Association of infant antibiotic exposure and risk of childhood asthma: A meta-analysis

Abstract

Background

Infant antibiotic exposure may be associated with childhood asthma development.

Objective

To examine and detail this association considering potential confounders.

Study design

PubMed, EMBASE, Web of Science, and the Cochrane Library were searched for publications from January 2011 to March 2021. Eligible studies were independently reviewed to extract data and assess quality. Random effect model was used to pool odds ratio (OR) and corresponding 95% confidence intervals (CIs).

Results

A total of 52 studies were included. The association of infant antibiotic exposure and childhood asthma was statistically significant for overall analysis (OR, 1.37; 95% CI, 1.29–1.45) and for studies that addressed reverse causation (RC) and confounding by indication (CbI) (1.19; 95% CI, 1.11–1.28). Significance remained after stratification by adjustment for maternal antibiotic exposure, medical consultation, sex, smoke exposure, parental allergy, birth weight, and delivery mode. In detailed analyses, macrolides (OR, 1.56; 95% CI, 1.31–1.86), antibiotic course≥5 (OR, 1.79; 95% CI, 1.36–2.36), exposure within 1 week of birth (OR, 1.82; 95% CI, 1.34–2.47), asthma developed among 1–3 years (OR, 1.84; 95% CI, 1.63–2.08), short time lag between exposure and asthma onset (OR, 2.05; 95% CI, 1.91–2.20), persistent asthma (OR, 2.61; 95% CI, 1.49–4.59), and atopic asthma (OR, 2.14; 95% CI, 1.58–2.90) showed higher pooled estimates.

Conclusion

Infant antibiotic exposure is associated with increased risk of childhood asthma considering confounding, and the association varied with different settings of exposure and outcomes. This highlights the need for prevention of asthma after early antibiotic exposure. Heterogeneity among studies called for caution when interpretation.

Introduction

Asthma, with an estimated worldwide prevalence of 14% among children,¹ is one of the most common children's chronic respiratory diseases. Asthma development is considered to be susceptible to genetic and environmental factors.² Human microbiota, as an environmental factor, play a key role in immune modulation, and are thus protective for asthma development.3, 4, 5 Recently, numerous studies suggested that infant antibiotic exposure can reduce the gastrointestinal microbiome diversity⁶ and affect the balance between microbiota and immune system during a period when microbiome changes rapidly.⁷ As a result, antibiotic use in infancy may be particularly considerable to childhood asthma development.

However, studies investigating the association between infant antibiotic exposure and childhood asthma development produced conflicting conclusions. This is because the association is subject to various confounding factors.⁸ Besides family, delivery, and environmental factors, remarkably, reverse causation (RC) and confounding by indication (CbI) are 2 strong confounders.^9,10 RC occurs when early-onset symptoms of asthma are treated with antibiotics. CbI occurs when antibiotics are prescribed for respiratory infection (RI), which alters the microbiome through a cascade of host immune responses,¹¹ and is an independent risk factor for asthma. Both of these factors will lead to false causal effect. Previous meta-analyses published in 2011 pooled studies that addressed RC and CbI showed small effect of antibiotic on childhood asthma.^12,13 However, they included studies with either relatively short follow-up or short exposure period or long interval between exposure and outcome, which may lead to an underestimated effect. Additionally, these reviews lack sufficient details such as antibiotic timing, age of asthma.

Growing number of studies that addressed RC, CbI, and some emerging confounders, as well as details on different settings of antibiotic exposure and asthma outcome have been published since then, with inconsistent findings. Given the conflicting evidence, limitations of the previous meta-analyses, and availability of new data, we conducted the present meta-analysis to examine the association of infant antibiotic exposure and risk of childhood asthma considering the effect of potential confounders, and to detail the association based on different settings of exposure and outcome.

Method

This meta-analysis and systematic review was conducted and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA).¹⁴

Search strategy

PubMed, EMBASE, Web of Science, the Cochrane Library, and Medline were searched for English publications from January 2011 to March 2021 as there were meta-analyses published in 2011. The following search terms were used: (antibiotic∗) and (child OR pediatric OR “early life” OR infan∗ OR neonatal) and (asthma OR wheezing OR wheeze). Additional studies were found by searching reference lists of relevant articles.

