Abstract
BACKGROUND AND OBJECTIVE: Recent evidence has linked childhood antibiotic use and microbiome disturbance to autoimmune conditions. This study tested the hypothesis that antibiotic exposure was associated with newly diagnosed juvenile idiopathic arthritis (JIA).
METHODS: We performed a nested case–control study in a population-representative medical records database from the United Kingdom. Children with newly diagnosed JIA were compared with age- and gender-matched control subjects randomly selected from general practices containing at least 1 case, excluding those with inflammatory bowel disease, immunodeficiency, or other systemic rheumatic diseases. Conditional logistic regression was used to examine the association between antibacterial antibiotics (including number of antibiotic courses and timing) and JIA after adjusting for significant confounders.
RESULTS: Any antibiotic exposure was associated with an increased rate of developing JIA (adjusted odds ratio: 2.1 [95% confidence interval: 1.2–3.5]). This relationship was dose dependent (adjusted odds ratio over 5 antibiotic courses: 3.0 [95% confidence interval: 1.6–5.6]), strongest for exposures within 1 year of diagnosis, and did not substantively change when adjusting for number or type of infections. In contrast, nonbacterial antimicrobial agents (eg, antifungal, antiviral) were not associated with JIA. In addition, antibiotic-treated upper respiratory tract infections were more strongly associated with JIA than untreated upper respiratory tract infections.
CONCLUSIONS: Antibiotics were associated with newly diagnosed JIA in a dose- and time-dependent fashion in a large pediatric population. Antibiotic exposure may play a role in JIA pathogenesis, perhaps mediated through alterations in the microbiome.
- CI —
- confidence interval
- IBD —
- inflammatory bowel disease
- JIA —
- juvenile idiopathic arthritis
- OR —
- odds ratio
- THIN —
- The Health Improvement Network
- URI —
- upper respiratory tract infection
What’s Known on This Subject:
The etiology of juvenile idiopathic arthritis (JIA) is poorly understood. A recent study suggested a link between antibiotics and JIA but did not examine the potential for confounding from infections or the role of antibiotic timing.
What This Study Adds:
Antibiotics were associated with newly diagnosed JIA in a dose- and time-dependent manner after adjusting for infection and other confounders. Antibiotics may play a role in the pathogenesis of JIA.
Juvenile idiopathic arthritis (JIA) is the most common rheumatic diagnosis in children, but its etiology remains unclear.1 Despite the presence of known genetic risk factors, population-based studies suggest that genetics explains only 10% to 25% of disease incidence.2,3 Environmental triggers such as viral infections have been suggested,4–6 but other studies have not substantiated these findings.7–10 One study found that hospitalization for infection in the first year of life was associated with an increased risk of developing JIA.11 However, this study focused on hospital-based diagnoses and not on antibiotic use, thus limiting its interpretability.
Disturbance of the human microbiome has been implicated in the development of chronic pediatric diseases, including inflammatory bowel disease (IBD).12,13 At least 1 category of JIA has been associated with distinct intestinal microbial populations.14 Antibiotics can alter some subjects’ intestinal microbiota for ≥6 months.15–17 Correspondingly, anaerobic antibiotic use has been associated with incident pediatric IBD.18 Another recent study suggested that childhood antibiotic exposure was associated with a subsequent diagnosis of JIA.19 Of note, this study controlled for age, gender, and geography but did not differentiate the effects of antibiotic use from the effects of infection. Furthermore, the impact of antibiotic timing was not examined in detail. The present study tested the hypothesis that antibiotics were associated with incident JIA in a large pediatric population in a dose- and time-dependent fashion after adjusting for confounding from infection and other variables.
Methods
Study Design and Data Source
A nested case–control study was performed by using The Health Improvement Network (THIN), a population-representative electronic medical records database from >550 practices of general practitioners across the United Kingdom.20 This design was chosen for its computational efficiency and unbiased estimates of incidence rate ratios, including when studying common time-varying exposures and rare outcomes.21,22 Data within THIN are collected during routine medical care by using Vision software (INPS, London, United Kingdom) and are anonymized for research purposes. These data contain information on demographic characteristics, diagnoses, specialist and hospital referrals, and outpatient prescriptions. The present study included data from 1994 through 2013. THIN has been validated for several diseases in pharmacoepidemiologic research,23 including JIA.24
Subject Selection
Eligible subjects were aged 1 to 15 years, born after initiation of the Vision software, and registered within 3 months of birth to capture lifetime outpatient prescriptions. Cases were defined by using diagnostic Read codes (analogous to International Classification of Diseases, Ninth Revision/Tenth Revision codes) validated for JIA.24 The study period spanned from registration to the date of the first JIA code (ie, the index date). Secondary case definitions were analyzed to improve the specificity of JIA diagnosis, consisting of the JIA code (same index date) plus: (1) anti-inflammatory prescription (nonsteroidal or glucocorticoid, 2 months before to 1 year after diagnosis) or disease-modifying antirheumatic drugs (postdiagnosis); (2) rheumatology referral; and (3) anti-inflammatory/disease-modifying antirheumatic drug prescriptions and/or rheumatology visits ≥3 months after diagnosis. Children with previous IBD, immunodeficiency, or non-JIA systemic rheumatic diseases (eg, lupus, vasculitis) were excluded.
