Early Life Exposure to Antibiotics and Autism Spectrum Disorders: A Systematic Review

In this review on pre- and postnatal antibiotic exposure and subsequent risk of autism/ASD, a total of 11 observational studies were included. In two identified cohort studies on prenatal antibiotic exposure performed in overlapping populations, a slightly increased risk of ASD/infantile autism after the use of various antibiotics was observed. However, 3 of 4 identified case–control studies did not report any significant association between prenatal antibiotic exposure and autism risk, whereas one reported a positive association based only on univariate analysis. Three identified cohort studies on early childhood antibiotic exposure reported conflicting evidence. In one cohort study, a small trend towards a reduced risk of ASD in infants exposed to various antibiotics in their first year of life was observed, which was not significant in secondary analyses based on sibling-controlled design. In another cohort study, a borderline increased risk of ASD was observed with exposure to certain types of antibiotics in a standard model but no association was observed in a between-within sibling model. Finally, in the third study on a cohort overlapping with the previous one, a significantly higher risk of autism after exposure to various classes of antibiotics was observed. Data from case–control studies on antibiotic exposure during childhood were more consistent. All three of them reported significant, positive associations between antibiotic use and later ASD diagnosis.

To our knowledge, no previous systematic review has examined antibiotic use during pregnancy and early childhood as a potential risk factor for ASD. A number of meta-analyses have reported data on various infections preceding ASD diagnosis (Modabbernia et al. 2017), and some have found significant positive associations with infections occurring during pregnancy (Jiang et al. 2016; Gardener et al. 2009). Additionally, one of those systematic reviews reported increased risk of ASD after prenatal exposure to different medications, including antibiotics, which were, however, not assessed separately (Gardener et al. 2009). In another meta-analysis, no associations between early postnatal infections and subsequent risk of ASD were found (Gardener et al. 2011). On the other hand, some observational studies have found evidence of a higher risk of mental disorders after various infections in older children and in adults (Benros et al. 2011; Kohler et al. 2017). Recently, more scientific attention is being drawn towards effects of early-life antibiotic exposure on long-term health outcomes, including reports of its negative influence on neurocognitive development (Slykerman et al. 2019) and risk of obesity (Miller et al. 2018).

The main strength of this review is its novelty, as no previous systematic review has focused on the association between early-life antibiotic exposure and ASD. Our study utilized a broad search strategy, complemented by an extensive reference hand search, so the possibility of missing important published data is low. A relatively narrow clinical question allowed us to focus in detail on the included studies and enabled a clear and meticulous presentation of findings. However, this review also has considerable limitations. Compared to the number of other potential environmental risk factors for ASD, antibiotic exposure turned out to be rarely assessed (Wang et al. 2017; Modabbernia et al. 2017). Thus, only 11 studies were finally included in the review, the majority of which were of case–control design. After division into groups depending on study type and timing of exposure, small groups of 2 to 4 studies were created. Methods of data reporting in these studies were heterogeneous, further limiting the capacity to perform either qualitative or quantitative synthesis of the data. Moreover, each of the included case–control studies was characterized by a significant potential risk of bias. Another limitation of this study is the inclusion of only published data, creating a risk of publication bias. Finally, we excluded a number of studies in which infections were explored as potential risk factors for ASD. Some types of these infections are almost always treated with antibiotics (e.g., urinary tract infections during pregnancy, severe infections in newborns), so we could have assumed them to be a proxy for antibiotic exposure. However, we decided to use only direct data on exposure.

Results of the included studies do not provide conclusive evidence to support the hypothesis that prenatal antibiotic exposure is associated with ASD. Although a number of positive associations was reported in two large cohorts (Atladottir et al. 2012; Wimberley et al. 2018), their populations were largely overlapping, so they cannot be treated as independent studies for the purpose of this review. Moreover, authors of the study with borderline significant results (Atladottir et al. 2012) emphasize, that within their analysis they made 106 comparisons with no adjustments for multiple testing, so this borderline significant association between prenatal antibiotic exposure and ASD could have been a chance finding. While this study was specifically designed to examine prenatal risk factors of autism and tested for a wide range of potential confounders, the other one focused on postnatal factors and hardly performed any adjustment in the prenatal exposure analysis. Because of that, in our opinion the results of the former study are more meaningful.

On the other hand, all of the case–control studies reported no association between prenatal antibiotic exposure and ASD, except for one that reported higher risk of autism in exposed patients (Mrozek-Budzyn et al. 2013). However, this study was performed in a relatively small group of children, and strangely, its finding concerning antibiotic exposure was not explored in a multivariate analysis. Moreover, compared to the literature (Santos et al. 2010; de Jonge et al. 2014) an unexpectedly low percentage of women in the control group (2.1%) was reported to have used antibiotics during pregnancy, which may suggest a bias within the exposure data.

Results of studies on postnatal antibiotic exposure are conflicting. One cohort study reported a reduced risk of ASD in children after antibiotic use (Hamad et al. 2018). However, since the main model was based only on data from administrative databases, the authors performed secondary analyses based on a sibling-controlled design. In those analyses, antibiotic use was not associated with the risk of developing ASD, hence, the authors concluded that the marginal association observed in the main model was unlikely to be meaningful. In another cohort study (Axelsson et al. 2019), the between-within sibling design analysis did not confirm the findings from the standard design analysis of higher ASD risk after use of certain types of antibiotics. We used data from these two studies to perform a meta-analysis, and its results revealed no significant association. The final included cohort study (Wimberley et al. 2018) reported a positive association between postnatal antibiotic exposure and autism risk. However, since its population largely overlapped with that of the previous cohort, the two cannot be treated as independent studies for the purpose of this systematic review. Besides the lack of a sibling analysis in the latter cohort, the difference in results might have originated from the differences in the exposure assessment. The authors of the latter study defined ‘antibiotic use preceding the diagnosis of autism’ as the antibiotic received before the first diagnosis of autism registered in the national healthcare databases. Median age for the first autism diagnosis in this study was 7.1 years, which is much later than both median age of ASD diagnosis reported in the literature and median age of first developmental concerns in ASD children (Baio et al. 2018). Therefore, even for the subgroup of children exposed to antibiotics before fifth birthday it is unclear, if the symptoms of autism were absent before the first antibiotic administration. In our opinion, this might have introduced considerable bias.

All of the remaining case–control studies reported a positive association between postnatal antibiotic exposure and ASD. All of these studies were characterized by considerable potential sources of bias, including small sample sizes (Grossi et al. 2018; Niehus and Lord 2006), self-selection bias (Niehus and Lord 2006; Bittker and Bell 2018) reflecting differences between those who choose to participate in a study and those who do not, potentially biased control group selection (Grossi et al. 2018; Niehus and Lord 2006), and/or lack of verification if the ASD diagnosis in the case group was made by a professional (Bittker and Bell 2018). Also, in all of these studies, the exposure data were based on parental reports, so a risk of recall bias (Coughlin 1990), especially over-reporting the exposure in ASD groups and under-reporting in control groups, cannot be ruled out.

Data on specific antibiotic types and risk of ASD are also inconsistent. In one study, an increased risk of ASD was reported after prenatal exposure to sulfonamides, penicillins and macrolides, but not cephalosporins (Atladottir et al. 2012). This is consistent with the evidence that macrolides induce distinctive changes in the microbiome composition of children (Korpela et al. 2016). On the other hand, expected microbiota alterations after the use of cephalosporins should be at least as strong as those after use of penicillins, due to cephalosporins’ broader spectrum (Biedermann and Rogler 2015). Accordingly, another study reported a stronger association between ASD and both pre- and postnatal exposure to broad spectrum antibiotics (cephalosporins, tetracycline, trimethoprim-sulfonamide combination, aminoglycosides, and quinolones) compared to moderate-spectrum antibiotics (Wimberley et al. 2018). However, no significant associations were found between autism risk and exposure to narrow-spectrum antibiotics, including macrolides. In two other cohort studies, no significant associations between ASD and individual types of antibiotics were described (Axelsson et al. 2019; Hamad et al. 2018). Individual antibiotic types were analyzed and reported differently in each of the studies included in our systematic review; therefore, a meta-analysis was not possible. According to the available evidence, some patterns of microbiota composition in individuals with ASD have been identified, mostly regarding the abundance of different Clostridia and Bacteroidetes species and an altered Bacteroidetes/Firmicutes ratio (Rosenfeld 2015). Such changes may be induced both by macrolides and broad-spectrum antibiotics (Korpela et al. 2016). On the other hand, microbiota alterations in different individuals might not follow a specific pattern even after use of the same antibiotic (Yassour et al. 2016). Apart from the disturbance of the microbiome-gut-brain axis, to our knowledge no well-documented alternative explanations for the antibiotic-ASD association have been proposed.

A question concerning all of the included studies is the difficulty in differentiating whether the observed effects were due to the antibiotics themselves or were due to the antibiotics acting as a proxy or mediator of an underlying infection. Only two of the included studies controlled for infections occurring during the same period as antibiotic exposure (Hamad et al. 2018; Guisso et al. 2018). None of the included studies investigated differences in risk of autism resulting from specific indications for antibiotic therapy. Three included studies reported a higher risk of autism after maternal respiratory infections (Atladottir et al. 2012; George et al. 2014), fever episodes and influenza (Atladottir et al. 2012), or “infections” in general (Guisso et al. 2018); two studies found significant positive associations between early childhood otitis media episodes and autism (Wimberley et al. 2018; Niehus and Lord 2006). Such associations of prenatal and early-life infections with autism have been documented before (Jiang et al. 2016; Atladottir et al. 2010), and they may also be supported by underlying biological mechanisms (Singh 2009; Ponzio et al. 2007; Rosenhall et al. 1999).

All of the included studies in which the assessment of exposure was based on parental reports collected after diagnosis of ASD have the possibility of recall bias. Knowledge of a variety of potential exposures that are considered as risk factors for ASD is common among parents of children with ASD (Chaidez et al. 2018); thus, they may tend to analyze the previous medical history of their children more and report medication use more often than parents of the control children. This is a similar effect to the one observed after past claims of the supposed vaccination-autism association (Andrews et al. 2002). This may suggest a greater reliability of antibiotic exposure data collected prospectively or derived from medical records.

Yet another problem in the interpretation of possible antibiotic-ASD associations is the fact that, despite stable antibiotic use in Europe and USA in recent years (European Centre for Disease Prevention and Control 2018; Klein et al. 2018), the prevalence of ASD is actually rising in these regions (Lyall et al. 2017). Nevertheless, given the heterogeneous etiology of autism, it is difficult to exclude the possibility that reduced antibiotic use is outweighed by other environmental factors contributing to ASD risk.

Currently available data from observational studies do not conclusively support the hypothesis that prenatal or postnatal exposure to antibiotics as a group is a risk factor for ASD. However, a portion of the results from large cohort studies may suggest a possible trend towards higher risk of ASD after use of certain types of antibiotics. More studies, most preferably of a prospective cohort design, controlling adequately for potential confounding factors including infections, are needed to verify this association.