Early and frequent exposure to antibiotics in children and the risk of obesity: systematic review and meta-analysis of observational studies

Background: This study aimed to systematically evaluate the available evidence on prenatal and early infancy antibiotic exposure and the association with overweight and obesity in later childhood. Methods: We conducted a comprehensive search of Embase, MEDLINE, and Web of Science for observational studies assessing prenatal and early antibiotic exposure on the risk of overweight and obesity. We independently assessed the risk of bias using the ROBINS instrument and the overall quality of evidence using the GRADE approach. Results: Our search identified thirteen observational studies including 554,983 participants; most studies were at moderate risk of bias. We found a statistically significant impact of early antibiotic exposure and the risk of being overweight later in childhood (OR 1.18; 95% CI 1.05 to 1.34) (very low quality evidence). We also found that early childhood antibiotic exposure was associated with the risk for childhood obesity (OR 1.14; 95% CI 1.04 to 1.24) (very low quality evidence). Conclusions: Very low quality evidence suggests that exposure to antibiotics early in life may be associated with an increased risk of being overweight and obese in later childhood. However, very low quality evidence raises serious questions about the plausibility of prenatal and early infancy antibiotic exposure being causally related to weight in children. PROSPERO registration: CRD42016050011 (14/12/2016)


Introduction
Infants and children are commonly and frequently prescribed antibiotics and up to 40% of infants are exposed either directly or through maternal intra-partum antibiotic prophylaxis 1,2 . In the United States it is estimated that a child receives approximately three courses of antibiotics by the age of two and 10 courses by the age of 10 years 3 and, often, these courses are prescribed for viral infections thus offering no therapeutic benefit 4 . The concerns with significant overuse of antibiotics are increased antibiotic resistance, increased rates of adverse drug reactions, such as rashes, fungal infections, and antibiotic-associated diarrhea. The intestinal microbiota begins development in utero and resembles adult microbiota by 2.5 years of age 4 . Thus, external exposures to antibiotics during this phase of microbiota development may potentially impact the normal colonization pattern, and the composition of the gut microbiota, both of which may play a role in health outcomes later in life, including weight gain and obesity 2,3,5 .
While the paradigm has been that the infant gut is sterile at birth, increasing evidence suggests that colonization may begin in utero with bacterial colonies detected in the placenta and meconium 6 . Thus, it is possible that prenatal or intrapartum antibiotic exposure may potentially affect these bacterial colonies in utero. Most notably, antibiotic exposure can affect the gut microbiota composition at any age; however, infants may especially be vulnerable, as the gut microbiota is highly unstable and dynamic during this period with greater interindividual variability, compared to that of an adult 7-9 . Cho et al. and Cox et al. have shown increased fat accumulation in mice who were treated with antibiotics early in age regardless of antibiotic class 10-12 . This corresponds with an absence of certain populations of prominent microbes, such as Lactobacillus, Allobaculum, Rikenellaceae, and Candidatus Arthromitus due to antibiotic exposure. This suggests a potential protective role of these bacteria against patient-important outcomes, such as weight gain and obesity 11,12 . As such, this paper aims to evaluate and summarize the available evidence on the potential impact of prenatal, intrapartum, and early childhood antibiotic exposure on the risk of overweight and obesity later in life.

Methods
This review was registered with PROSPERO -International Prospective Register of Systematic Reviews on the 14 th December 2016 under the number CRD42016050011 13 .

