Early-Life Exposure to Antibiotics and Risk for Crohn’s Disease: A Nationwide Danish Birth Cohort Study

Abstract

Background

Early-life antibiotic use can alter the intestinal flora and modify the risk of developing Crohn disease (CD), but rigorous epidemiological evidence is limited, with inconsistent results.

Methods

We identified all children born in Denmark from 1995 to 2009 and followed them from birth until death, emigration, a diagnosis of CD, or January 1, 2013. Using Cox regression, we assessed the association between antibiotic exposure in the first year of life and subsequent risk for CD, adjusting for sex, degree of urbanization, birth order, birth year, route of delivery, gestational age, smoking during pregnancy, intake of nonsteroidal anti-inflammatory drugs in the first year of life, and family history of CD.

Results

During a median 9.5 years (9.3 million total person-years), CD was diagnosed in 208 of 979,039 children. Antibiotic use in the first year of life was associated with a higher risk of CD (adjusted hazard ratio, 1.4; 95% confidence interval [CI], 1.1-1.8), with the highest risk with ≥6 courses of antibiotics (adjusted hazard ratio, 4.1; 95% CI, 2.0-8.5). A family history of CD did not modify these risk associations. The cumulative risk of CD at the 11th birthday for children exposed to antibiotics in their first year of life was 0.16‰ (95% CI, 0.11‰–0.22‰) compared to 0.11‰ (95% CI, 0.08‰–0.15‰) for children unexposed to antibiotics in their first year of life.

Conclusions

Antibiotic use in the first year of life is associated with a modestly increased risk for CD, although the absolute risk is very low.

Introduction

Interactions between gut commensal bacteria and host immune function are increasingly recognized as playing crucial roles in the pathoetiology of many inflammatory diseases,¹ including Crohn’s disease (CD). Although specific genetic mutations in CD are associated with dysregulation of the gut immune response,² the 50% discordance rates for CD between monozygotic twins suggests that other factors are involved.³

The hygiene hypothesis proposes that early-life environmental exposures are key determinants in the development of many diseases.⁴ In support of this hypothesis, route of delivery (ie, vaginal or caesarean), urbanization, birth order, and childhood infections are associated with the risk of developing CD.⁴

The worldwide exponential increase in CD incidence in recent decades⁵ may in part be traced to changing perinatal influences. Parallel to this rise in CD incidence, studies have documented the indiscriminate use of antibiotics in pediatric populations.^{6, 7} Given the potential of antibiotics to disrupt the intestinal microbial flora, an association between early-life antibiotic exposure and subsequent risk for CD seems plausible. Studies consistently show an increased risk for CD among those who receive antibiotics in childhood,^8-14 but limitations include small sample sizes, short follow-up, and use of nonvalidated inflammatory bowel disease (IBD) diagnoses. A recent population-based study from Manitoba, Canada, found a positive association between childhood gastrointestinal infections (viral and bacterial) and the risk of IBD, but a direct effect of antibiotics on the risk of IBD could not be established, possibly in part because data on antibiotic exposure were only available for a small number of children.¹⁵ Rigorous epidemiological evidence is limited, and no studies have explored the possible role of genetic predisposition in the association between antibiotic use and IBD risk.

The aim of this study was therefore to examine an association between early-life exposure to antibiotics and the subsequent risk of developing CD in a large, national birth cohort with long-term follow-up using validated outcome definitions.

METHODS

Setting

This study was conducted within the entire Danish population, consisting of approximately 5 million people. In Denmark, all pregnancies are prospectively registered in a national registry, and perinatal and pediatric health care is either fully or partly reimbursed by the national health authorities.¹⁶

Ethics Statement

The study was approved by the Danish Data Protection Agency.