Study selection

Retrieved articles were imported to Endnote. After removing duplications, 2 authors independently reviewed the articles and determined appropriateness for the inclusion. Where consensus on eligibility could not be achieved, a third author was involved. The inclusion criteria were: 1) Studies examined the association of infant antibiotic exposure and childhood asthma or wheezing; 2) Studies were observational studies; 3) Studies provided risk estimates (odds ratio [OR], risk ratio [RR], hazard ratio [HR]) and 95% confidence intervals (CIs); 4) Antibiotic exposure occurred within 3 years of life; and 5) Age of asthma or wheezing ranged from 1 to 18 years. The exclusion criteria were: 1) Studies only examined the effect of antibiotic on asthma exacerbation; 2) Estimate of mixed exposure or outcome was given, rather than a specific estimate of antibiotic exposure or asthma outcome; and 3) Prenatal and postnatal antibiotic exposures were mixed for analysis. Asthma was defined as persistent or recurrent broncho-obstructive symptoms with a good response to bronchodilators and assessed either by questionnaires or doctor's diagnosis or anti-asthmatic drug prescriptions. Studies reporting wheezing as the outcome were also included in the main analysis. Nevertheless, given the fact that not all wheezes are truly a condition of asthma, we conducted detailed analyses stratified by asthma definition and age of “asthma” onset.

Data extraction

Two authors independently extracted data from all eligible studies using a standardized data collection form. The extracted data included: first author, publication year, country/region, race, study design, sample size, number of cases, antibiotic exposure timing, age of asthma, diagnosis and measurement of exposure and outcome, effect size and corresponding 95% CI, correction for RC and CbI, and adjustment model. If an article contained separate studies, each study was deemed as a single record and its effect size was extracted. If more than 1 article reported the findings of the same study, the effect size of study reporting longer follow-up was extracted. If there were multiple estimates in a study, the effect size that was judged to be the least biased (eg, adjusted for the most confounders, addressed RC and CbI) was extracted for analysis. For gender-stratified studies, effect sizes of boy and girl were extracted respectively. For indication-stratified studies, effect size of antibiotic prescribing for non-respiratory infection was extracted.

Statistical analysis

Meta-analysis was conducted using STATA 15.0 to generate pooled ORs. When HRs or RRs were provided, we assumed HRs were similar to RRs as previous studies did.^12,15 Since the prevalence/incidence rate of asthma is generally low among included studies, we assumed that ORs approximated the RRs.^16,17 Heterogeneity was assessed by Cochrane Q test (P < 0.05 indicated statistically significant) and Higgins I² test (I²>50% and I²>75% indicated moderate and high heterogeneity, respectively). Due to the detected heterogeneity, random effect models were used for all analyses. Random effect models allow both within-study and between-study variances, and are more conservative.¹⁸ We conducted the following analyses to evaluate the association of infant antibiotic exposure and childhood asthma: 1) An overall analysis including the least biased estimates of all studies and stratified analyses by measurement of exposure and outcome, study design type, HR/RR/OR, and country/region; 2) Subgroup analyses based on the adjustment for the common factors; 3) Analysis exclusively for studies that addressed RC and CbI. Studies that additionally adjusted for common factors were further included in a restricted analysis; and 4) Detailed analyses regarding antibiotic type, course, timing, asthma definition, phenotype, age of onset, and time lag between exposure and asthma onset. Antibiotic course refers to the frequency of interval antibiotic prescription within the defined exposure time. To describe a dose-response relationship between antibiotic course and risk of childhood asthma, we assigned a value to each antibiotic course category using the midpoint of category (eg, value of 1.5 for the category of 1–2 antibiotic course) or 1.5 times the lower limit of open-ended category (eg, 7.5 for the category of ≥5 course of antibiotic) to fit a linear or non-linear model.¹⁹

Meta-regression was performed for region, publication year, study type and quality, sex ratio, exposure timing, outcome definition, asthma onset age, measurement, measures of effect size, and common confounders that were used as stratified factors in the results below (eg, RC/CbI) to identify sources of heterogeneity. Sensitivity analysis was performed to examine the stability of the pooled results by removing one study each time and assessing the change of the pooled estimates. Publication bias was evaluated using Begg's and Egger's test (P < 0.10 was considered significant). Funnel plot was also presented. Study quality was assessed using the Newcastle–Ottawa Quality Assessment Scale (NOS) for cohort and case-control studies and Agency for Healthcare Research and Quality scales for cross-sectional studies. Although some items of NOS has been questioned²⁰ and a ROBINS-I tool is recommended for non-randomized studies,²¹ we choose NOS in this review because it can be simply applied to assess the quality of case-control studies and cohort studies separately. By comparison, NOS almost covered the range of bias assessed by ROBINS. The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) criteria were also used to assess the quality of evidence for studies that addressed RC and CbI on the following domains: risk of bias, inconsistency, imprecision, indirectness, and publication bias.²²