Search strategy
An experienced clinical librarian (TAW) conducted a literature search in the following electronic databases from inception to June 2018: Embase (Ovid), MEDLINE (Ovid), MEDLINE In-Process & Other Non-Indexed Citations (Ovid), MEDLINE ® Epub Ahead of Print (Ovid), and Web of Science. The search terms included database-controlled vocabulary and keywords for the concepts of "antibacterial agents" AND "children" AND "microbiome" OR "microbiota" AND "health outcomes" (e.g. weight, obesity, diabetes). We also supplemented the search by reviewing bibliographies of review articles and other eligible clinical studies to ensure that studies that were not identified by the search strategy were included. Clinicaltrials.gov was searched for unpublished and ongoing trials. No language or date restrictions were applied. See extended data for full search strategies 14 .
Study selection Inclusion criteria. Observational and experimental study designs were eligible, including cohort, cross-sectional, and casecontrol studies, data modelling studies, and randomized controlled trials. Further, to be eligible, studies had to examine maternal prenatal, intrapartum, and child (birth to 18 years) exposure to antibiotics and at least one of our target health outcomes of interest: overweight (body mass index [BMI] ≥ 25 kg/m 2 ), or obesity (BMI ≥ 30 kg/m 2 ). The comparison group included children that were not exposed to antibiotics. Studies with both fixed and variable follow-up periods were eligible. Studies or subgroups within studies that included premature infants, infants born with low birth weight or comorbidities, such as cystic fibrosis and Crohn's disease were excluded.
Screening, data extraction and quality assessment. Blinded to the journal of publication and results, two teams of independent reviewers (AS with either LL or HK) screened titles and abstracts of the studies to determine eligibility. Full-text articles were retrieved and assessed for further eligibility assessment. Discrepancies were resolved by discussion and, when necessary, additional input from a third reviewer (BCJ, KC).
Two reviewers (AS with either KC or HK) extracted data independently using a standardized data extraction form. The following data was extracted: study design, study setting, demographic information, antibiotic regimen (frequency and duration of exposures, type of antibiotic/class), and outcome data for each of our target outcomes including reported time points and duration of follow-up.
As no randomized trials were identified, the Risk of Bias In Non-randomized Studies (ROBINS) instrument for different types of observational studies 15 was used to assess study validity. Two reviewers (AS and KC) independently evaluated each observational study included for risk of bias (RoB). The ROB-INS instrument consists of seven total RoB questions, with two questions under the pre-intervention domain, one question on the intervention domain and four questions under the postintervention domain 15 . Response options for each question included 'low RoB', 'moderate RoB', 'serious RoB', and 'critical RoB'. We modified the instrument; wherein low and moderate risk of bias was classified as 'low risk of bias' and serious and critical risk of bias was classified as 'high risk of bias'. For each study, if one or more questions was judged to be high or critical risk of bias, the overall study was deemed to be at high risk of bias 15 . Any disagreements regarding data extraction or risk of bias items was resolved through discussion, and, when necessary, with an experienced methodologist (BCJ).
We applied the GRADE (Grading of Recommendations, Assessment, Development, and Evaluations) approach 16 for rating the overall quality of evidence for the outcomes of interest. In particular, observational studies were considered low quality evidence, but may be rated up for three reasons: (1) when a large magnitude of effects exists (e.g. OR <0.5 or >2.0), (2) when there is a dose-response gradient or (3) when all plausible confounding or other biases may be working against the observed effect 16 . Observational studies without these characteristics were considered low quality evidence. In addition, if studies were limited by risk of bias, inconsistency, indirectness, imprecision or publication bias the overall quality of evidence may to rated down to very low quality 17 . The quality of evidence for each main outcome was determined after considering each of these elements, and categorized as either high (highly confident that the true effect lies close to that of the estimate of the effect); moderate (moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different); low (confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect); very low (very little confidence in the effect estimate: the true effect is like to be substantially different from the estimate of effect) 17 .

Data analysis
Meta-analysis was conducted using Stata 14 (Stata Corp., College Station, TX, USA). Effect estimates and corresponding confidence intervals were extracted from eligible articles and we calculated the adjusted odds ratio (OR) with corresponding 95% confidence interval with the generic inverse variance method was used to determine relative effects 18 .
We conducted two primary meta-analyses, one for overweight and one for obesity with obesity including overweight. Random effects models were used due to the anticipated heterogeneity between studies. We used the DerSimonian and Laird method for estimating tau-squared and subsequent adjustment for effect size 19 .
Heterogeneity was assessed with the I-squared measure and the corresponding Q statistical test 18 . To further explore heterogeneity, we conducted a priori subgroup and sensitivity analyses 20 . Our three subgroups of interest included (1) the timing of exposure (prenatal versus early exposure), (2) number of antibiotic exposures (1 to 2 versus 3 or more), and (3) time point of outcome assessment (less than 7 years versus 7 years of age or more). Subgroups on timing and number of antibiotic exposures were stated a priori 13 , while the time point of assessment was a posthoc subgroup chosen based on the distribution of outcome assessments among eligible studies. We also conducted a sensitivity analysis on study design limitations (risk of bias). Low and moderate risk of bias was classified as low risk of bias and serious and critical risk of bias was classified as high risk of bias.