Birth Cohort

We constructed a cohort of all children born in Denmark in the period from 1995 to 2009 who were alive and living in Denmark at their first birthday. These children were identified from the Danish Medical Birth Registry, a prospectively maintained health database established in 1973. It contains data on prenatal factors (eg, maternal age, parity, smoking status), route of delivery (vaginal or caesarean), and newborn characteristics (eg, date of birth, weight, length, gestational age, and Apgar score) for all infants born in Denmark.¹⁷ At birth, all infants are assigned a civil registration number: a personal, unique identifier that allows accurate linkage of individual-level data among all national health registries. The Civil Registration System issues this number and maintains data regarding name, address, marital status including spouse, place of birth, kinship, profession, death, and emigration.¹⁸

Exposure

The exposure of interest was antibiotic use in the first year of life, defined as either an antibiotic outpatient prescription or a hospital contact (i.e. admission or outpatient consultation) likely to reflect antibiotic use. Data on antibiotic prescriptions were collected from the Register of Medical Product Statistics, where information on antibiotic type was based on the Anatomical Therapeutic Chemical classification system (see Supplementary Table 1 for codes). The Register of Medical Product Statistics contains complete data on outpatient drug prescriptions made in Denmark since 1995. Data include date of the prescription, Anatomical Therapeutic Chemical code, and the number of daily doses in the prescription.¹⁹

From the Danish National Patient Registry (DNPR),¹⁶ we collected data on hospital contacts with serious infections. The DNPR is a nationwide register that contains records on 99.4% of all hospital discharges that occurred after January 1, 1977. Since 1995, the register has also captured outpatient contacts, covering virtually all specialist medical care in Denmark. For both inpatient and outpatient contacts, recorded data include the civil registration number, admission and discharge dates, treating hospital and department, invasive procedures performed during the contact, and up to 20 diagnoses, including 1 mandatory discharge diagnosis. From the inception date of the current cohort on January 1, 1995, all diagnoses were coded according to the International Classification of Diseases, 10th edition, and we defined serious infections as those for which antibiotics were highly likely to have been used (see Supplementary Table 1 for codes).

Outcome

Because the accuracy of a single CD diagnosis in the DNPR is unknown, we performed a validation study on probable CD diagnoses among children from the central region of Denmark, defined as those with at least 1 CD diagnosis (International Classification of Diseases-10 code K.50) reported to the DNPR after the individual’s first birthday. We identified 64 children with potential CD, and a pediatric gastroenterologist (AL) reviewed all patients using the Porto criteria to confirm the diagnoses.²⁰ Of the 64 potential patients with CD, 27 were confirmed. We then calculated the sensitivity and positive predictive value (PPV) of different diagnostic algorithms using combinations of data from the DNPR, the National Pathology Registry, and the Register of Medical Product Statistics. The National Pathology Registry contains data on all pathological tests performed in Denmark since the registry’s inception in 1997, and diagnoses are recorded using a uniform classification system (ie, SNOMED).²¹

The sensitivity of a single CD diagnosis reported to the DNPR was 100%, but the PPV was only 42%. Defining CD as at least 3 CD diagnoses recorded in the DNPR resulted in a more accurate definition (sensitivity 89%, PPV 92%). Consequently, we used this definition, and the date of the third CD diagnosis after the first year of life defined the date of CD onset.

Possible Confounding Factors

Although no strong environmental risk factors have been identified for CD, we chose the following perinatal, demographic, and socioeconomic factors as potential confounders a priori, based on previous research⁴ and biological plausibility: route of delivery (caesarean or vaginal), maternal age, gestational age (preterm: <37 weeks; near-term: 37-38 weeks; term: ≥38 weeks), maternal smoking during pregnancy, sex assigned to the infant, degree of urbanization of birthplace (with “urban” defined as living in 1 of the 5 largest cities in Denmark: Copenhagen, Aarhus, Odense, Aalborg, and Esbjerg), and birth order. Information on these variables was collected from the Danish Medical Birth Registry.

Although nonsteroidal anti-inflammatory drugs (NSAIDs) are infrequently prescribed to infants, they may be associated with an increased risk for IBD.²² For this reason, we also included administration of NSAIDs (see Supplementary Table 1 for code) within the first year of life.

Comparison Cohort Members

The most important risk factor for CD is a family history of the condition.²³ For this reason, we included family history as a positive control so that a strong association between family history of CD and risk of CD in the child would represent strong support for a valid study design and outcome definition. For this purpose, we identified first-degree relatives of the infant from the Danish Medical Birth Registry and cross-linked individual-level data with the DNPR to identify those who had a diagnosis of CD.

Beta-agonists are widely used in outpatient pediatric care, and their use has not been associated with the development of either ulcerative colitis (UC) or CD. We therefore included the occurrence of any beta-agonist administration (see Supplementary Table 1 for code) during the first year of life as a negative control, assuming that the absence of an association between beta-agonist therapy and subsequent CD would indicate a reliable methodology and analytic strategy.