Results

Literature search

We identified 1521 non-duplicate articles. After applying the inclusion and exclusion criteria, 112 articles were reviewed in detail, of which 47 articles were eligible for the present meta-analysis 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69 (Fig. 1). Among these articles, 5^{25,39,46,56,60} included 2 separate studies in the same article. As a result, totally 52 studies out of 47 articles involving 2 742 140 patients were included. The study by Gupta et al⁴⁹ met the exclusion criteria by exceeding age of outcome limits (0–21 years), but it was retained because its calculated mean age was 6 years.

Fig. 1.

Flow diagram of selecting studies.

Study characteristics

The characteristics of the included studies were summarized in Table S1. Most exposure was limited to the first 2 years, with a wider range of asthma onset age (1–21 years). The majority of studies were prospective studies (n = 22) and adopted information measurement of self-report (n = 32). All included studies reported adjusted effect size except for 2 studies.^47,66 Quality scores of the included studies were presented in Table S1 and displayed in Fig. S1. Overall, 21 studies were of medium quality and 31 studies were of high quality.

Association of infant antibiotic exposure and childhood asthma

The overall analysis pooling from the least biased estimates of all studies showed that infant antibiotic exposure was significantly associated with childhood asthma (OR, 1.37; 95% CI, 1.29–1.45; I² = 97.4%, P < 0.01) (Table 1). Cross-sectional studies pooled a higher OR (1.48; 95% CI, 1.33–1.64, I² = 45.0%, P = 0.04), and was followed by retrospective studies (OR, 1.37; 95% CI, 1.27–1.47, I² = 87.4%, P < 0.01). Prospective studies indicated the lowest but still significant pooled OR (1.32; 95% CI, 1.22–1.43, I² = 96.6%, P < 0.01).

Table 1.

The overall analysis and subgroup analyses of childhood asthma risk associated with infant antibiotic exposure.

Subgroup types	Number of studies(n)	Heterogeneity		Pooled OR (95%CI)
		I2 (%)	P-value
Overall analysis	52	97.4	<0.01	1.37 (1.29–1.45)
OR	38	95.6	<0.01	1.44 (1.31–1.59)
RR	3	99.5	<0.01	1.16 (0.85–1.59)
HR	11	95.0	<0.01	1.33 (1.21–1.46)
Measurement of exposure and diagnosis
Databasea	20	98.8	<0.01	1.33 (1.22–1.44)
Self-report	32	86.4	<0.01	1.45 (1.32–1.59)
Study design typeb
Prospective studies	22	96.6	<0.01	1.32 (1.22–1.43)
Databasea	10	98.1	<0.01	1.32 (1.20–1.46)
Questionnaire	12	64.8	<0.01	1.38 (1.16–1.64)
Retrospective studies	11	87.4	<0.01	1.37 (1.27–1.47)
Databasea	8	89.6	<0.01	1.36 (1.21–1.53)
Questionnaire	3	63.3	0.07	1.41 (1.29–1.54)
Cross-sectional studies	14	45.0	0.04	1.48 (1.33–1.64)
Study country/region
US	12	99.2	0.01	1.34 (1.17–1.52)
Europe	23	95.5	<0.01	1.36 (1.24–1.50)
Asia	12	75.0	<0.01	1.38 (1.24–1.54)
Otherc	5	37.4	0.17	1.47 (1.22–1.77)
Adjustment for RC
Yes	35	98.0	<0.01	1.34 (1.25–1.44)
No	17	83.6	<0.01	1.47 (1.29–1.68)
Adjustment for CbI
Yes	27	97.9	<0.01	1.29 (1.19–1.41)
No	25	77.9	<0.01	1.43 (1.35–1.53)
Adjustment for gender, smoke exposure, and parental allergy
Yes	13	68.3	<0.01	1.35 (1.22–1.50)
No	39	96.5	<0.01	1.36 (1.27–1.46)
Adjustment for gender, birth weight, and delivery mode
Yes	10	89.5	<0.01	1.23 (1.15–1.32)
No	42	95.9	<0.01	1.41 (1.31–1.51)
Adjustment for maternal antibiotic exposure
Yes	6	90.7	<0.01	1.25 (1.08–1.46)
No	46	97.6	<0.01	1.39 (1.30–1.48)
Adjustment for medical consultation
Yes	8	94	<0.01	1.16 (1.00–1.35)
No	44	97.4	<0.01	1.42 (1.33–1.51)
Qualityd
High	31	97.3	<0.01	1.30 (1.22–1.38)
Medium	21	97.5	<0.01	1.56 (1.29–1.89)