Results
Among 12,091 articles identified, 13 studies were deemed eligible for this review (Figure 1), including 11 cohort studies 2,21-25,26-30 , one cross-sectional study 31 , and one nested case-control study 1 . Five studies analyzed the effect of prenatal exposure to antibiotics 23,24,26,30,31 while the remaining eight studies evaluated the impact of early childhood (birth to 2 years) antibiotic exposure and the risk of overweight and obesity 1,2,21,22,25,[27][28][29] . Included studies ranged in size from 97 29 to 312,702 participants 25 with a total sample size among all included studies of 554,983. The median age for weight assessment was 7 years (range 2 years to 18 years). A detailed summary of all 13 study characteristics is shown in Table 1.

Risk of bias assessment
Using ROBINS instrument to independently assess risk of bias for weight-related outcomes, 11 studies were rated as 'moderate' for risk of bias 1,[21][22][23][24][25][27][28][29][30]31 and two studies were rated as 'serious' for risk of bias 2,26 . The studies rated as serious risk of bias were missing significant participant outcome data, with exclusion of participants introducing bias in the study. Despite all studies having adjusted for potential confounders, each included study was rated as moderate risk of bias for confounding. The choice and number of variables adjusted for differed considerably between studies. In some instances, the adjustment factors were theoretically not potential confounders (e.g. ethnicity), while in other instances, potential confounders such as socioeconomic status were not adjusted for. The complete risk of bias assessment for each individual study is provided in Table 2.

Overweight
We found statistically significant effect of antibiotics on the risk of being overweight (OR 1.18; 95% CI 1.05 to 1.34, 8 studies, 125,533 participants, I 2 =58.5%) ( Figure 2). Using the estimated population risk of being overweight 32 , the risk difference between exposed and non-exposed was equivalent to 26 more overweight cases per 1000 children (95%CI 7 more to 47 more cases). Our GRADE assessment indicates that the overall quality of evidence for the risk of being overweight was very low due to serious indirectness (Table 3).

Obesity
We found an overall significant effect of antibiotics on the risk of obesity (OR 1.14; 95% CI 1.04 to 1.24, 7 studies, 497,209 participants, I 2 =76.8%) (Figure 3). Based on the estimated population risk of obesity in children 14 , the risk difference between those exposed and non-exposed to antibiotics was equivalent to 9 more obesity cases per 1000 (95%CI 3 more to 15 more cases). Again, our GRADE assessment indicated that the overall quality of evidence for the obesity was very low due to serious inconsistency between studies (Table 3).
Subgroup and sensitivity analysis for overweight outcome Prenatal vs. early infancy antibiotic exposure. For prenatal timing of exposure, the effect of antibiotics on the risk of being overweight was not statistically significant, random effects (OR = 1.13, 95% CI 0.97 to 1.32, 4 studies, 54865 participants, I 2 =57.5%) ( Figure 4). For early infancy exposure, we found a significant effect of antibiotics on the risk of being overweight (OR = 1.26, 95% CI 1.0 to 1.57, 4 studies, 70668 participants, I 2 =54.8%). Our meta-regression analysis showed no significant difference between subgroups with regard to the effect of antibiotics on the risk of being overweight outcome (p =0.518).