Ten control subjects without inclusion or exclusion diagnoses were matched according to age and gender to each case subject at the index date by using incidence density sampling. To limit practice-based confounding based on general practitioners’ lack of awareness of JIA, control subjects were randomly selected from practices caring for at least 1 subject with JIA. Control subjects were not matched on practice (except sensitivity analyses) to capture variability in antibiotic prescribing across practices. Matching on practice can cause overmatching when the same prescribing physicians treat both case and control subjects.25
Exposure and Covariate Data
Exposures to systemic antibacterial antibiotics prescribed by general practitioners were examined, characterized according to type, date, and dose (courses or weeks prescribed). Prescriptions written on different days at any time before diagnosis were considered a new course. When course duration was unclear, the mode was imputed according to age and drug class. Additional analyses examined antibiotics categorized as antianaerobic versus non-antianaerobic or by individual antibiotic classes18 (Table 1). Antibiotics were also stratified according to the presence of enterohepatic circulation, using biliary/fecal excretion as a proxy,26–28 with the theory that such drugs might preferentially affect intestinal microbiota. Systemic nonbacterial (eg, antifungal, antiviral) antimicrobial agents were analyzed for comparison.
Characteristics of the Study Population
Potential confounders were demographic variables, comorbidities, previous infections, maternal autoimmunity, and other factors such as hospitalization (Table 1). Diagnoses including infections were identified by using Read codes. Antibiotic courses and hospitalizations within ±1 week of an infection were attributed to that infection. Outpatient visits were tabulated over 2 years starting 2.5 years before the index date, excluding 6 months before diagnosis to limit biased ascertainment of this covariate. For the same reason, outpatient visits and hospitalizations for JIA symptoms (eg, joint swelling, limp) were excluded.
Subjects were matched with mothers by using an algorithm modified from a previous approach.29 Briefly, family relationships are not explicit in THIN due to anonymization, but deidentified household numbers are recorded. Subjects were matched with female subjects from the same household who were aged 12 to 50 years at subjects’ birth. If matches remained ambiguous, codes for pregnancy, labor/delivery, and postnatal period were identified within 270 days before to 90 days after subjects’ birth dates, and eligible maternal age was restricted to 15 to 45 years. Only subjects matched to a unique female were analyzed in models with maternal data.
Statistical Analysis
The association between antibiotic prescription and JIA was determined by conditional logistic regression, accounting for matching. Associations were expressed as odds ratios (ORs) with 95% confidence intervals (CIs). After unadjusted analysis, potential confounders were included in a multivariable model if associated with the outcome in univariate analysis with P < .2; they were retained if they changed the OR by ≥10% or had a P value <.05. Variables with >10% missing data were excluded from the multivariable analysis.
Antibiotic dose models (based on number of courses or weeks prescribed) examined dose as an ordinal or continuous variable, the latter to test for trend. The effect of timing of the first and last antibiotic exposure during the study period was examined. Primary and secondary analyses were repeated based on secondary case definitions. Additional secondary analyses were performed for each antibiotic category (antianaerobic versus other) and class. To consider confounding by indication (whereby infections rather than antibiotics could be associated with JIA risk), we compared the risk of treated and untreated upper respiratory tract infections (URIs) because patients with these common conditions may or may not warrant and receive antibiotic treatment.