Abbreviations: OR, odds ratio; CI, confidence interval; RC, reverse causation; CbI, confounding by indication.

^aDatabase studies often define a asthma diagnosis using a combination of International Classification of Diseases code and drug prescription records.

^bAs 2 studies did not report adjusted effect size,^47,66 1 study included adult patients,⁴⁹ and 1 study involved premature infants,³⁵ these studies were excluded in the first subgroup analysis to reduce the sources of heterogeneity.

^cOther countries/regions included Angola, Costa Rica, Cuba, Brazil, and Canada.

^dFor cohort studies and case-control studies, score of 4–6 was considered as medium quality and 7–9 as high quality. For cross-sectional studies, score of 4–7 was considered as medium quality and 8–11 as high quality

We further stratified studies by adjustment for common confounders (Table 1). The pooled OR in the subgroup adjusted for medical consultation (OR, 1.16; 95% CI, 1.00–1.35), CbI (OR, 1.29; 95% CI, 1.19–1.41), sex, birth weight and delivery mode (OR, 1.23; 95% CI, 1.15–1.32), maternal antibiotic exposure (OR, 1.25; 95% CI, 1.08–1.46), RC (OR, 1.34; 95% CI, 1.25–1.44), sex, smoke exposure and parental allergy (OR, 1.35; 95% CI, 1.22–1.50) and studies of high quality (OR, 1.30; 95% CI, 1.22–1.38) showed lower estimates. However, the association of infant antibiotic exposure and childhood asthma still remained in all subgroups.

Confounding by RC and CbI

Studies that addressed RC and CbI pooled an attenuated but statistically significant association (1.19; 95% CI, 1.11–1.28; I² = 95.3%, P < 0.01) (Table 2) (Fig. S2). To further analyze whether this significance was biased by other confounding, we restricted studies to those additionally adjusting for common factors. Results indicated that studies additionally adjusting for maternal antibiotic exposure (OR, 1.11; 95% CI, 1.01–1.23), sex, birth weight and delivery mode (OR, 1.15; 95% CI, 1.11–1.20), and sex, smoke exposure and parental allergy (OR, 1.18; 95% CI, 1.15–1.21) had lower risk. When stratified by study type, the association remained in prospective studies (OR, 1.24; 95% CI, 1.12–1.37, I² = 97.4%, P < 0.01), while almost disappeared in retrospective studies (OR, 1.15; 95% CI, 1.00–1.32, I² = 77.7%, P < 0.01) and disappeared in cross-sectional studies (OR, 1.11; 95% CI, 0.99–1.24, I² = 0.0%, P = 0.60).

Table 2.

The particular analysis and restricted analysis for studies that addressed RC and CbI.

Subgroup types	Number of studies(n)	Heterogeneity		Pooled OR (95%CI)
		I2 (%)	P-value
Studies that addressed RC and CbI	23	95.3	<0.01	1.19 (1.11–1.28)
OR	15	57.6	0.01	1.14 (1.06–1.23)
RR	2	94.8	<0.01	1.25 (0.86–1.82)
HR	6	87.2	<0.01	1.20 (1.09–1.32)
Study type
Prospective studies	12	97.4	<0.01	1.24 (1.12–1.37)
Retrospective studies	7	77.7	<0.01	1.15 (1.00–1.32)
Cross-sectional studies	4	0.0	0.60	1.11 (0.99–1.24)
Measurement of exposure and diagnosis
Database	12	97.6	<0.01	1.19 (1.09–1.30)
Self-report	11	43.7	0.06	1.21 (1.06–1.37)
Study country/region
US	8	55.7	0.03	1.16 (1.07–1.25)
Europe	9	95.9	<0.01	1.16 (1.02–1.33)
Asia	5	72.8	<0.01	1.30 (1.04–1.61)
Adjustment for maternal antibiotic exposure
Yes	4	74.4	<0.01	1.11 (1.01–1.23)
No	19	96.1	<0.01	1.22 (1.12–1.34)
Adjustment for gender, smoking, and parental allergy
Yes	6	22.5	0.27	1.18 (1.15–1.21)
No	17	91.3	<0.01	1.18 (1.07–1.30)
Adjustment for gender, birth weight, delivery mode
Yes	5	67.1	0.02	1.15 (1.11–1.20)
No	18	87.2	<0.01	1.22 (1.12–1.34)

Abbreviations: OR, odds ratio; CI, confidence interval; RC, reverse causation; CbI, confounding by indication.

