Introduction

With advances in medicine, antibiotics are commonly prescribed around the world [1]. The special physiology of pregnant women makes them more susceptible to infection, such as urinary tract infection. Thus, antibiotics are often used during the pregnancy period. It is estimated that up to 40% of pregnant women receive antibiotics prior to delivery [2, 3]. The prevalence of atopic diseases, such as food allergy, atopic dermatitis, asthma, and allergic rhinitis, has also increased globally in recent decades as a result of industrialization [4,5,6]. These allergic diseases not only seriously affect the quality of patients’ lives, but also cause a huge personal and socioeconomic burden [7].

A possible link has been suggested between the increasing use of antibiotics during pregnancy and the occurrence of atopic illnesses. The composition of an infant’s gut microbiome contributes to her subsequent immunological development. Alteration of the microbiome could lead to subsequent allergy diseases and obesity later in life [8,9,10]. The maternal microbiome determines the initial composition of the infant’s microbiome. Some studies reported that maternal antibiotic exposure during pregnancy could change infants’ microbiome [11, 12]. A matched case–control study found prenatal antibiotic exposure was associated with an increased risk of asthma [13]. However, large-scale studies on prenatal antibiotic exposure and atopic diseases later in life are still lacking.

The aim of this study was to explore the association of antibiotic exposure during pregnancy with childhood atopic diseases from a nationwide, population-based perspective.

Materials and methods

Study design and data source

This was a nationwide, population-based cohort study. Taiwan’s National Health Insurance Research Database (NHIRD) was the main source of data. Taiwan’s National Health Insurance (NHI) system was launched in 1995. It is a single-payer program with mandatory enrollment. The current coverage rate is 99.99% of Taiwan’s population (approximately 23.5 million). In 2002 , the NHIRD was established for research purposes. It contains all claims data from the NHI [14,15,16]. Since 2015, the Health and Welfare Data Center (HWDC) of Taiwan’s Ministry of Health and Welfare (MOHW) further integrated NHIRD with other health-related databases [17]. In this study, the pairing of mothers and children was achieved by linking the NHIRD with the Taiwan Maternal and Child Health Database (MCHD) of Taiwan’s Health Promotion Administration (HPA). The main data analyzed in this study were obtained from ambulatory care expenditures by visit (CD) files and inpatient expenditure by admission (DD) files from the NHIRD. Antibiotic exposure records were acquired from inpatient order (DO) files. For privacy protection and database reliability, Taiwan’s Ministry of Health and Welfare (MOHW) requires investigators to conduct on-site analysis. During the study period, diagnoses in the NHIRD were coded by the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) format.

Study population

This nationwide cohort study only enrolled first-time pregnancies during the study period, from 2004 to 2010. We excluded infants of multiple delivery, preterm delivery, and death before 5 years old. Finally, a total of 906,942 infants were enrolled in the study cohort (Fig. 1). The cohort was followed up until the end of 2016. All children in this cohort were followed up for at least 5 years.

Fig. 1
figure 1

Composition of the study cohort. Only the first child in each family was enrolled. Premature infants and children of early death were excluded from analysis

Full size image

Exposure to antenatal antibiotics

Antenatal antibiotic exposure was defined as a mother who received a medication with an ATC code (J01A, J01B, J01C, J01D, J01E, J01F, J01G, J01M, J01R, and J01X) during pregnancy. The timing of prescription (first, second, and third trimesters) and cumulative numbers of prescriptions were also recorded. We restricted the subgroup analysis to the timing of the initial exposure.

Outcome measurement

During the study period, the International Statistical Classification of Diseases and Related Health Problems, Ninth Revision, Clinical Modification (ICD9-CM) was used for coding of each diagnosis. Children who visited the outpatient department more than twice or were admitted once with a primary diagnosis of asthma (ICD-9 code 493.9), allergic rhinitis (ICD-9 code 477.9), or atopic dermatitis (ICD-9 code 691.8), were regarded as having an atopic disease.

Covariates

Maternal age, mode of delivery, maternal comorbidities, maternal allergic diseases, pregnancy-related complications, and infants’ gender were collected as potential confounders.