Outcome assessment based on follow-up time point.
When the risk was assessed among those less than 7 years, we found a statistically significant effect of antibiotics on this risk of being overweight (OR = 1.22, 95% CI 1.05 to 1.42, 4 studies, 30952 participants, I 2 =0%). When the risk was assessed at 7 years or more, the effect of antibiotics on the risk of being overweight was not statistically significant, random effects OR = 1.18, 95% CI 0.98 to 1.43, 4 studies, 94581participants, I 2 =78.1%) ( Figure 5). Meta-regression analysis showed no significant difference between subgroups with regards to the effect of antibiotics on the risk of being overweight (p = 0.699).
Number of antibiotic exposures. For 1 to 2 antibiotic exposures, the effect of antibiotics of antimicrobials on the risk of being overweight was not statistically significant (OR = 1.30, 95% CI 0.94 to 1.81, 3 studies, 53941 participants, I 2 =84.1%) ( Figure 6). For 3 or more antibiotic exposures, the effect of antimicrobials on the risk of being overweight was not statistically significant, random effects (OR = 1.29, 95% CI 0.92 to 1.83, 3 studies, 53941 participants, I 2 =67.7%). Similarly, our meta-regression analysis showed no significant difference between subgroups for antibiotic exposure and overweight risk (p =0.937).

Sensitivity analysis: Risk of bias assessment.
For low risk of bias studies, we found a statistically significant effect of antibiotics on the risk of being overweight (OR = 1.18, 95% CI 1.02 to 1.37, 7 studies, 110992 participants, I 2 =61.6%) (Figure 7).
Subgroup and sensitivity analysis for obese outcome Prenatal vs. early antibiotic exposure. For prenatal timing of exposure, we found a significant effect of antibiotics on obesity (OR 1.32; 95% CI 1.01 to 1.73, 3 studies, 53,945 participants, I 2 =80.2%) and a non-significant effect for early infancy exposure (OR 1.07; 95% CI 0.99 to 1.16, 5 studies, 443,264 participants, I 2 =68.4%) ( Figure 8). However, subgroup analysis showed no significant difference between prenatal and early infancy subgroups on the risk of obesity (p = 0.353).

Number of antibiotic exposures.
For those with 1 to 2 antibiotic exposures, the risk of obesity was not statistically significant (OR 1.03, 95% CI 0.98 to 1.09, 4 studies, 70,199 participants, I 2 =0%), while for those with 3 or more exposures, we found no statistically significant effect (OR 1.29, 95% CI 0.92 to 1.83, 4 studies, 70,199 participants, I 2 = 76.1%) ( Figure 10). Again, subgroup analysis showed no significant difference between 1 to 2 exposures and 3 or more exposures (p = 0.085). The  One or more prognostic variables also predicts the intervention received at baseline. b Selection Bias: When exclusion of some eligible participants is related to both intervention and outcome, there will be an association between interventions and outcome even if the effect of interest is truly null. c Information Bias: Bias introduced by either differential or non-differential misclassification of intervention status. d Confounding: Bias that arises when there are systematic differences between experimental intervention and comparator groups in the care provided, which represent a deviation from the intended intervention. e Selection Bias: Bias that arises when later follow-up is missing for individuals initially included and followed (e.g. differential loss to follow-up that is affected by prognostic factors) f Information Bias: Bias introduced by either differential or non-differential errors in measurement of outcome data. Such bias can arise when outcome assessors are aware of intervention status, if different methods are used to assess outcomes in different intervention groups, or if measurement errors are related to intervention status or effects. g Reporting Bias: Selective reporting of results from among multiple measurements of the outcome, analyses or subgroups in a way that depends on the findings.
subgroup results, however, reached statistical significance with 3 or more antibiotics having a larger associated OR for obesity later in life.

Discussion
In total, 13 observational studies with over 554,983 participants were identified that examined the association between prenatal or early childhood antibiotic exposure and the risk of weight-related outcomes. Overall, based on very low quality evidence, antibiotic exposure may be associated with overweight and obesity. The absolute risk increase between those exposed and non-exposed to antibiotics was equivalent to 26 more  overweight cases per 1,000 children followed, and 9 more obesity cases per 1,000 children followed.
We specified a priori subgroup hypotheses to assess the stability of association across population and intervention variables thought to potentially modify our outcomes of interest. We found a statistically significant association between prenatal antibiotic exposure and obesity, between late antibiotic exposure (>7 years) and obesity and between higher frequency of exposures to antibiotics (3 or more scripts) and the risk of obesity. When we conducted a test of interaction for each of our subgroups, although none were statistically significant, the frequency of exposure was almost significant (p = 0.085), suggesting that higher antibiotic frequency may increase the risk of obesity later in life. This finding corresponds to our a priori hypothesis 13 that with higher exposure comes a higher risk of obesity, potentially attributable to a more frequent disruption of the composition of the gut microbiota 7-9 .
The quality of evidence for each outcome was determined using the GRADE criteria 17 . For the primary outcome, overweight, the quality of evidence was categorized as very low because of serious indirectness related to the measurement of weight, there were 5 different definitions (e.g. BMI from 25 to <30; BMI *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI:
Confidence interval; RD: Risk difference; OR: Odds ratio GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.

Very low quality:
We are very uncertain about the estimate.         > 85 percentile). For the obesity health outcome, the quality of evidence was categorized as very low due to indirectness (the reference control group was both normal weight and normal weight plus overweight) and inconsistency between studies (P < 0.001, I 2 = 76.8%), and our subgroup analyses did not explain the observed heterogeneity. Despite authors' adjustment for a variety of known or suspected confounders, observational studies are prone to residual confounding and after considering each of the GRADE criteria, including the possibility of rating up for magnitude of effect and for dose-response, the evidence provides only very low quality evidence.
The association between antibiotic exposure in early life and increased risk of being overweight or obese is also supported by observations from animal studies. A study conducted by Angelakis et al. (2013) 33 reported an increase in fat accumulation in mice when the mice were exposed to antibiotics at the weaning stage. The underlying mechanisms may be due to an upregulation of genes involved in lipogenesis, as well as antibiotic-induced changes in gut microbiota composition resulting in increased production of short-chain fatty acid, an additional energy source which results in imbalance of energy regulation, contributing to obesity 34,35 . In other studies, it has been suggested that microbial colonization may begin in utero through exchange of placental bacteria from mother to fetus; thus, placental transfer of antibiotics consumed during pregnancy may enter fetal circulation resulting in microbial changes similar to that of early life antibiotic exposure 10,36 .
Our paper has several strengths. First, our review included a systematic literature search of three primary databases and gray literature sources and is the most comprehensive review to date including 13 studies on weight-related outcomes. Second, an a priori design was published as a written protocol and registered with PROSPERO 13 . We closely followed our a priori protocol, however, we did make some post-hoc analysis and presentation decisions including the addition of a subgroup on follow-up time (up to 6 years vs 7 or more years of age). Third, we independently assessed the quality of the evidence for each outcome using the GRADE approach, allowing us to document the uncertainty we have in attributing antibiotic exposure as a causal risk factor. Fourth, we quantitatively analyzed our a priori subgroups of interest including prenatal versus early infancy antibiotic exposure, and frequency of exposure to explore hypothesized effect modifiers 1,22,26 , suggesting based on the available data, that frequency of exposure deserves further study.
This systematic review also has two major limitations. First, the search for included articles in this review was completed in June 2018. However, the discussion (below) has been updated to include recent evidence from studies published since spring 2018. Second, there was significant heterogeneity across the included studies with respect to the number and types of antibiotics used, the timeframe between antibiotic exposure and weight assessment, and the method of weight assessment. Although we minimized confounding by using the most adjusted analyses from each study in our meta-analyses, spurious results due to residual confounding remains a plausible explanation for all associations. That is, given that all the included studies were observational in nature, there is a risk of uncontrolled confounding factors despite multivariable adjustment 37 .
The potential relation between early infancy antibiotic exposure and weight issues for children is a burgeoning field of investiga-  39 conducted a systematic review and meta-analysis of 19 total studies. The outcomes of interest were childhood overweight or obesity, difference in childhood body mass index (BMI) and weight between the exposed group and the non-exposed group. The study found a significant association between early childhood antibiotic exposure (<2 years) and risk of childhood weight gain and obesity (OR 1.05, 95% CI 1.04 to 1.06). The study also conducted a subgroup analysis involving time point of exposure. Among infants exposed to antibiotics before 6 months of age, the odds of more weight gain was 11%, while for infants exposed to antibiotics after 6 months of age, the odds of more weight gain was 7%. However, unlike our review, Rasmussen et al. In addition to having a publicly available protocol, our study is the only systematic review and meta-analysis to rate the quality of the evidence for each outcome using the GRADE approach, and to present data as an absolute risk difference, an approach to presenting results that has been shown to be more intuitive and helpful for decision-makers 40,41 .
Our findings must be interpreted in the light of a recent crosssectional study 42 of 284,211 participants that included siblings and twins in New Zealand and that analyzed prenatal and early infancy antibiotic exposure (first two years) and its association with obesity at 4 years. The study found that both prenatal and early infancy exposure to antibiotics were independently associated with obesity at 4 years in a dose dependent manner. For the child's exposure, the OR for the association between antibiotic exposure and obesity was 1. Overall, our systematic review suggests that prenatal and early childhood antibiotic exposure is associated with an increased associated risk of overweight and obesity among children and adolescents, potentially independent of more established early determinants of obesity. Our findings suggest that the first few years of life may represent a critical window of development, where external exposures, such as antibiotics, may program metabolic pathways and obesity risk through mechanisms involving the gut microbiota; however, experimental studies are needed to establish the impact of antibiotics, particularly frequent antibiotic use, on the gut microbiota and how this may impact weight-gain in humans. Until registered protocol-driven higher quality cohort studies with explicit plans for adjustment and statistical analysis that better demonstrate an antibiotic dose-response curve or controlled clinical trials (e.g. watchful waiting for otitis media) with long-term follow-up are conducted to confirm or refute these findings, very low quality evidence raises serious questions about the plausibility of prenatal and early antibiotic exposure being causally related to weight in children.