Database studies often define a asthma diagnosis using a combination of International Classification of Diseases code and drug prescription records.

In the overall analysis and the RC, CbI-adjusted analysis, the subgroup of studies reporting RR lost the significance (Table 1, Table 2). However, significant association maintained across all subgroups stratified by exposure/outcome measurement and study country/region, with a trend of stronger association observed for studies using questionnaires and studies conducting in Asia (Table 1, Table 2).

Antibiotics: type, course, and exposure timing

Penicillin was the most frequently prescribed antibiotic type among included studies, with the exception of cephalosporin being most commonly used in Japan (Table S1). Macrolides were associated with a greater risk of developing asthma (OR, 1.56; 95% CI, 1.31–1.86) than penicillin (OR, 1.33; 95% CI, 1.15–1.54) and cephalosporin (OR, 1.39; 95% CI, 1.20–1.62) (Table 3). An increased risk of asthma was observed when antibiotic course increased from 1 to 2 (OR, 1.29; 95% CI, 1.18–1.42) to 3–4 (OR, 1.79; 95% CI, 1.49–2.14), and a plateau when increased to 5 courses (OR, 1.79; 95% CI, 1.36–2.36) (Table 3), revealing a non-linear dose-response relationship (P < 0.01) (Fig. 2). Subgrouping studies by exposure timing produced different effects (Table 3) (Fig. 3). The strongest effect was observed when exposure was in the first week of life (OR, 1.82; 95% CI, 1.34–2.47). A slightly stronger effect was observed when exposure was in the second (OR, 1.42; 95% CI, 1.27–1.59) and third year (OR, 1.54; 95% CI, 1.30–1.82). However, when RC and CbI were considered, reduced effect was observed in the second year (OR, 1.25; 95% CI, 1.07–1.47).

Table 3.

Detailed analyses regarding antibiotic type, course, exposure timing, age of asthma onset, asthma outcome, and time interval.

Subgroup types	Number of studies(n)	Heterogeneity		Pooled OR (95%CI)
		I2 (%)	P-value
Antibiotic type
Penicillin	10	97.4	<0.01	1.33 (1.15–1.54)
Cephalosporin	11	98.0	<0.01	1.39 (1.20–1.62)
Macrolides	11	97.8	<0.01	1.56 (1.31–1.86)
Antibiotic course
1-2	15	94.7	<0.01	1.29 (1.18–1.42)
3-4	14	97.8	<0.01	1.79 (1.49–2.14)
≥5	8	98.4	<0.01	1.79 (1.36–2.36)
Antibiotic exposure timing
1 week	3	0.0	0.61	1.82 (1.34–2.47)
6 months	10	98.9	<0.01	1.43 (1.25–1.63)
6 months-1 year	6	63.3	0.02	1.31 (1.14–1.50)
1 year	11	72.0	<0.01	1.41 (1.25–1.59)
2 years	10	59.8	<0.01	1.42 (1.27–1.59)
Adjusted for RC and CbI	5	64.2	0.03	1.25 (1.07–1.47)
Unadjusted for RC and CbI	5	0.0	0.87	1.65 (1.48–1.84)
3 years	5	56.9	0.05	1.54 (1.30–1.82)
Age of asthma onset
1–3 years	8	95.1	<0.01	1.84 (1.63–2.08)
Adjusted for RC and CbI	3	68.9	0.04	1.83 (1.48–2.28)
Unadjusted for RC and CbI	5	97.0	<0.01	1.83 (1.51–2.22)
4–6 years	11	60.3	<0.01	1.28 (1.18–1.39)
Adjusted for RC and CbI	4	0.0	0.60	1.17 (1.14–1.21)
Unadjusted for RC and CbI	7	51.8	0.05	1.37 (1.20–1.57)
>6 years	4	64.4	0.04	1.11 (0.96–1.28)
Time interval
0–2 years	7	93.0	<0.01	2.05 (1.91–2.20)
2–5 years	7	89.6	<0.01	1.38 (1.17–1.62)
5–6 years	4	0.0	0.99	1.17 (1.13–1.21)
Asthma definition
Asthma	41	96.7	<0.01	1.39 (1.30–1.49)
Transient asthmaa	3	97.2	<0.01	2.46 (1.50–4.04)
Persistent asthmab	3	96.7	<0.01	2.61 (1.49–4.59)
Late-onset asthmac	3	0.0	0.67	1.11 (1.06–1.15)
Wheezing	7	90.9	<0.01	1.48 (1.14–1.93)
Atopic asthma	4	0.0	0.70	2.14 (1.58–2.90)