Statistical analysis

The data were retrieved and analyzed using the SAS statistical package (version 9.4; SAS Institute, Cary, North Carolina, USA). Demographic data were described by the mean with standard deviation, or frequency and percentage. Continuous variables were compared using the independent t-test. The Pearson’s Chi-square test was applied for analyzing categorical data. Cumulative incidences of atopic diseases between groups were compared by the Kaplan–Meier method. Cox regression model was applied for calculating the hazard ratios of antibiotic prescription after adjusting for potential confounders. A p value less than 0.05 was considered statistically significant.

Results

The cumulative incidences of atopic diseases

Of the 900,584 enrolled children, 359,891 (40.0%) were exposed to prenatal antibiotics. A comparison of the demographic data of these two groups revealed that the antibiotic exposure group had a slightly younger age of pregnancy, more Cesarean sections, more maternal comorbidities, more maternal allergic diseases, more pregnancy complications, and more male babies (Table 1). At the end of the study, the cumulative incidences of atopic diseases of the antibiotic exposure group were: 29.5% for atopic dermatitis, 30.5% for asthma, and 56.4% for allergic rhinitis. In the non-antibiotics group, the cumulative incidences were: 26.4% for atopic dermatitis, 28.4% for asthma, and 52.8% for allergic rhinitis (Fig. 2). The adjusted hazard ratio of antibiotics exposure during pregnancy to childhood atopic diseases were 1.12 for atopic dermatitis, 1.06 for asthma, and 1.08 for allergic rhinitis. All of them reached statistical significance (Table 2). Univariate analysis and actual numbers in each category are listed in Table 3.

Table 1 Characteristics of study subjects
Full size table
Fig. 2
figure 2

Cumulative incidences of atopic diseases with or without prenatal antibiotics: A atopic dermatitis; B asthma; C allergic rhinitis. Prenatal antibiotics exposure increases the cumulative risk in all three atopic diseases. CI confidence interval, HR hazard ratio

Full size image
Table 2 Adjusted hazard ratios of prenatal antibiotics for childhood atopic diseases by Cox regression models*
Full size table
Table 3 Univariate analysis of factors associated with childhood atopic diseases
Full size table

Timing of prenatal antibiotic exposure and childhood atopic diseases

To investigate how the timing of antibiotic prescription affected the incidences of childhood atopic diseases, we further stratified the infants into three groups according to their first-time exposure to antibiotics during the pregnancy course. After adjusting for confounders, including maternal age, mode of delivery, preterm delivery, maternal comorbidity, maternal allergic disease, pregnancy-related complications, and neonatal gender, the hazard ratios for asthma, allergic rhinitis, and atopic dermatitis were 1.07, 1.09, and 1.13 for the first trimester, 1.06, 1.06, and 1.07 for the second trimester, and 1.02, 1.04, and 1.06 for the third trimester. Although all these hazard ratios reached statistical significance, the timing of exposure did not affect the magnitude of risk for childhood atopic diseases (Table 4).

Table 4 Adjusted hazard ratios of prenatal antibiotics for childhood atopic diseases by Cox regression models*
Full size table

Cumulative number of times of prenatal antibiotic exposure and childhood atopic diseases

We further stratified the children according to their cumulative number of times of prenatal antibiotics exposure to test if a dose-dependent effect existed. After adjusting for confounders, including maternal age, mode of delivery, preterm delivery, maternal comorbidity, maternal allergic disease, pregnancy-related complication, and neonatal gender, the hazard ratios for asthma, allergic rhinitis, and atopic dermatitis were 1.04, 1.06, 1.08 for one exposure, 1.06, 1.09, 1.13 for two exposures, and 1.11, 1.12, 1.20 for exposure more than 3 times. A trend was revealed showing the higher the number of times an infant was prenatally exposed to antibiotics, the higher the hazard ratio was for childhood atopic diseases (Table 4).

Types of delivery and risk for childhood atopic diseases

We stratified the children according to their types of delivery. After adjusting for potential confounders, including maternal age, mode of delivery, preterm delivery, maternal comorbidity, maternal allergic disease, pregnancy-related complication, and neonatal gender, the hazard ratios for asthma, allergic rhinitis, and atopic dermatitis were 1.07, 1.08, 1.12 for vaginal delivery and 1.06, 1.08, 1.12 for Cesarean section (Table 5). The risk raised by antibiotics exposure was not modified by types of delivery.