Data availability
Underlying data All data underlying the results are available as part of the article and no additional source data are required. pre-planned analyses of subgroups are different in the final paper than in the PROSPERO record [i.e. pre-planned analyses not done: subgroup analyses for boys versus girls; the effect of broad vs. narrow spectrum of antibiotics will be analyzed for all outcomes; we will assess the potential impact of early exposure (less than 3 years of age) to antibiotics versus later exposure (3 years of age or greater) (but done for 7y vs >=7y)].

○
In line with the above commentaries, the cited sentence should be changed accordingly ("a priori design was published as a written protocol and registered with PROSPERO13. We closely followed our a priori protocol").
2a: In the methods section, the main outcomes of interest (overweight and obesity) are defined as BMI >25 kg/m2 for overweight and >30 kg/m2 for obesity, which is typically used in adults. It should be explained why the authors did not consider (in the methods) different, more appropriate outcomes for children definitions of overweight/obesity (i.e. >1SD for overweight and >2SD for obesity in children 5-19y in accordance to WHO). As most of the included studies were reporting the main outcome in percentiles -therefore the methods section should be changed accordingly.
2b: In the PROSPERO protocol, pre-specified outcomes were 1. Overweight 2. Obesity. Please clarify what was behind the decision to group the second outcome as Overweight + Obesity? Why was the pre-planned outcome of "obesity only" not reported at all? There is at least a theoretical chance that the overweight effect was big enough to make a grouped outcome of overweight + obesity also significant, even if obesity alone was non-significant (alone). It would be of interest for all readers to complement this review with analysis for the obesity outcome only (as was preplanned).
3: As the literature search was conducted >24 months apart of completion of the review, and the Reviewers report other relevant studies in their discussion, one should consider updating the search, as following guidance from the Cochrane Handbook: "Reviews that are out of date and do not incorporate all the available evidence risk providing misleading information to decision makers and other stakeholders." https://training.cochrane.org/handbook/current/chapter-iv.
The flow diagram shows that the authors identified 12116 (12091+25). However, in the manuscript there is the number 12091 -please clarify.

1.
Could the authors provide the To increase readability, numerical references should be added in Table 1 near each included study.