Multiple sensitivity analyses were performed (Supplemental Table 5). To further consider confounding by infection, analyses compared the rate of infection between case subjects and control subjects not prescribed antibiotics. Study periods ending 4, 8, and 12 months before the index date were examined to determine whether infections closer to diagnosis were more strongly associated with JIA. To address possible confounding by local environmental factors and practice patterns, analyses were repeated after matching case subjects and control subjects according to practice along with age and gender. Protopathic bias was explored, whereby antibiotics are given for early JIA symptoms, by using as the index date the first joint symptom (eg, stiffness, limp) or rheumatology visit that preceded diagnosis (if known). Because inpatient medications are not well captured in THIN, analyses assumed that antibiotics were received with infection-related or all hospitalizations. Finally, the role of unmeasured confounding was also examined by using the rule out method; this method tests whether a theoretical confounder could nullify results over a range of covariate prevalence and associations with exposure and outcome.30
All analyses were performed by using Stata version 12.1 (Stata Corp, College Station, TX). Hypothesis tests were 2-sided with a type I error of 0.05. P values were derived from regression models. There was no adjustment for multiple comparisons in secondary or sensitivity analyses. This study of anonymized data was exempted by the University of Pennsylvania institutional review board and was approved by THIN’s scientific review committee.
Results
Characteristics of the Study Population
There were 152 cases of JIA diagnosed in 454 457 children meeting the inclusion criteria, within 3.1 million person-years of follow-up (Fig 1, Supplemental Table 6). These data yielded an incidence of 4.9 per 100 000 person-years (female subjects: 6.4 per 100 000; male subjects: 3.6 per 100 000) (Supplemental Table 7).
Flowchart of subject selection. The diagram displays the numbers of patients excluded from the THIN cohort and reasons for exclusion, leading to selection of case subjects with JIA and age- and gender-matched control subjects randomly selected (10:1) at the time of diagnosis.
Because of matching, case subjects and control subjects did not differ in age or gender (Table 1). Previous autoimmunity, including psoriasis and uveitis, was more prevalent among case subjects (P = .002). Case subjects more commonly had a history of infection (P = .01) and hospitalization for infection (P = .01), and they had more infections (P = .03) than control subjects. Case subjects also had more outpatient visits (P < .001). Mothers of case subjects were nearly twice as likely to have autoimmune diseases (P < .001). Cesarean delivery may have been more common among case subjects (P = .07), but this variable had considerable missing data (37%).
Association Between Antibiotics and JIA
After adjustment for matching, other autoimmune conditions, and previous infection, receipt of ≥1 antibiotic prescription was associated with an increased risk of developing JIA (adjusted OR: 2.1 [95% CI: 1.2–3.5]) (Table 2). Adjustment for specific infection types (eg, URI) or number of infections did not appreciably change this association (adjusted OR range: 2.0–2.5). The magnitude of the association increased with additional antibiotic courses (test for trend: P < .001) (Table 3). Models that analyzed dose in weeks prescribed produced a similar dose response.
Association of Any Antibiotic Prescription With JIA
Association of Antibiotic Prescription Dose With JIA
In terms of timing of antibiotic exposure, age of first prescription did not significantly modify the relationship between antibiotic use and JIA (test for interaction: P = .50). In contrast, timing of last exposure was important. Antibiotics prescribed within 1 year of diagnosis (and within 6 months of first joint symptom or rheumatology referral) showed the strongest association with JIA (Table 4). In contrast, untreated infections were not associated with JIA during any time period.
Effect of Timing of Exposure on the Association With JIA
Secondary Analyses
Results of repeat analyses using secondary case definitions were similar (Tables 2 and 3). The association between antibiotics and JIA was also similar for antianaerobic and non-antianaerobic antibiotics, for drugs with and without enterohepatic circulation, and for individual drug classes, although the association for cephalosporins was weak and not significant (Supplemental Table 8). Notably, nonbacterial antimicrobial drugs lacked association with JIA. In exploring possible confounding by infections, having multiple antibiotic-treated URIs was strongly associated with JIA, but there was no association between JIA and multiple untreated URIs. An association with treated URIs persisted after excluding cases of acute otitis media, pharyngitis, and sinusitis.
Sensitivity Analyses
Number of infections was not associated with JIA among exposed (adjusted OR: 1.02 [95% CI: 0.98–1.05]) and unexposed (adjusted OR: 1.06 [95% CI: 0.80–1.42]) subjects. However, in an unmatched sensitivity analysis of unexposed subjects, URI was associated with JIA. This association weakened and was not significant when using an index date 4 to 12 months before JIA diagnosis.
When control subjects were matched on practice, the association between antibiotics and JIA was similar to the original data set, if modestly stronger (Supplemental Table 9). Results were also similar when the first joint symptom or rheumatology referral was used as the index date, and when hospitalization (for infection or otherwise) was considered equivalent to an antibiotic course. Based on the strength of the association (OR: 2.1), only prevalent (20%–60%) and strong (OR: 5–10 with exposure and outcome) unmeasured confounders could explain and nullify these results, calculated by using the rule out method.30 Adjustment for noninfectious hospitalization alone and in combination with maternal autoimmunity and visits had minimal effect on the main results.
Discussion
Previous studies indicate that antibiotic exposure may possibly predispose children to chronic diseases, including IBD18 and JIA.19 Our study supports the hypothesis that antibiotic exposure is associated with an increased risk of developing JIA. This effect was significant after adjusting for confounders such as infection. These findings were also dose dependent, strongest within 1 year of diagnosis, specific to antibacterial antimicrobial agents, and robust to numerous sensitivity analyses. Together, these results suggest a possible role for antibiotics in JIA pathogenesis. This public health finding is potentially important, considering that approximately one-quarter of antibiotics prescribed for children, and an estimated one-half of antibiotics for acute respiratory infections, may be unnecessary and potentially avoidable.31,32
The human microbiome plays important roles in immune regulation and self-tolerance.33 Disturbance of the intestinal microbiome has been linked to several autoimmune diseases, including IBD,34,35 rheumatoid arthritis,36–40 and at least 1 category of JIA.14 After antibiotic treatment, commensal microbial populations often recover within ∼3 months, including in young children.17,41 However, rates of microbial recovery are variable, and incomplete recovery of certain taxa may persist for ≥6 months, particularly after repeated antibiotic exposures.15–17,42 Our finding that antibiotic exposure is most strongly associated with JIA within 1 year of diagnosis (and 6 months of first symptom/referral) supports the hypothesis that antibiotic-induced microbiome dysregulation could precipitate JIA in children predisposed to this disease.
In addition to a causal relationship, alternative explanations for this association are protopathic bias (treatment of early JIA symptoms with antibiotics) and confounding from infection. We found no evidence of protopathic bias in a sensitivity analysis that used first joint symptom or rheumatology referral as the index date. Unfortunately, we could not study or exclude subjects with systemic JIA, who present with symptoms including fever and rash that might initially be confused with infection and treated with antibiotics. However, systemic JIA usually presents acutely and only comprises ∼5% to 10% of all JIA diagnoses.43 This JIA category, therefore, is unlikely to fully explain the association with antibiotics, including exposures 6 to 12 months before diagnosis.
Confounding from infection could occur if infections triggered JIA or if an inherited or acquired immunodeficiency preceded the diagnosis of JIA. We found a significant association between URIs and JIA among unexposed subjects in the months preceding diagnosis. For some subjects, this finding could represent an early predisposition to infection, a short-term infectious trigger, a process exacerbating subclinical arthritis, or an event that merely brought arthritis symptoms to medical attention. Antibiotic-untreated infections can also precipitate changes in children’s microbiota.41 JIA has rarely been reported in association with a definable immunodeficiency,44 but children diagnosed with JIA are at higher risk for serious infection independent of disease treatment.45 In our study, case subjects had more office visits than control subjects and were more likely to be hospitalized for infection. It is unclear whether more severe infections led to more antibiotics, which then triggered autoimmunity in susceptible subjects, or whether JIA-associated immune dysfunction caused more severe illnesses beforehand, for which antibiotics were a marker. Because office visits and hospitalizations could be on the causal pathway between antibiotics and JIA, we did not adjust for these variables in the primary analyses. Of note, the lack of association of JIA with nonbacterial antimicrobial agents argues against the immunodeficiency hypothesis. Indeed, these findings, and the stronger association of JIA with antibiotic-treated URIs than with untreated URIs, support a causal model for antibiotics.
Our study has several strengths. Outpatient prescription data are comprehensive in THIN, and most antibiotics in general pediatrics are prescribed in the outpatient setting, ensuring near complete capture of antibiotic exposure in these children followed up from early infancy. The failure of some subjects to take some or all of the prescribed medication would likely bias results toward the null, suggesting that our study may underestimate the true effect. Although inpatient prescription data were not available, models assuming inpatient antibiotic exposure were similar. Infections were important confounders in the relationship between antibiotics and JIA, and timing of last antibiotic exposure proved important. Our analyses incorporating these factors made this study unique compared with an earlier report.19 JIA itself has previously been validated with 86% positive predictive value.24 The incidence (overall and gender-specific) of JIA in our cohort was similar to earlier published estimates from other western European countries using current classification criteria.46,47 Secondary case definitions designed to increase the specificity of JIA diagnoses yielded similar results. Finally, multiple sensitivity analyses were consistent and robust to a wide range of assumptions.