Table 5 Adjusted hazard ratios of prenatal antibiotics for childhood atopic diseases, stratified by types of delivery*
Full size table

Discussion

This nationwide, population-based cohort study reveals that prenatal antibiotic exposure increases the risk of childhood atopic disease. A dose-dependent effect was revealed by the positive correlation between the cumulative number of times antibiotics were prescribed and the risk of atopic diseases. The increased risk of atopy associated with antibiotic exposure was not affected by different trimesters. This study provides comprehensive evidence that the pathogenesis of childhood allergic diseases may begin in early pregnancy, according to population-based data.

Antibiotic exposure in mid-to-late pregnancy was consistently associated with childhood asthma in a Danish birth cohort study [18]. Trimester effects have also been reported in several smaller-scale studies [19,20,21]. However, a meta-analysis revealed a positive association between prenatal antibiotic use in every trimester and the occurrence of childhood asthma [22]. In our study, trimester effects were assessed, but no significant differences in hazard ratios were found among the trimesters. This apparent inconsistency in findings might be explained by the different follow-up periods and different disease definitions used in the studies. Prenatal antibiotic exposure has also been reported to increase the risk of atopic dermatitis and hay fever [23, 24]. However, the trimester effects were not analyzed. Our study also reported that risk of allergic rhinitis was positively associated with prenatal antibiotic exposure. Respiratory tract infections were the most common indication for prenatal antibiotics use in our study (Appendix Table 6). Subgroup analysis by different kinds of antibiotics is added in Appendix Table 7. There are no significant differences between groups. Only quinolone shown borderline statistical significance in asthma. Vaginal delivery exposes the newborn to the maternal gut microbiota directly during birth, which may have a protect effect than in cases of cesarean section.

More studies are needed to elucidate the mechanism underlying the positive association between use of prenatal antibiotics and childhood atopic diseases. The hygiene hypothesis may partially explain it [25]. According to the hygiene hypothesis the microbiota, i.e., the composition of the intestinal flora, which is established early in life, plays a crucial role in the development of the immune system in children [26]. The association between gut microbiota and allergic diseases has been reported in a number of studies [27, 28]. The microbial colonization of the fetus has been reported to occur as early as 11 weeks of gestation [29]. Thus, by inducing reductions and alterations in the fetal intestinal microbiota, exposure to antibiotics during pregnancy may affect immune system development, thereby increasing the likelihood of chronic disease [30, 31]. Animal studies have also shown that antibiotics could induce the transition from TH1/TH2 balance to TH2-dominant immunity. Nevertheless, oral administration of intestinal flora could prevent this process from developing [32, 33]. The risk of childhood asthma increases as the cumulative number of courses of prenatal antibiotics increases, according to a Canadian cohort study [34]. A dose-dependent effect has also been reported in a claims data analysis [31]. In our study, a similar trend was noted in all childhood allergic diseases. This further supports the notion that prenatal antibiotics may be causally linked with childhood atopic diseases, and that this relationship is not the result of the phenomenon of confounding by indication [35, 36].

The correlation between antibiotic exposure during pregnancy and childhood allergic diseases may be confounded by many factors. Maternal characteristics such as maternal age, maternal history of allergy, maternal smoking, delivery mode, and maternal education level have all been reported [23, 34, 35, 37, 38]. The strongest confounder may be maternal allergic disease, because atopy has a strong hereditary tendency. The strongest predictor of childhood atopic diseases is genetic inheritance from parents. If we include all siblings in this study. The analysis might be confounded by family clusters [39, 40] So, we included only the first child in each family. Preterm infants usually have more medical care need. So, we excluded them to prevent the confounding effect. If children did not survive more than 5 years, short follow-up time would confound the outcome analysis. As a result, we did not involve those infants of early death.

Our study had certain limitations. The data source was national health insurance claims data, which do not include laboratory data. The disease diagnosis was mainly decided by physicians’ coding. The validity of the diagnoses could not be confirmed because personal identification data are not permitted to be released from the data center. Thus, certain misclassifications may have existed. Because we used Cox regression model to analyze the cumulative hazard ratio between groups. However, Cox regression model (proportional hazard model) can only calculate the hazard ratio. Risk difference calculation can count the attributable risk proportion. It may be more valuable in public health policy making.