Study characteristics

Five cohort articles were identified examining the relationship between prenatal antibiotic exposure and childhood overweight or obesity (Table 1). The studies were conducted in the United States (n = 4) or in Denmark (n = 1) and the sample sizes ranged from 436 to 39,615 mother–child pairs. The prevalence of prenatal antibiotic exposure was 16% when intra-partum antibiotics were excluded,^{Reference Mueller, Whyatt and Hoepner23} 57.5% in the study that included intra-partum antibiotics^{Reference Cassidy-Bushrow, Sitarik and Levin21} and 25.2%–60.4% in the other three studies where it was unclear whether or not intra-partum antibiotics were included.^{Reference Wang, Liu and Zhang20, Reference Mor, Antonsen and Kahlert24, Reference Poulsen, Pollak, Bailey-Davis, Hirsch, Glass and Schwartz22} Childhood overweight/obesity ranged from 17.1% at 2 years of age^{Reference Cassidy-Bushrow, Sitarik and Levin21} to 24.7/8.9% (overweight/obesity) at 4 years of age^{Reference Wang, Liu and Zhang20} to ranges of 10.9%–25.2% at the ages of 7 and older.^{Reference Wang, Liu and Zhang20, Reference Mueller, Whyatt and Hoepner23, Reference Mor, Antonsen and Kahlert24}

Study findings

There were generally positive trends found in the relationship between any prenatal antibiotic exposure and childhood overweight, although this relationship was significant in just two of the five studies: that is, overweight: prevalence rate (PR): 1.26 (1.10–1.45); obesity: PR: 1.29 (1.03–1.62) in a Danish sample of 7–16 year olds^{Reference Mor, Antonsen and Kahlert24} and relative risk (RR): 1.77 (1.25–2.51) in 7 year olds from a sample who had mothers identifying as African-American or Dominican^{Reference Mueller, Whyatt and Hoepner23} (Table 1). Another study did not find an association with overweight or obesity categories, but with an increased BMI z-score at 3 years of age.^{Reference Poulsen, Pollak, Bailey-Davis, Hirsch, Glass and Schwartz22}

The significance of the relationship found in the study conducted by Mor et al.^{Reference Mor, Antonsen and Kahlert24} was mainly driven by exposure to three or more antibiotic prescriptions. Although the other three studies found a general insignificant relationship between prenatal antibiotic exposure and childhood overweight/obesity, they all did reveal significant relationships under certain conditions, that is, the exposure consisted of multiple antibiotic prescriptions in a sample where mothers were overweight^{Reference Poulsen, Pollak, Bailey-Davis, Hirsch, Glass and Schwartz22} and the exposure was multiple, mainly broad-spectrum antibiotic administration when overweight was measured at a later childhood age (7 versus 4 years of age^{Reference Wang, Liu and Zhang20}). Other conditions were that antibiotic exposure had occurred only during certain trimesters [i.e. only first trimester^{Reference Cassidy-Bushrow, Sitarik and Levin21} or only second trimester for overweight at 7 years of age (but not at 4 years)^{Reference Wang, Liu and Zhang20}].

Three of the four studies that had examined evidence for a dose–response relationship found a significant relationship only after multiple courses (i.e. 2+ courses when outcome was measured at age 7^{Reference Wang, Liu and Zhang20}, 3+ courses^{Reference Mor, Antonsen and Kahlert24} and 3+ courses by overweight mothers^{Reference Poulsen, Pollak, Bailey-Davis, Hirsch, Glass and Schwartz22}). Two studies examined antibiotic types and found that sulfonamides, trimethoprim and mixed antibiotics had a stronger effect size^{Reference Mor, Antonsen and Kahlert24} and that macrolides were associated with a higher BMI z-score.^{Reference Cassidy-Bushrow, Sitarik and Levin21}

Risk of bias in studies

Most studies had a medium risk of bias, usually scoring a 7 out 9, when assessed according to the NOS. Studies generally lost points by not taking important potential confounders into account (such as delivery mode, postnatal antibiotics or maternal and childhood infections) or by using self-reported as opposed to medically diagnosed measurements. There were other issues, however, that we believed could not be captured in the NOS scale, such as ambiguity at times about the types of analyses that had been conducted, and missing information, such as whether or not intra-partum antibiotics (administered just prior to or during delivery) had been included as part of the exposure prenatal antibiotics.