This study has several limitations. Although JIA has been validated, specific JIA categories have not been validated and are usually unspecified in THIN. This absence made it difficult to discern whether antibiotics were associated only with particular JIA categories. If this association were category-specific, then the true effect could be larger for certain forms of JIA and null for others. Similarly, race and ethnicity correspond to differences in JIA presentation48 but are poorly captured by THIN data. The study’s inclusion of children registered early in life also identified a relatively young cohort. This design limited the study’s generalizability to older children and adolescents, and compromised our ability to study the interaction between age and antibiotic exposure on JIA risk, including age of diagnosis. In addition, we could not detect substantial differences across antibiotic classes, as was previously shown for pediatric IBD18 and JIA in another study that reported relatively greater risk for cephalosporins and clindamycin19; our study was not powered to address this specific issue. Finally, despite our efforts to account for infection and our many analyses implicating antibiotics in the development of JIA, we cannot definitively rule out residual confounding from infections.
Conclusions
The present study found that treatment with antibiotics was associated with the development of JIA in a large general pediatric population. This relationship was dose dependent, strongest within 1 year of diagnosis, specific for antibacterial agents, persisted after adjustment for infection, and was robust to numerous assumptions. These findings suggest a potential role for antibiotics in the pathogenesis of JIA, perhaps mediated through changes in the microbiome. If this association is causal, antibiotics could be considered a potentially modifiable risk factor for JIA, especially in light of the overprescribing of antibiotics to children, particularly for respiratory tract infections. Children with JIA may also be at risk for more infections before diagnosis due to immune dysfunction, and a causal role for infections remains a possibility. Further research is necessary to confirm these findings in other populations; to determine whether this association depends on age, JIA category, and antibiotic drug class; and to investigate underlying mechanisms.
Acknowledgment
The authors thank Molly Collins, MD, for her critical review of the manuscript.
Footnotes
- Accepted May 11, 2015.
- Address correspondence to Daniel B. Horton, MD, Office of the Chancellor, Rutgers Biomedical & Health Sciences Child Health Institute of New Jersey, 4th floor, 89 French St, New Brunswick, NJ 08901. E-mail: daniel.horton{at}rutgers.edu
Dr Horton conceptualized and designed the study, conducted the analyses, and drafted the initial manuscript; Drs Scott and Rose provided input in study design and critically reviewed the manuscript; Dr Haynes provided input in study design, assisted with data extraction, and critically reviewed the manuscript; and Drs Lewis, Putt, and Strom provided input in study design, provided guidance on analyses, and critically reviewed the manuscript. All authors approved the final manuscript as submitted.
Dr Horton's current affiliation is Rutgers Biomedical and Health Sciences, New Brunswick, New Jersey.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: All phases of this study were supported by grants from the National Institutes of Health (NIH): NIH NRSA T32-GM075766 Clinical Pharmacoepidemiology Training Grant (Dr Strom [Principal Investigator], Dr Horton); NIH NRSA F32-AR066461 Individual Fellowship (Dr Horton); NIH K08-DK095951 (Dr Scott); and NIH K24-DK078228 (Dr Lewis). Funded by the National Institutes of Health (NIH).
POTENTIAL CONFLICT OF INTEREST: Dr Rose served on the data and safety monitoring committee for Bristol-Myers Squibb for a trial of the pediatric formulation of abatacept for the treatment of juvenile idiopathic arthritis, and Dr Strom has consulted for the following antibiotic manufacturers (on matters unrelated to the article): Abbott, AbbVie, AstraZeneca, Bayer, Bristol-Myers Squibb, GlaxoSmithKline, Lundbeck, Roche, Sanofi, Takeda, and Teva. Dr Lewis has served as a consultant for the following antibiotic manufacturers (on matters completely unrelated to this manuscript): AbbVie, AstraZeneca, Janssen Pharmaceuticals, Medimmune, Merck, Takeda, Shire. He has served on a Data and Safety Monitoring Board for clinical trials (unrelated to antibiotics) sponsored by Pfizer, another antibiotic manufacturer. He has also served as a consultant for Nestle Health Science and Rebiotix, companies studying therapies for intestinal health (on matters completely unrelated to this manuscript). The other authors have indicated they have no potential conflicts of interest to disclose.
COMPANION PAPER: A companion to this article can be found on page e492, online at www.pediatrics.org/cgi/doi/10.1542/peds.2015-1296.
References
- Copyright © 2015 by the American Academy of Pediatrics
In this issue
