FormalPara Key Points

Compared with past use, current use of antibiotics in children was associated with an increase in the risk for liver injury.

Substantial differences in risk estimates have been found among individual antibiotics.

Paediatricians should be aware of the potential increase of liver enzymes in children taking ceftriaxone, sulphamethoxazole/trimethoprim combination or clarithromycin, even for short periods.

1 Introduction

Antibiotics are the most common drug class causing liver injury in the general population [1–4]. Antibiotic-induced hepatotoxic reactions are usually idiosyncratic, unpredictable and present a poorly understood pathogenesis [5–7]. The diagnosis of liver injury is challenging due to heterogeneous clinical manifestations, ranging from transient, mild, asymptomatic liver function abnormalities to rare, potentially fatal, acute liver failure [5]. Moreover, especially for antibiotics, causality assessment is difficult because the indication for antibiotic treatment acts as confounder. In fact, it is known that bacterial infection may lead to changes in hepatic enzymes, representing an underlying cause of liver injury per se [7].

The diagnosis of antibiotic-induced liver injury is even more of a challenge in the paediatric setting because of the age-dependent maturation of the cytochrome P450 enzymes involved in the antibiotic (and overall medicines) metabolism [8].

Although several case reports in paediatrics suggest that antibiotic-induced liver injury in children would be likely, at least as in the general population [7], only a few studies addressed this issue specifically in paediatrics [1, 2, 9–11]. Moreover, spontaneous adverse drug reaction (ADR) reporting systems showed that antibiotics are most frequently implicated in hepatic ADRs in children and adolescents [1]. However, since these data lack denominator [12–14], findings from spontaneous reports may also be explained by the widespread use of these drugs in the paediatric population [15]. Moreover, a previous population-based study investigating the power of real-world data mining on electronic healthcare databases to assess hepatic drug safety in paediatric outpatients showed that antibiotics had the highest risk of acute liver injury compared with non-use of any drug (age- and sex-adjusted relative risks: 25.9, 95% confidence interval [CI] 13.4–50.0, for clarithromycin; 18.6, 95% CI 11.3–30.6, for amoxicillin/clavulanic acid; and 7.5, 95% CI 3.4–16.8, for amoxicillin) [2]. However, confounding by indication and protopathic bias could not be fully excluded.

Given that antibiotics are more commonly prescribed in children than in adults, and since no studies have so far examined antibiotic-induced liver injury specifically in the paediatric population, we conducted a large, multi-database, population-based, case-control study to investigate the risk estimates of liver injury associated with individual antibiotics in children and adolescent outpatients.

2 Methods

2.1 Source Population

We selected all children and adolescents younger than 18 years of age from three longitudinal electronic primary care databases in two European countries: (i) Pedianet, a family paediatrician (FP) registry; (ii) Health Search–IMS Longitudinal Patient Database (HSD), a general practice (GP) registry from Italy; and (iii) the Integrated Primary Care Information (IPCI), a GP database from The Netherlands.

All three databases contain anonymous data on patient demographics, reasons for clinic visits, medical diagnoses by GP/FP and specialist, hospitalizations, drug prescriptions, and laboratory and other diagnostic findings. In The Netherlands, the paediatric population receives medical care from GPs, while, in Italy, medical care is provided by FPs (up to 14 years of age) and GPs (over 14 years of age). These databases are representative of the Italian and Dutch paediatric populations and have been proven to be valid data sources for pharmacoepidemiological studies [1, 15–19]. The study period ran from 1 January 2000 to 31 December 2008.