Heterogeneity between the studies on prenatal antibiotic exposure made them difficult to compare. There was heterogeneity with respect to the exposure (i.e. inclusion/exclusion of intra-partum antibiotics or inclusion not mentioned), various ages of outcome (ranging from age 2 in one study to 7–16 years in another study), the type of overweight measurements (i.e. BMI odd ratios versus z-scores) and the number and types of potential confounders taken into account.

Study characteristics

The four systematic reviews that had been retrieved were conducted in Denmark, China, United Kingdom and Canada each covering between 12 and 17 publications on infant antibiotic administration. The latest date of the search period conducted by these four systematic reviews was October 2017. Nine publications were covered by all 4 reviews and 12 publications by at least 3 reviews (see Box S2).

Main findings

The four systematic reviews varied in their overall conclusions (see Table 2). One concluded there was substantial evidence for infant antibiotic exposure increasing the risk of childhood overweight/obesity,^{Reference Shao, Ding and Wang33} two concluded there was a small association between infant antibiotic exposure and childhood overweight/obesity^{Reference Miller, Wu, Pedersen, de Klerk, Olsen and Burgner49, Reference Rasmussen, Shrestha and Bjerregaard34} and one concluded that there was no conclusive evidence of associations between infant antibiotic exposure and childhood overweight/obesity.^{Reference Partap, Allcock, Parker, Gurdasani, Young and Sandhu35} Reported pooled effect sizes found by the meta-analyses conducted in three of the systematic reviews were RR: 1.21 (1.09–1.33) for overweight and RR: 1.18 (1.12–1.25) for obesity (based on any exposure <24 months and any age of outcome from 2 to 12 years),^{Reference Shao, Ding and Wang33} odds ratio (OR): 1.11 (1.02–1.20) [based on any exposure (0–2 years) and any age of outcome from >2 to 12 years]^{Reference Rasmussen, Shrestha and Bjerregaard34} and OR: 1.05 (1.00–1.11) [based on earliest age of exposure (<24 months), latest age of outcome (2–18 years) and lowest number of prescriptions in each publication].^{Reference Miller, Wu, Pedersen, de Klerk, Olsen and Burgner49}

Table 2.

Summary of systematic reviews examining infant antibiotic exposure and childhood overweight or obesity

Notable influential factors with greater effects, as reported by these reviews, were multiple doses of antibiotics (a few studies within the reviews also reporting dose–response relationships), broad-spectrum antibiotics (macrolides being most consistently reported^{Reference Miller, Wu, Pedersen, de Klerk, Olsen and Burgner49}) and male gender.

Other factors investigated were the follow-up times. Miller et al.^{Reference Miller, Wu, Pedersen, de Klerk, Olsen and Burgner49} found that shorter follow-up times (<5 years) were significant, whereas longer follow-up times up to 10 years and above 10 years had positive effects, but were insignificant.

The investigation of maternal overweight was only mentioned by Partap et al.^{Reference Partap, Allcock, Parker, Gurdasani, Young and Sandhu35} who found that in the two publications conducting sub-studies according to maternal weight, a significant association was only found in mothers of normal weight. Partap et al. also mentioned two twin studies that showed an insignificant relationship between infant antibiotics and childhood overweight.

The authors of the four systematic reviews all reported the potential confounders that had been adjusted for in each study; factors that were commonly adjusted for included gestational age, birth weight, maternal age, socio-economic status and breastfeeding. All authors of the systematic reviews acknowledged the likelihood of there being more potential confounders influencing the relationship between infant antibiotic administration and overweight/obesity, such as childhood infections.

Assessment of studies

Using the critical appraisal tool by Aromataris et al.^{Reference Aromataris, Fernandez, Godfrey, Holly, Khalil and Tungpunkom29} and the ROBIS to assess the methodology and synthesis, it was apparent the four systematic reviews varied in the criteria they adhered to (Tables S1a and S1b). Partap et al. were the most and Shao et al. the least stringent with regard to the methodology, study syntheses and evaluations. Three out of the four reviews assessed the studies using the NOS, with varying scores and evaluations. Shao et al. rated their included studies the most favourably, assigning scores from 7 to 9 to all the studies and evaluating scores of at least 7 as low risk of bias. More details on how the scores were obtained were not reported in their review. Their more relaxed criteria with regard to assessing articles, pooling findings from heterogeneous studies, interpreting publication bias and including them in their meta-analysis likely contributed to their more assertive conclusions that there is a causal relationship between early-life antibiotics and childhood overweight.

Miller et al. and Rasmussen et al. also assessed the included articles using the NOS and in general had similar proportions of scores [Miller: low risk (four studies), moderate risk (eight studies) and high risk (three studies) and Rasmussen: low risk (two studies), moderate risk (eight studies) and high risk (three studies)].

Miller assessed the ‘comparability’ category of the NOS (covering the adequacy of adjusting for confounders) in the studies they reviewed much more favourably, giving 12/15 of their studies the maximum of two points if relevant potential confounders had been adequately accounted for. Their criteria were adjusting for at least one childhood and at least one maternal risk factor to gain the full two points. They did mention in their discussion, however, that (untreated) infections should be treated as potential confounders in future studies. Rasmussen et al. only awarded 2 of 13 included studies the full 2 points for ‘comparability’, due to inadequate adjustment for confounding, although in their list of potential confounders, they did not include childhood infections. They did mention in their discussion, however, that infections were likely to be of importance and that most studies did not address this. The studies rated as high risk of bias were not included in their meta-analyses. Other points were often lost by the authors of these systematic reviews for inadequate reporting of loss-to-follow-up or non-response rates and self-reported data.

The most stringent assessment of publications was conducted by Partap et al.^{Reference Partap, Allcock, Parker, Gurdasani, Young and Sandhu35} who also had the most cautious conclusions regarding the association between infant antibiotic exposure and childhood overweight. The authors mentioned not conducting a meta-analysis, due to the heterogeneity of all the studies. Although they did not use a standardised method, such as the NOS to assess the studies, they did consider important potential confounders that studies should have adjusted for to obtain more reliable results. Important confounders mentioned in their review were early childhood infections, delivery mode, exclusive breastfeeding, preterm birth and birth weight, maternal smoking, BMI and socio-economic status, current wheeze/asthma, lifestyle and siblingship. None of the studies adjusted for all of these factors. They concluded that most studies did not sufficiently examine confounding by indication. Other critical points raised by this review were small sample sizes and parent-reported antibiotic use.

The three studies who conducted meta-analyses also created funnel plots to test for publication bias. One study concluded there was no indication of publication bias (Shao et al., 2017),^{Reference Shao, Ding and Wang33} one study mentioned there was possible publication bias, but this would unlikely affect the pooled results^{Reference Miller, Wu, Pedersen, de Klerk, Olsen and Burgner49} and one study concluded there was some potential publication bias.^{Reference Rasmussen, Shrestha and Bjerregaard34}

Three of the four reviews recommended further studies using more standardised and similar measures in order to make comparability across the publications more clear and reliable.

The studies that were assessed by the authors of the systematic reviews had similar quality issues to the assessments we had made of the observational studies on prenatal antibiotic exposure, the majority scoring a 7 when assessed by the NOS (medium quality). Similar issues were encountered, including much heterogeneity in the studies (e.g. in exposure/outcome measurements and potential confounders), and the use of self-reported data.