Abbreviations: OR, odds ratio; CI, confidence interval; RC, reverse causation; CbI, confounding by indication.

^aTransient asthma: began and resolved before 3 years of age.

^bPersistent asthma: began before 3 years of age and persisted through to 4–7 years of age.

^cLate-onset asthma: began after 3 years of age

Fig. 2.

Linear and non-linear fit model demonstrating a dose-response relationship between antibiotic exposure course and childhood asthma.

Fig. 3.

Association of antibiotic exposure timing and risk of childhood asthma.

Asthma: age of onset, time interval and definition

Association of infant antibiotic exposure and risk of childhood asthma was strong at younger age (OR for 1–3 years, 1.84; 95% CI, 1.63–2.08) and became insignificant at >6 years old (OR, 1.11, 95% CI, 0.96–1.28) (Table 3) (Fig. 4). Reduced effect and heterogeneity was observed in the subgroups that adjusted for RC and CbI. Besides, the longer time interval between exposure and asthma onset, the lower risk of asthma was reported (OR for 0–2years, 2.05, 95% CI, 1.91–2.20 and OR for 5–6years, 1.17, 95% CI, 1.13–1.21). Pooling studies using definite asthma diagnosis showed similar pooled OR (1.39; 95% CI, 1.30–1.49) to that of the overall analysis. Significantly higher pooled OR appeared in studies reporting wheezing (OR, 1.48; 95% CI, 1.14–1.93), persistent asthma (OR, 2.61; 95% CI, 1.49–4.59), and atopic asthma (OR, 2.14; 95% CI, 1.58–2.90) (Table 3).

Fig. 4.

Age-stratified analysis of antibiotic exposure and risk of asthma.

Meta-regression analysis, sensitivity analysis, publication bias and GRADE evidence

Meta-regression analysis showed that study quality (P = 0.02), adjustment for medical consultation (P = 0.02), and adjustment for RC and CbI (P = 0.04) could be the sources of heterogeneity. Sensitivity analysis suggested robustness of our findings (Fig. S3). Begg's and Egger's test did not detect significant publication bias (P = 0.61 and P = 0.17, respectively). Funnel plot showed an asymmetry among small studies, suggesting that there may be some small study effect (Fig. S4). According to the GRADE criteria, the quality of evidence for studies that addressed RC and CbI was low due to study design and heterogeneity identified among studies (Table S2).

Discussion

In this meta-analysis, we showed that infant antibiotic exposure was associated with development of childhood asthma. Correction for RC, CbI, and other common confounders attenuated the association but did not change the significance. In the detailed analyses, macrolides, course of 3–4, exposure within 1 week of birth, asthma developed among 1–3 years, short time interval, persistent asthma, and atopic asthma phenotype had greater effect. The heterogeneity among studies and the low value of the GRADE assessment called for caution when consider the results.

Infant antibiotic exposure has been associated with childhood asthma development in recent decades. However, whether this association is biased by other influential factors is worth considering. Acknowledged factors among studies include sex, smoke exposure, familial allergy, birth weight, and delivery mode. Additionally, maternal antibiotic exposure during pregnancy may induce childhood asthma by interfering the neonatal microbiota, and thus confounds the effect of infant antibiotic exposure.^70,71 Medical consultation may be a reflection of the susceptibility to diseases under various risk factors⁷² and was considered as potential confounder.¹³ Pooling studies that adjusted for the above confounders showed that infant antibiotic exposure was still associated with childhood asthma. Besides, RC and CbI are two highlighted factors that strongly confounded the association between infant antibiotic exposure and childhood asthma. In the present analysis, studies that addressed RC and CbI showed an increased asthma risk of 18% when exposed to infant antibiotic with a narrower 95%CI, which indicates a more precise estimate. This result is in line with a meta-analysis including non-retrospective studies that addressed RC and CbI, which showed a small but significant association.¹² When we restricted studies to those that addressed RC, CbI and other fore-examined confounders, statistical significance remained. These findings indicate that the association between infant antibiotic and childhood asthma was relatively stable and only partly biased by these confounders.