2.2 Case and Control Selection

From this population, we excluded all children with clear competing causes of liver injury, including viral infections, hepatic neoplasm, autoimmune hepatitis, neonatal jaundice, genetic hepatopathy, biliary tract diseases and abdominal trauma. Details on case definition, identification and validation have been previously described [2]. In brief, by applying a similar stepwise approach across all three databases, all potential cases were extracted using (i) International Classification of Primary Care (ICPC) codes (as used in the IPCI database) or International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes (as used in the Pedianet and HSD databases) for hepatic reactions/signs (i.e. hepatitis, liver failure, hepatic steatosis, hepatic cirrhosis, hepatic necrosis, hepatomegaly, or jaundice); (ii) specific keywords for free-text search; and (iii) laboratory age-specific values of liver function tests (i.e. alanine aminotransferase [ALT], aspartate aminotransferase [AST], alkaline phosphatase [AP] and total bilirubin). For each potential case, the complete electronic medical record history (including results of laboratory data, ultrasound and other diagnostic tests, as well as hospital discharge summaries and specialists’ letters) was then manually validated by four medically trained investigators, blinded to the drug exposure. Based on Council for International Organizations of Medical Sciences (CIOMS) criteria and in accordance with previously published evidence [20–24], cases of liver injury were defined as (i) any age-specific increase of more than two times the upper limit of normal (ULN) range for liver function tests, i.e. ALT, AST, AP or total bilirubin, or their combinations; or (ii) diagnosis of liver injury confirmed by either a specialist, GP or FP, or via ultrasound evidence [20, 21]. Specific search terms, such as jaundice and hepatomegaly, which are suggestive of liver injury but are not sufficient by themselves to confirm a diagnosis of livery injury, were considered only in association with other more specific symptoms/signs (e.g. abnormal liver enzyme values, steatosis). Children with elevation of biochemical liver tests (i.e. ALT, AST, or AP) less than two times the ULN, or with isolated increases of γ-glutamyltransferase, were excluded as cases [21, 25]. In case of uncertainty, cases were reviewed by expert medical doctors with the aim of reaching consensus on whether the case was indeed indicative of liver injury [2].

The index date of the event was defined as the earliest date of the hepatic symptoms/signs (i.e. fatigue, weakness, anorexia, nausea, jaundice, dark urine, light stools, itching and bloating) or, in the absence of these, the date of abnormal liver tests immediately preceding the diagnosis.

Within the same underlying study population, we selected up to 100 control participants for each case at the same index date, through incidence density sampling, according to which the likelihood of being selected as a control is proportional to the person-time [26]. Controls were matched to the corresponding case on index date, year of birth, sex, and database.

2.3 Exposure Definition

Antibiotic exposure (Anatomical Therapeutic Chemical [ATC] code J01*) was evaluated based on prescription data from these healthcare databases. To estimate the association between antibiotic use and liver injury, we created exposure categories based on timing and duration of use. Exposure was categorized as current if the index date fell during antibiotic exposure or within 15 days after the end of the prescription (i.e. carryover period), recent if the last prescription ended within 16–90 days before the index date, or past if it ended more than 90 days before. No use was defined as the absence of a prescription before the index date.

2.4 Patient Comorbidities

Several comorbidities were addressed as potential risk factors for liver injury. Patient medical histories were screened for diagnoses of diabetes mellitus, hypoglycemia, obesity, hyperlipidemia, nutrition-related disorders, hyper- or hypothyroidism, hypertension, or alcohol intake or smoking within 1 year before the index date, while congenital diseases were identified any time before the index date. We also considered concomitant use (i.e. within 3 months of the index date) of other potential hepatotoxic medications, including antimycotics, drugs for the treatment of tuberculosis, drugs for acid-related disorders, anticonvulsants, drugs for respiratory disorders, paracetamol and its combinations, nervous system drugs (such as psycholeptics and psychoanaleptics), non-steroidal anti-inflammatory drugs (NSAIDs), or immunosuppressants.

2.5 Main Analyses

We compared characteristics of cases and controls using conditional logistic regression. Covariates associated with liver injury in the univariate analysis at a p-value <0.10, and those that changed the point estimate of the association between antibiotics and liver injury by more than 10%, were included in the final adjusted model [27–29].

