.65–1.50)	0.87 (0.54–1.40)
Cumulative antibiotic duration (days)
0	368 556	477	1.00 [Reference]	1.00 [Reference]
1–7	85 069	112	0.93 (0.71–1.22)	1.00 (0.73–1.37)
8–14	142 259	190	0.98 (0.79–1.21)	0.95 (0.74–1.21)
15–21	68 965	112	1.28 (0.95–1.72)	1.30 (0.91–1.84)
>21	81 286	121	1.08 (0.81–1.43)	1.05 (0.75–1.48)

a

Adjusted for sex, region, SES, maternal age at delivery, prenatal infections, prenatal antidepressants use, size for gestational age, childhood medical conditions (epilepsy, infections, neonatal jaundice, asthma and a diagnosis with other developmental disability disorder), birth complications, mode of delivery, multiple birth, breastfeeding initiation, year of birth, season of birth, and birth order.

*

P-value <0.05.

Sensitivity analyses

No major shift in risk estimates was observed in the planned sensitivity analyses. The association between early childhood antibiotics exposure and ASD was consistent across the different sensitivity analyses with changes in exposure, outcome definitions and other parameters (Figure 2).

Figure 2

Forest plot with sensitivity analyses risk estimates: adjusted hazards ratios for the risk of ASD are presented for each of the planned sensitivity analyses.

Sibling cohort analysis

The sibling cohort included 80 225 subjects with 57 063 sibling pairs discordant in exposure status (Figure 1). In this cohort, 1012 subjects developed ASD during a median follow-up of 9.1 (IQR 5.6–12.9) years. For baseline characteristics of the sibling cohort, see Supplementary Table 6, available as Supplementary data at IJE online. Early life antibiotic exposure was not associated with ASD in the sibling-controlled analysis (adjusted HR 1.03, 95% CI 0.86–1.23). No substantial variation in the risk association was observed in all secondary analyses, and all estimates remained non-significant (Table 3).

Discussion

The main analysis of this large population-based cohort study, consisting of all births occurring in Manitoba over an 18-year period, showed a trend towards reduced risk of ASD in infants exposed to antibiotics in their first year of lives compared with those not exposed. This marginal association was observed in males, and in those residing in urban areas when stratified by sex and region. In secondary analyses, risk reduction appeared to be significant in those exposed to macrolides and penicillins. Number of antibiotic courses or cumulative duration on antibiotics was not associated with ASD risk. Several sensitivity analyses modifying various exposure and outcome parameters affirmed the robustness of the risk estimates in the main model. Despite the statistically significant findings in the main models, we do not believe the observed association was clinically meaningful.

There were a few causes for concern with the findings from our main model. Whereas the risk estimate was robust, the confidence interval approximated null. There was also an absence of a dose-response effect when antibiotic exposure was stratified by cumulative duration or number of antibiotic courses, which cast further uncertainty on the observed association between early life antibiotic exposure and ASD. The most critical issue affecting the validity of the findings from the primary analyses, however, was the fact that the main model could not account for confounding due to environmental, genetic and other familial or social factors known to be major contributors in the ASD disease model. To address these outstanding confounders, we explored the association in subsequent analyses using a sibling cohort.

In the analyses based on a sibling-controlled design, antibiotics use in first year of life was not associated with risk of developing ASD in both unadjusted and adjusted models. The associations remained non-significant after stratifying by sex and region and when exposure was examined by the number of courses, cumulative duration and antibiotic class. The discrepancy in results between the models based on the overall birth cohort and the sibling cohort suggested high susceptibility of the former to systematic confounding by familial factors that cannot be identified or measured in administrative databases. This led us to conclude that the marginal association between early life antibiotic exposure and ASD observed in the main model was unlikely to be meaningful and that, based on the sibling-controlled design, antibiotic exposure in infant years did not appear to be associated with ASD.

Antibiotics are the most frequently prescribed medications for children, where over-prescription and inappropriate use are often observed.^37–40 Accordingly, identifying antibiotics’ association with ASD, if any, would be of public health interest. Previous research suggested a role for microbiota in the development of ASD through disruption of the gut-brain axis.^20–24,29 One research group examined gut microbiota of infants exposed to antibiotics perinatally and found significant changes in both the diversity and the quantities of microbiota composition at 1 year of age.²⁷ Hence, it has since been stipulated that early life antibiotics exposure could induce long-term changes to microbiota composition. Previous observational studies reported an increased risk of several childhood diseases, including asthma, inflammatory bowel disease, idiopathic juvenile arthritis, obesity and eczema,^41–51 with early life antibiotic-induced microbiota changes as the proposed aetiology. On the contrary, we did not observe this expected increase in ASD rates in children exposed to antibiotics in their first years of life. This could be attributed to the magnitude of microbiota changes being too small to impair neurodevelopment, or that a specific profile of microbiota changes is responsible for neurodevelopmental disorders.

Our study has several strengths. We used 18 years of population-based data, allowing for a large sample size and long follow-up period. In addition, we examined a comprehensive set of potential confounders and ASD predictors for inclusion in the model. Most importantly, we conducted a sibling-controlled analysis to account for unmeasured confounding due to genetics and shared environmental factors. Since we did not exclude any children on basis other than Manitoba Health Insurance registration, the results are likely generalizable.

We identified a few potential limitations in our study. First, drug dispensation may not accurately reflect actual drug use, potentially resulting in misclassification of the exposure, which could bias the results towards the null. We also did not consider antibiotic exposure in the hospital setting, due to poor database quality. We did, however, include in-hospital pharmaceutical data in our sensitivity analysis, and the risk estimates were similar. Second, although theASD identification algorithm used was adopted from previous research, it has not been independently validated.A non-differential misclassification of the outcome is therefore possible, which could bias the results toward the null. We explored this prospect in a sensitivity analysis using a stricter algorithm to identify ASD, and the risk estimates remained consistent. Also, it was difficult to estimate the appropriate exposure time window between antibiotic use and ASD diagnosis, because little is known about the underlying mechanism and latency period associated with impairment in neurodevelopment. In subsequent sensitivity analyses, we varied antibiotic exposure to within the first 6 and 18 months of life, and the risk estimates were not sensitive to changes in exposure time window. Finally, despite all our efforts to control for confounding by including multiple relevant covariates in the model and by using a sibling cohort, there was still a potential for unmeasured confounding from covariates that could not be obtained from the Repository. For example, we were not able to fully control for both maternal and paternal genetic contribution, as our sibling cohort used maternal siblings only. Also, we could not examine other factors that may impact on microbiota composition, such as diet, use of probiotics, metabolic changes etc. The complexity of gut microbiota and consequences following any changes make the association with childhood diseases a challenging one to investigate, and warrant further studies to examine such risk on both biological and population levels.

Conclusion

Our results suggest that antibiotics use during the first year of life is not associated with development of ASD. Lack of dose-response and, most importantly, lack of association in the sibling-controlled analysis confirm main analysis findings. We also found that the risk of ASD is highly susceptible to confounding by unmeasured shared familial factors. A sibling-controlled analysis limited such unmeasured confounding to a certain extent, and should be considered in all future observational studiesexamining risk of ASD.

Funding

This work was supported by the University of Manitoba and the Evelyn Shapiro Award for Health Services Research.

Acknowledgement

We would like to thank Charles Burchill and Heather Prior from the Manitoba Centre for Health policy for their valuable support. The authors acknowledge the Manitoba Centre for Health Policy for use of data contained in the Manitoba Population Research Data Repository under project #H2016: 244 (HIPC# 2016/2017–11). The results and conclusions are those of the authors, and no official endorsement by the Manitoba Centre for Health Policy, Manitoba Health, Seniors and Active Living or other data providers is intended or should be inferred. Data used in this study are from the Manitoba Population Research Data Repository housed at the Manitoba Centre for Health Policy, University of Manitoba, and were derived from data provided by Manitoba Health, Seniors and Active Living, Winnipeg Regional Health Authority, Manitoba Department of Families, Healthy Child Manitoba and Manitoba Education and Training.

Conflict of interest: None declared.