Prospective studies with full consideration for RC and CbI are expected to provide casual inferences. In this meta-analysis, significant association remained in prospective studies that addressed RC and CbI, supporting a possible cause-and-effect relationship between antibiotics and asthma. However, this result is inconsistent with previous meta-analyses.^12,13 This discrepancy may be due to the short antibiotic exposure window of those studies, which can underestimate the true effect of infant antibiotic because exposure occurred in the second or third year of life also associated with increased risk of subsequent asthma.^23,37,47 Besides, over the past decade, increased access to medical counseling has led to more asthma detection.⁷³ Reduced exposure to farms or fields lead to reduced microbio-diversity in children.⁷⁴ In view of the above reasons, a causal effect of infant antibiotic on risk of childhood asthma in this study, or at present, seems to be rational. If there is a true causality, attention should be drawn to reduce unnecessary antibiotic prescription given that the risk of developing asthma increased by 24% when exposed to infant antibiotic and a fact that a huge number of children prescribed antibiotics.

The association between infant antibiotic exposure and childhood asthma can be elucidated by the microbiota hypothesis. Microbiota plays a key role in the immune modulation by the function of protecting from pathological bacteria, differentiating lymphocytes, and promoting Th1 phenotype development.⁷⁵ Antibiotic exposure can reduce the airway and gastrointestinal microbiome diversity,⁶ affect the balance between microbiota and immune system, and thus increase the risk of asthma development. In view of that, antibiotic type, course, and exposure timing may all modify the association through microbiota-mediated mechanism. Based on the previous studies, broad-spectrum antibiotics (e.g. penicillin, cephalosporin, macrolides) tend to show greater effect,^66,76 among which macrolides showed the greatest pooled effect in the present meta-analysis. This result can be supported by the finding that macrolides cause greater disorder to the microbiome.⁷⁷ Regarding antibiotic course, related studies reported that the risk of asthma increased with antibiotic course in a dose-response fashion.^47,51 However, a non-linear relationship was further revealed in the present meta-analysis.¹² This phenomenon has also been reported in a previous meta-analysis with nearly the same pooled estimate for >4 and 3–4 courses. However, explanation was not given in that study. We infer that this nonlinear relationship may be due to the “antibiotic resistance” of microbiome, meaning that microbiome may be more sensitive to preliminary antibiotic exposure.^78,79 This hypothesis needs to be elucidated in future studies. The first 6 months after birth are crucial for the development of a healthy microbiome.⁷ Antibiotics exposed in this period can profoundly affect the microbiome composition⁸⁰ and lead to increased risk of asthma, which is accordant with our results that exposure in the first week of life and in the first 6 months of life showed a relatively high risk. Unexpectedly, risk was higher when exposure at the second and third years than at the first year, which possibly arises from the cumulative effect of antibiotics. Besides, when studies limited their exposure window to the first year of life, effect of antibiotics exposed within the first year could be underestimated by classifying children who received antibiotics in their second and third year as unexposed children. Increased incidence of RC and CbI in later exposure may also account for it, as the effect of exposure at the second year reduced when adjusted for RC and CbI.