To minimize the effect of confounding by indication, past use of any antibiotic served as the reference category, instead of non-use of antibiotics [30]. Children who have never been treated with antibiotics during the study period may be healthier than those receiving antibiotics, and ultimately this may result in an overestimation of the risk of liver injury for all antibiotic users. Several multivariate models have been run as primary analyses to estimate the odds ratios (ORs), together with 95% CIs, as a measure of the association between liver injury and current use of antibiotics grouped as follows: (i) antibiotics overall; (ii) antibiotics by class; (iii) individual antibiotics.

2.6 Sensitivity and Subgroup Analyses

To rule out possible effects of outcome misclassification, we repeated all the analyses in a dataset restricted to patients for which liver injury was strictly defined as more than twice higher than the ULN of laboratory parameters and confirmed by specialists.

To better address confounding by indication due to current infections, a sensitivity analysis was conducted using current use of amoxicillin as a reference group. Amoxicillin was chosen since it is the most commonly prescribed antibiotic in children [31] and is considered to be ‘non-hepatotoxic’ when used as a single ingredient compared with its combination with clavulanic acid [15, 32].

To investigate exposure misclassification, we changed the risk window from 15 days to 0 days (i.e. no carryover period at all). In order to explore the impact of age as an effect modifier, we stratified the analysis by age category.

All analyses were conducted using SPSS version 20 (IBM Corporation, Armonk, NY, USA). We used a p value of <0.05 as the threshold of statistical significance, except for the selection of the covariates to be included in the final multivariate models (p < 0.10). Wald’s test was used to compare characteristics among cases and controls.

3 Results

3.1 Main Analysis

In the database source population of 429,772 children and adolescents (<18 years of age) in Italy (n = 145,706 from Pedianet and n = 190,772 from HSD) and The Netherlands (n = 93,294 from IPCI), we identified 938 cases of liver injury after exclusion of all clear competing causes. These cases were matched to 93,665 controls. Case characteristics are described in the electronic supplementary table.

The majority of cases were males (58.2%) with a mean age of 11.3 years (standard deviation 5.1). Cases had a greater burden of comorbidities such as diabetes, hyperlipidaemia, obesity, thyroid disorders or congenital disease than controls. Children with liver injury were more likely to be currently exposed to acid-suppressant drugs, anticonvulsants, NSAIDs, psycholeptic agents, paracetamol, and anti-asthmatics than children without liver injury (Table 1).

Table 1 Demographic and clinical characteristics of cases of liver injury and matched controls from the paediatric population identified
Full size table

Compared with past use, current use of antibiotics overall (adjusted OR [ORadj] 3.22, 95% CI 2.57–4.03) was significantly (p < 0.001) associated with an increased risk of liver injury. This association, although less strong, was also observed for recent use of any antibiotic (ORadj 1.53, 95% CI 1.24–1.89; p = 0.043). With regard to different antibiotic classes, we found some heterogeneity, with the lowest risk estimates being those for penicillins (ORadj 2.83, 95% CI 2.06–3.90) and the highest for fluoroquinolones (ORadj 13.87, 95% CI 4.81–39.95) [Table 2].

Table 2 Associations between the use of antibiotica therapeutic classes and risk of liver injury in the paediatric population identified
Full size table

Table 3 shows the risk estimates of liver injury for each individual antibiotic compared with past use of any antibiotic. Except for azithromycin, the risk was significantly increased for current use of each antibiotic (p < 0.005), varying from the lowest ORadj of 1.86 (95% CI 1.08–3.21) for amoxicillin to the highest ORadj of 24.16 (95% CI 11.78–49.54) for cotrimoxazole (i.e. sulphamethoxazole/trimethoprim) and 26.70 (95% CI 12.09–58.96) for ceftriaxone.

Table 3 Associationsa between individual antibioticsb and the risk of liver injury in the paediatric population identified
Full size table

3.2 Sensitivity and Subgroup Analyses

When considering current use of amoxicillin as the reference group, we found a statistically significant increase in the risk of liver injury for ceftriaxone (ORadj 14.35, 95% CI 5.58–36.87), cotrimoxazole (ORadj 12.98, 95% CI 5.34–31.53) and clarithromycin (ORadj 2.29, 95% CI 1.04–5.07). The association was still observed for all other antibiotics, such as amoxicillin/clavulanic acid, cefaclor, cefixime, ceftibuten, however results were not statistically significant due to limited statistical power (Table 4).

Table 4 Association between individual antibioticsa and the risk of liver injury in paediatric outpatients using current use of amoxicillin as the comparator
Full size table

To estimate the effect of potential misclassification of exposure, we removed the carryover period in a sensitivity analysis, which yielded an increase in the risk of liver injury by approximately 20% for current use of amoxicillin/clavulanic acid (ORadj 3.34, 95% CI 1.74–6.43), 50% for current use of amoxicillin (ORadj 2.82, 95% CI 1.51–5.28), 10-fold for current use of clarithromycin (ORadj 46.41, 95% CI 30.86–69.77) and 25-fold for ceftibuten (ORadj 93.22, 95% CI 43.59–199.36) [Table 5].

Table 5 Effect of exclusion of the carryover period on the association between individual antibioticsa and the risk of liver injury in the paediatric population identified
Full size table

When restricted to only cases of abnormal liver parameters confirmed by a specialist, the associations between current use of antibiotics and liver injury did not change substantially (Tables 2, 3). In terms of individual antibiotics, the increased risk was confirmed for amoxicillin/clavulanic acid and ceftriaxone, and became stronger for cefixime and clarithromycin. This analysis was limited by statistical power given the few cases exposed to cefaclor, ceftibuten, and cotrimoxazole. The analyses were not stratified by age category because of the very low number of exposed cases among each age subgroup.

4 Discussion

In this population-based, multi-database, case-control study, we found that current use of antibiotics in children and adolescents was associated with a threefold increased risk of liver injury when compared with past use. Risk estimates differed among antibiotic classes and varied even more among individual antibiotics belonging to the same subclass. Although these associations may be partly explained by confounding due to current infections, the use of cotrimoxazole, ceftriaxone, and clarithromycin still remained associated with a higher risk of liver injury when the potential effect of confounding by indication has been ruled out.

To the best of our knowledge, no other paediatric population-based studies have addressed the association between liver injury and individual antibiotic use specifically providing risk estimates. Thus, our results can only be compared with adult data, descriptive studies from drug/induced liver injury registries [11] and results from signal detection analyses [1, 10].

Fluoroquinolones, sulfonamides, tetracyclines, cephalosporins, macrolides and penicillins have been associated with liver injury [6, 33–36]. Of course, variations on risk estimates across antibiotic classes depend on different pharmacodynamics and pharmacokinetics which play a crucial role in their manifestations of liver injury [6, 25, 32, 35].

Our results confirmed the high risk of ceftriaxone-induced hepatitis or elevated liver enzymes, as already described in a few case reports in children/adolescents [37–39]. Moreover, the high risk is also supported by our previous finding from signal detection analysis on ADR spontaneous reporting systems in children [1]. Clinical manifestation of the ceftriaxone-induced hepatitis may represent a direct toxic effect, an idiosyncratic reaction, or a cholestatic injury associated with its calcium precipitation, which is known to typically occur after 9–11 days of treatment [6, 33, 34]. Cotrimoxazole-induced liver injury is well-described in adults and has also been detected as an hepatotoxic signal in children from ADR spontaneous reporting system analysis [18, 40–43]. As in adults, the typical presentation is sudden occurrence of fever and rash, followed by jaundice within a few days or weeks of starting the medication, and the typical pattern of serum enzyme elevations is mixed or cholestatic and often asymptomatic. The mechanism underlying sulphonamide liver injury is probably immunoallergic [40].

In contrast to previous evidence [25, 44, 45], our results showed different hepatotoxic profiles amongst macrolides. A higher risk was observed for rokitamycin (withdrawn from the Italian market in 2013) and clarithromycin, while the association was not significant for azithromycin. The effect of reducing carryover time on the risk estimate for clarithromycin is consistent with the proposed mechanism suggesting the short-term onset of liver injury [45, 46].

Consistent with existing evidence, amoxicillin with clavulanic acid is associated with a higher risk of liver injury than amoxicillin alone, supporting the potential role of clavulanic acid in the toxic pathway [2, 25, 32]. Nevertheless, we cannot definitely exclude an increased risk, however small, of liver injury associated with amoxicillin use.

4.1 Strengths

First, given the large study population identified from three longitudinal, nationally representative GP and FP databases, the results can be largely generalized to the paediatric population in these countries. Second, these electronic registries are maintained for daily routine healthcare purposes and the exposure is prospectively collected, thus limiting the possibility of recall bias. Third, we were able to adjust the analyses for many potential confounders because of the availability of clinically relevant information in the study databases. In addition, we confirmed, in the paediatric setting, some risk factors for liver injury only known in adults, such as underlying diabetes, obesity, hyperlipidaemia, thyroid imbalance or congenital diseases. Fourth, confounding by indication is a main concern when studying the association between antibiotics and liver injury. Accordingly, the sensitivity analysis in which current exposure to amoxicillin was used as the reference category allowed to control for this potential confounding because amoxicillin is the most frequently used antibiotic in children [31] and is usually considered less hepatotoxic than other antibiotics [25, 32]. Thus, the risk estimate during amoxicillin exposure can be regarded as a proxy of the background risk of liver injury [32].

4.2 Limitations

This study has some potential limitations because of its observational nature. Due to the limited number of exposed cases, we could not explore the effect of heterogeneity by country. Residual confounding due to unmeasured severity of infection cannot be excluded. Moreover, although we carefully excluded viral infections as underlying disease, they still may represent the non-documented indication for antibiotic prescription [47–49].

With regard to liver injury case selection, we adopted a very sensitive search strategy, as in previous database studies investigating the same association and thereafter manually validating all automatically detected potential cases. Nevertheless, it is likely that outcome misclassification (if any) is randomly distributed among those exposed and unexposed to antibiotics, thus again eventually leading to risk dilution.

As we used outpatient prescription and no dispensing data, we might have misclassified the exposure. However, if present, such a bias would likely be non-differential between cases and controls, thus underestimating the actual risk.

We could not exclude the potential effect of diagnostic bias on the risk estimates because children exposed to specific well-known hepatotoxic antibiotics might receive liver function tests more likely than children exposed to other drugs. Moreover, the analyses were not stratified by dosage; however, since antibiotics are usually responsible for idiosyncratic liver injury reactions, i.e. ‘not dose-related’ by definition [50], it is unlikely that the risk of liver injury is influenced by the dose of antibiotic.

The system of medical record databases did not allow to collect, and then to explore, the over-the-counter medications, such as paracetamol, well-known to be hepatotoxic in children. Thus, although we were able to identify paracetamol as a potential risk factor for liver injury in children, despite the low number of cases and controls exposed, we failed to test it as an effect modifier.

Lastly, the limited number of cases exposed to individual antibiotics resulted in wide CIs, particularly for cephalosporin antibiotics. As a result, their risk estimates need to be interpreted with caution.

5 Conclusion

The use of antibiotics in paediatric outpatients is associated with an increased risk of liver injury, with substantial differences in risk among individual antibiotics. In particular, after several analyses, the potential risk of liver injury in children was found to be associated with current exposure to ceftriaxone, cotrimoxazole and clarithromycin. Paediatricians should be aware of this risk when using these antibiotics, even if for short periods. From a methodological point of view, this study demonstrates that combining data from different databases is crucial in paediatric postmarketing surveillance to provide the large sample size required for the adequate assessment of drug safety profiles in routine clinical care. However, a larger and more heterogeneous sample size is needed to investigate safety in terms of less commonly used antibiotics, or even other medications.