Statistical Analysis

Children in the cohort were followed from their first birthday to a diagnosis of CD, death, emigration, or end of the observation period on January 1, 2013. In a primary analysis, we used Cox proportional hazards modeling to estimate the hazard ratio (HR) of antibiotic exposure or serious infections in the first year of life for subsequent CD, adjusting for route of delivery, maternal age, gestational age, maternal smoking during pregnancy, sex of the infant, degree of urbanization, birth order, and NSAID use.

In a secondary analysis, we included antibiotic exposure or serious infections in the first year of life as an ordinal variable on Cox regression (1-2 courses of antibiotics/serious infections, 3-5 courses of antibiotics/serious infections, ≥6 courses of antibiotics/serious infections). In this analysis, administration of antibiotics occurring within 7 days before or after a diagnosis of infection was not considered a separate outcome.

Early antibiotic use may be associated with antibiotic use later in life as a result of parent or physician preferences or a tendency to develop infections. For this reason, we included serious infections and antibiotic administration (as defined above) occurring beyond the first year of life as a time-dependent covariate in a sensitivity analysis.

We also performed analyses separately for the different antibiotic subgroups. For this purpose, exposure was categorized according to the type of antibiotic prescribed most recently. Furthermore, we performed the analyses separately for those with and without a family history of CD, to assess a potential effect modification of a genetic predisposition on the association between antibiotic exposure and risk of CD.

To ascertain the number of newly diagnosed CD that could be attributed to antibiotic use, we calculated the population-attributable fraction (PAF)²⁴ based on the cumulative risk for CD at 10 years and used the following formula:

where Ppop was the proportion of children exposed to antibiotics and RR was the adjusted relative risk for CD associated with antibiotic use.

Cumulative risk for CD after the first year of life was calculated using death as a competing risk and illustrated as cumulative functions, stratified by exposure to antibiotics within the first year of life.

RESULTS

Study Population

In the period 1995 to 2009, a total of 979,039 children were born and survived to their first birthday in Denmark, of whom 476,694 (49%) were females and 3487 had a family history of CD. The distributions of perinatal and demographic variables are listed in Table 1, according to exposure to antibiotics or serious infections within the first year of life.

Within the first year after birth, a total of 400,162 (41%) children received at least 1 antibiotic prescription; 79,124 (20%) of these children received 3 to 5 prescriptions, and 9372 (2%) received ≥6. From 1995 to 2009, the proportion of children receiving antibiotics in their first year of life remained relatively stable. Similarly, we found no differences when looking at the antibiotic subgroups separately (data not shown).

A total of 40,169 (4%) children had experienced at least 1 serious infection within the first year of life. There were 1 to 2 serious infections among 37,822 (94%) children, 3 to 5 serious infections among 2209 (5%) children, and ≥6 serious infections in 138 (1%) children, with no differences between calendar years. There were 12,401 children in the exposed cohort who were included exclusively for diagnoses of infection with no antibiotic prescription, whereas 372,394 children were included in the exposed cohort based solely on antibiotic prescriptions without any diagnoses of infection and 27,768 children who had both been diagnosed with infections and been prescribed antibiotics in their first year of life. Beta-agonist medication had been prescribed to 74,386 (7.6%) children within their first year of life.

Risk of Antibiotic Use for Development of CD

During 9,262,962 person-years of observation (median, 9.5 years), 208 children were diagnosed with CD, most (61.1%) between ages 10 and 15 years. Among the children who were prescribed antibiotics or diagnosed with a serious infection in their first year of life, 102 (1 in 4044) were diagnosed with CD, compared with 106 (1 in 5343) of those who were not. On crude analysis, the HR for CD was 1.4 (95% confidence interval [CI], 1.1-1.8) comparing children receiving antibiotics or experiencing a serious infection within their first year of life with children who did not. Compared to unexposed children, the risk was highest for those who had ≥6 antibiotic prescriptions or serious infections in that first year (HR, 4.1; 95% CI, 2.0-8.5) and for those who had 3 to 5 (HR, 1.1; 95% CI, 0.6-1.8) or 1 to 2 antibiotic prescriptions or serious infections (HR, 1.4; 95% CI, 1.0-1.8; data not shown). The HR for CD associated with beta-agonist use in the first year of life was 1.3 (95% CI, 0.8-2.3).

On adjusted analyses, the associations between the covariates of interest and the risk of CD remained largely unchanged (Table 2). The covariate most strongly associated with the risk of CD was a family history of CD (adjusted hazard ratio [aHR], 14.0; 95% CI, 6.9-28.4).

On sensitivity analysis adjusting for serious infections or antibiotic prescriptions beyond the first year of life as a time-dependent variable, the aHR for CD among children receiving antibiotics or experiencing serious infections in the first year of life was virtually identical to the primary analysis (1.3; 95% CI, 0.9-1.8).

When analyzing different subgroups of antibiotics separately, prescriptions of beta-lactam antibiotics (aHR, 1.4; 95% CI, 1.1-1.9) or macrolides, lincosamides, and streptogramins combined (aHR, 2.0; 95% CI, 1.1-3.5) within the first year of life were associated with an increased risk for CD. For the other subgroups of antibiotics, the number of exposed children was too small to support analyses.

The association between an antibiotic prescription or serious infection in the first year of life and the risk of being diagnosed with CD was similar for those with a family history of CD (aHR, 1.4; 95% CI, 0.4-5.9) compared with those without a family history of CD (aHR, 1.4; 95% CI, 1.0-1.8; Supplementary Table 2). Including both diagnoses of CD and UC in first-degree relatives as a “family history of IBD” did not markedly change the associations.

When we restricted the analysis to children with at least 10 years of follow-up after their first birthday, the association between antibiotic exposure or serious infections the first year of life was virtually unchanged (aHR, 1.3; 95% CI, 1.0-1.9).

PAF

The PAF associated with antibiotic use within the first year of life was 17% at 10 years.

Cumulative Risk of CD

Cumulative risks of CD after the first year of life among children exposed and unexposed to antibiotics, respectively, are illustrated in Fig. 1. The risk for CD among exposed children was 0.16‰ (0.11‰–0.22‰) at their 11th birthday, compared to 0.11‰ (0.08‰–0.15‰) for unexposed children.

Discussion

Environmental factors are increasingly recognized as important determinants in the etiology of CD.³ Here we found that antibiotic use or serious infection occurring within the first year of life was associated with a modestly increased CD risk, although the absolute risk for CD was very low. The association was strongest for penicillins, macrolides, lincosamides, and streptogramins. Children receiving ≥6 antibiotic prescriptions or experiencing ≥6 serious infections within their first year of life had the highest risk. A family history of CD did not modify the effect of antibiotic use or serious infections on the risk of subsequent CD.

The association between antibiotic use or serious infections and CD that we identified was smaller than that previously reported in a Danish nationwide cohort study covering the period 1995 to 2003. In that study, the risk ratio of being diagnosed with CD was 3.41 for children prescribed antibiotics compared to those who were not.⁸ However, this estimate was based on only 50 children being diagnosed with CD, and the cases of CD were not validated, which may have affected the precision of the estimate. Other studies that have explored the association between antibiotic exposure and risk of being diagnosed with CD have also relied on nonvalidated CD case definitions.^{9, 11-13} In addition, positive associations between medications used as negative controls (eg, drugs for cardiovascular or neurologic disease) and CD⁹ also suggest the possibility of misclassification of cases of CD and/or other potential biases. Most studies on antibiotic exposure and risk for CD are either case-control studies^{9, 11, 12} that could have been biased by reverse causality or are based on selected patients from referral centers,¹¹ which can diminish the transferability of results.

A large Swedish birth cohort study found an association of in utero exposure to antibiotics with a 2.5-fold higher risk of developing very-early-onset CD compared to those who had not been exposed.¹³ For antibiotic exposure in infancy, the risk of very-early-onset CD was only 11% higher compared to unexposed children. That study also was limited by a nonvalidated outcome definition, and only 20 children were diagnosed with CD. Note that the cumulative risk for CD that we found, regardless of exposure to antibiotics, was very low. Our analysis suggests that approximately 17% of diagnoses of CD in our cohort could have been prevented with complete elimination of antibiotics exposure. This information adds an important public health aspect to our study in the sense that antibiotic use possibly contributes only moderately to CD risk. Conversely, no other modifiable factor has been identified with a higher PAF. The decision to prescribe antibiotics should, among many other things, be weighed against a low probability of increasing the individual child’s risk of CD.

Mutations in genes that are involved in the intracellular killing of bacteria and the clearance of antigens are common in CD,² suggesting that the effects of disrupting gut commensal bacteria composition may be augmented in genetically susceptible individuals. Our finding of similar associations between antibiotic use and risk of CD among children with and without a family history of CD argues against a significant interaction between genetic factors and antibiotic use.

Although evaluating the association between exposure to antibiotics and the risk of UC was beyond the scope of our current study, we note that a recent population-based study from Sweden found prior exposure to antibiotics to be associated with an adjusted odds ratio of 1.74 for being diagnosed with UC in adulthood.¹⁴ In contrast, a meta-analysis based on 8 case-control studies and 3 cohort studies reported comparable risks of being diagnosed with UC for individuals exposed to antibiotics compared to individuals with no prior exposure to antibiotics.²⁵ Although the association between exposure to antibiotics in early childhood and development of UC is less certain, a recent study suggests that the risk is not increased.¹³ Further population-based studies using validated exposures and outcomes will be needed to further evaluate early life risk factors for UC.

The strengths of this study lie in the inclusion of a large nationwide sample, a long follow-up period, the use of a validated definition of CD, and a high number of patients with CD. The negative association between beta-agonist use and the risk of CD suggests that our analyses were unlikely to have been biased by a higher likelihood of having quiescent CD diagnosed in children in regular contact with health care services. Similarly, and consistent with previous studies,²³ the observation that a family history of CD was the covariate most strongly associated with CD also indicates robust analyses.

Our study also had some important limitations. Reverse causation, with symptoms of CD mistakenly treated as a gastrointestinal infection with antibiotic therapy, may in part explain our findings. This potential confounding is, however, not likely to be significant because the association was most pronounced for penicillins and macrolides, lincosamides, and streptogramins, which are usually prescribed for upper respiratory tract infections or related symptoms. However, the subgroup analysis of the association between antibiotic type and risk of CD should be considered exploratory and hypothesis-generating, and further studies are needed to verify this finding. Given that most children were diagnosed between ages 10 and 15 years, it is also unlikely that CD would go undiagnosed for so many years, further arguing against reverse causation. It should also be kept in mind that the findings from our study are not necessarily transferable to an adult population, and it is possible that the effects of antibiotics on the adult gut microbiome are different from the effects on the infant gut microbiome.

It was not possible to determine whether outpatient antibiotic prescriptions actually reflected antibiotic use among the children in this study. This potential misclassification could have led to an underestimation of the association between antibiotic exposure and risk of CD.

Reporting of fatherhood to the Danish Medical Birth Registry is based on nonvalidated information from the mother, so some children may have had a biological father other than the person recorded in the registry. The association between family history of CD and the risk of subsequent CD in our study should consequently be interpreted with this potential misclassification in mind.

Conclusions

Exposure to antibiotics in the first year of life was associated with a modestly increased risk of future CD, although the absolute risk for CD was very low in this study.

Abbreviations

Abbreviations
 
• aHR

  adjusted hazard ratio

 
• CD

  Crohn disease

 
• CI

  confidence interval

 
• DNPR

  Danish National Patient Registry

 
• HR

  hazard ratio

 
• IBD

  inflammatory bowel disease

 
• NSAID

  nonsteroidal anti-inflammatory drug

 
• PAF

  population-attributable fraction

 
• PPV

  positive predictive value

 
• UC

  ulcerative colitis

Author contributions: Aksel Lange, Rune Erichsen, Henrik Toft Sørensen, and Michael Kappelman contributed to the concept and design of the study and the acquisition and interpretation of data. Trine Frøslev and Buket Öztürk contributed to the design of the study and analyzed and interpreted the data. Anders Mark-Christensen contributed to the concept of the study, interpreted the data, and drafted the manuscript. He is also the guarantor of the article. All authors critically revised the manuscript for important intellectual content, approved the final manuscript as submitted, and agree to be accountable for all aspects of the work.

Supported by: National Institutes of Health (5K08DK088957).

Acknowledgments

We thank San Francisco Edit for their work proofreading and editing the manuscript.

References