The role of infant antibiotic exposure in childhood asthma varied with age of asthma onset. We observed a stronger effect at age of 1–3 years, while an insignificant effect at age ≥6 years. Besides the high incidence of RC at early age, reason for the result may be that antibiotics may exert the greatest effect on microbiome shortly after the exposure.^56,81 This is supported by the time interval-stratified analysis that a higher risk of asthma onset is observed on a short time scale after exposure. These findings underscore the importance of protecting microbiota and preventing asthma following early-life antibiotic exposure and in the early stage after exposure. Besides, an accurate diagnosis of asthma for children younger than 3 years old is difficult⁵¹ and is often confused with wheezing. Wheezing is common in infants and mostly disappears with age,⁸² and wheezing from a RI is often treated with antibiotics,⁸³ both of which lead to an overestimated effect in studies using “wheezing” as outcomes. We also observed a high pooled estimate for studies analyzing atopic asthma. Atopic asthma and non-atopic asthma are 2 phenotypes of childhood asthma, which have different risk factors.⁸⁴ Non-atopic asthma seems to be more related to early infection and genetics.⁸⁵ Current studies mostly linked antibiotics with various childhood atopic diseases,^86,87 and thus antibiotics may have greater roles in atopic asthma. Similarly increased risk was observed in persistent asthma phenotype, although with a broad 95%CI due to small sample size. This significance became understandable when considering the long-standing and irreversible damage to the development of microbiota and subsequent immune system caused by antibiotic exposure in the childhood.^88,89 Future studies are expected to detail the association of infant antibiotic with asthma phenotype and how it varies with age, because this will help with risk factor identification and prevention of different asthma phenotype.

There is high heterogeneity among included studies. In addition to stratified analyses and meta regression on confounding factors, we also examined the assumption that HR, RR, OR are similar. Although the pooled estimates of studies using HR and OR were consistent, the subgroup of RR lost its significance. We think this may be due to the small number of included studies and the great heterogeneity in this subgroup. Therefore, caution is still required to interpret the findings. Besides, we also conducted stratified analyses by region. Significant association maintained among subgroups, with higher estimates reported in subgroup of Asia. This may contradict to our hypothesis, as the prevalence of asthma for Asian was reported lower than that for Americans or Europeans.^2,90 However, studies also showed that compared with White, South Asian exposed to infant antibiotic had higher risk of childhood asthma development, while not for East Asian.²⁶ Future studies examining this association among different regions and ethnicities are required.

Compared with previous meta-analyses, our findings are extensive and novel in that more influential factors were considered and combined with RC and CbI to analyze the effect of confounding. Besides, our study extended antibiotic exposure window to the second and third year. Accordingly, to avoid the confounding by RC, we conducted a series of stratified analyses. Furthermore, our detailed analyses of different settings of exposure and outcome can help to understand the association in depth, as well as provide clinical implications for asthma prevention. However, limitations include the heterogeneity among studies, pooling studies collectively using ORs, and again, caution is required when interpret the results. Besides, different adjusted models were adopted among the included studies and it was not possible to take into account all the confounders together in one subgroup analysis. Additionally, there are still several confounders, such as air pollution and socio-economic status that were not examined in the present analysis. Finally, underestimated effect of antibiotic cannot be ruled out since the actual antibiotic exposure may be overestimated in database studies that using prescriptions filled as the measurement of exposure.

Conclusions

This meta-analysis indicated that infant antibiotic exposure is associated with an increased risk of childhood asthma. Common confounders could only explain part of the association. This study further supports the reduction of unnecessary antibiotic prescription during infancy, and underlines the importance of asthma prevention in the early stage following early-life antibiotic exposure. Future studies into the effect of antibiotic on different asthma phenotypes are required, and caution is needed when interpret the results.

Abbreviations

OR, odds ratio; RR, relative risk; HR, hazard ratio; CI, confidence interval; RC, reverse causation; CbI, confounding by indication; BMI, body mass index; NICU, neonatal intensive care unit; NOS, Newcastle–Ottawa Quality Assessment Scale; AHRQ, Agency for Healthcare Research and Quality; RI, respiratory infection; GRADE, the Grades of Recommendation, Assessment, Development and Evaluation; PRISMA, the Preferred Reporting Items for Systematic Reviews and Meta-Analysis.

Funding/Support

This work was supported by the National Natural Science Foundation of China (Grant 81400072 and Grant 82172543), the Natural Science Foundation of Shandong Province (Grant ZR2020MH006), and the Key Research and Development Program of Shandong Province (Grant No. 2019GSF108198).

Ethical approval of studies

Not applicable.

Informed consent

Not applicable.

Authors’ consent for publication

All authors have reviewed and consented to publication.

Data availability statement

All information can be available in the paper.

Author contribution

Jingjing Wang searched the articles. Haixia Wang, Yizhang Li, Mo Yi and Yuanmin Jia assessed the article quality. Zeyi Zhang was a major contributor in writing the manuscript. Ou Chen is responsible for review and modification of the manuscript.

Declaration of competing interest

The authors have no conflicts of interest relevant to this article to disclose.

Appendix A. Supplementary data

The following is the Supplementary data to this article: