1 Introduction

A previous epidemiologic study performed on a liver disease registry documents that the aetiology of acute liver failure is not determined, i.e. idiopathic, in almost 50 % of paediatric patients, is drug-related in around 15 %, while other causes (i.e., autoimmune hepatitis or hepatopathy due to metabolic diseases or viral infections) account for less than 10 % each [1]. Published data on the incidence and characteristics of less severe acute liver injury (ALI) are scarce because initially the disease is asymptomatic and, thus, difficult to recognize [2]. So far, information on clinical features of ALI are only available from disease registries [3–5].

The increasing availability of electronic healthcare record (EHR) and claims databases allows for the conduct of population-based epidemiologic studies in larger populations, including children and adolescents [6]. To date, no data are available on the pattern of disease manifestation, severity of the features, and incidence of ALI, either idiopathic or potentially drug-attributed, in children and adolescents.

In this retrospective, population-based study we explored the incidence of idiopathic ALI in a general paediatric population (<18 years old) from three longitudinal EHR databases in two European countries, Italy and The Netherlands. Our main aims were to quantify the incidence of idiopathic ALI and to describe its clinical features. Additionally, we investigated the role of drug exposure as the underlying potential cause of apparently idiopathic ALI by applying data mining techniques on the pooled data from the three databases.

2 Patients and Methods

2.1 Source Population

A retrospective cohort study was conducted combining data from three longitudinal EHR databases in two European countries: Pedianet (from 2000 to 2008) and Health Search/CSD Longitudinal Patient Database (HSD; from 2002 to 2008) in Italy and the Integrated Primary Care Information database (IPCI; from 2001 to 2007) in The Netherlands [7]. In The Netherlands, general practitioners (GPs) serve as gatekeepers to medical care for all patients, including children and adolescents. In Italy, children receive medical care by family paediatricians (FPs) up to the age of 14 years, and thereafter by GPs. All these databases have been proven valid data sources for pharmacoepidemiological studies [8–13]. Pedianet was set up in the year 2000 and comprises healthcare records of around 150,000 children 0–14 years old provided by 150 FPs distributed all over Italy. HSD is a GP database that was established in 1998 by the Italian College of General Practitioners. HSD currently contains records of around 1.4 million patients (190,772 patients <18 years) from over 900 GPs throughout Italy. Pedianet and HSD were combined together in the analysis to represent the whole Italian paediatric population (<18 years). IPCI is a Dutch GP database that was set up in 1992 and currently contains records from 600 GPs, covering approximately 1 million patients (93,294 patients <18 years), with an age and gender distribution that is representative for The Netherlands.

All three databases contain anonymous data on patient demographics, reasons for visits, diagnoses from GPs/FPs and specialists, hospitalizations, drug prescriptions, laboratory and other diagnostic findings for the paediatric population. Symptoms and medical diagnoses are either registered as free text or coded using ICPC (International Classification of Primary Care) in IPCI and ICD-9-CM (International Classification of Disease, 9th revision, Clinical Modification) in HSD and Pedianet. Drug prescription data include details on product name, formulation, dosing regimen and indication of use. All drugs are coded according to the Anatomical Therapeutic Chemical classification system [14].

2.2 Study Population

The retrospective study population comprised a dynamic cohort of all children and adolescents aged <18 years, who were registered with GPs/FPs in any of the three databases for at least 6 months. Newborns in the study period immediately entered the cohort (no prior history required). Follow-up of patients started from cohort entry until one of the following events, whichever occurred first: diagnosis of liver injury, death, transferring out of the practice, end of the study period (31 December 2008) or 18 years of age. Patients with a diagnosis of liver injury (irrespective of the cause) prior to the study entry were excluded from the study. The study period ran from 2000 to 2008.

2.3 Outcome Definition and Case Ascertainment

According to the CIOMS (Council for International Organizations of Medical Sciences) criteria and previous evidence [15–20], a liver injury was defined as an increase of more than two times the upper limit of the normal (ULN) range in alanine aminotransferase (ALT) or a combined increase in aspartate aminotransferase (AST), alkaline phosphatase (AP) and total bilirubin, provided that one of them was twice the upper limit of the respective normal range. Jaundice and hepatomegaly, which are suggestive of liver injury but are not sufficient by themselves for the diagnosis, were considered only in association with other specific symptoms/signs (e.g. abnormal liver enzyme values, steatosis). Patients with elevation of biochemical liver tests ≤2 ULN only in presence of isolated increase of gamma-GT or neonatal jaundice were excluded. According to our aims, we excluded all clear competing causes of liver injury: (i) viral infections (Hepatitis A virus [HAV], Hepatitis B virus [HBV], Hepatitis C virus [HCV], Cytomegalovirus [CMV], and Epstein-Barr virus [EBV]); (ii) hepatic neoplasm; (iii) autoimmune hepatitis; (iv) genetic and metabolic disorders-related hepatopathy (e.g., hemochromatosis, α1-antitrypsin deficiency, Wilson Disease, Gilbert Syndrome, etc.); (v) biliary tract diseases (i.e., biliary atresia, gallstones, cholangitis, and cholecyctitis); and (vi) abdominal trauma documented with imaging. Cases with drugs as potential cause were not excluded, as it is difficult to assess causality [18, 21].

ALI was identified through a similar stepwise approach across all three databases. First, all the potential cases of liver injury were extracted through a very broad automated search using both free text and diagnostic codes for hepatitis, liver failure, hepatopathy, hepatic steatosis, hepatic cirrhosis, chronic liver disease, or hepatic necrosis. Second, all of these records were independently validated by four medically trained investigators, two for each country, who were blinded to the drug exposure and native speakers of the language of the concerned database. For each potential case, the whole clinical diary history, including results of laboratory data, ultrasound and other diagnostic test, as well as hospital discharge summaries and specialists’ letters were reviewed. All cases of chronic liver injury were excluded. Based on validation, cases were classified as definite, with a diagnosis confirmed by a specialist or laboratory data/ultrasound evidence; and possible, with a diagnosis made by a GP/FP. In addition, cases of isolated hepatomegaly were assessed separately. In case of disagreement between the two assessors, a third expert medical doctor arbitrated.

All cases were further characterized, whenever possible, based on the source of diagnosis. According to the CIOMS criteria [15] and to Benichou [16], with respect to biochemical hepatic function, the pattern of liver injury was defined as (a) hepatocellular (or cytolytic), with an increase of more than twice the ULN of ALT; (b) cholestatic, with an increase of AP; or (c) mixed, when both ALT (or AST) and AP increased. Similarly, we categorized the degree of severity as mild (more than twice the ULN), moderate (more than three times the ULN), severe (more than five times the ULN), and most severe (more than eight times the ULN) [22].

The index date was defined as the onset of the first symptom/sign related to liver injury (i.e., fatigue, weakness, anorexia, nausea, jaundice, dark urine, light stools, itching, bloating and abdominal pain). In our analysis, we only included incident cases, i.e. newly diagnosed ALI during the study period and no diagnosis in the past six months.

2.4 Drug Exposure

Drug exposure was evaluated based on prescription data from healthcare databases. As drug prescriptions are locally coded using the national product codes, which differ among countries, we defined drug exposure based on the Anatomical Therapeutic Chemical (ATC) classification system of the WHO. The ATC code was used as the drug code in the input files for Jerboa©. Cases were considered as exposed to a certain drug if the prescription for that drug overlapped the date of onset/ascertainment of liver injury, based on the estimated duration of exposure. Each database owner estimates the duration covered by each prescription according to the legend duration (if dosing regimen is available), or is otherwise based on the defined daily dose (DDD) [23].

2.5 Statistical Analysis

Our main analyses focused on definite cases. Jerboa© software was used to estimate the incidence rates of cases of idiopathic ALI, as both pooled and country-specific. The validity of the software has been previously described [23–25]. Briefly, Jerboa© elaborates input files with the common data model (including patient unique identifier and demographics, follow-up time, exposure and outcome information) from different databases and produces as the output several parameters [e.g. incidence rates (IRs), relative risks (RRs), prescription rates (Rx/PYs)], stratified by age, sex and calendar years [6, 23]. Age was categorized into four categories: <2 years, 2 to ≤5 years, 6 to ≤11 years and 12 to <18 years.

To calculate the incidence rates, we considered the number of first event of ALI after a 1-year run-in period as the numerator and the number of cumulated person-years of the study population at risk of developing the first event of ALI, as the denominator. IRs were expressed as rate/100,000 PYs plus 95 % CI. In addition, we compared IRs between the two countries and calculated by direct method the standardized IR in Italy using the distribution of Dutch population as a reference.

To study the effect of outcome misclassification, we conducted a sensitivity analysis combining definite and possible cases as well as cases of isolated hepatomegaly.

In order to investigate the associations between drug exposure and apparently idiopathic ALI, we linked prescribing data from all combined databases to case occurrence in a temporal manner, by estimation of IRs of ALI during exposure and comparing the drug-specific rates with background rates. We only considered the definite cases for this analysis. Based on the background IR of ALI, using Mantel–Haenszel method, we calculated pooled age- and sex-adjusted rate ratios (RRs, with 95 % CI), for each drug associated with at least two event occurring its use (ATC V level) using the non-use of the drug of interest as comparator [6, 23].

3 Results

During the period 2000–2008, the study cohort comprised 429,772 subjects (of which 68 % were from Italy and 32 % were from The Netherlands) aged <18 years.

After exclusion of all clear causes of liver injury and of chronic hepatitis through case-by-case validation (Fig. 1), we identified 1326 cases of idiopathic ALI of which the majority originated from Italy (1237 cases, 93.3 %) (Table 1). In the two Italian databases, 731 cases (59.1 % of 1237) were classified as definite, 285 (23.0 %) as possible, while 221 (17.9 %) were cases of isolated hepatomegaly. In The Netherlands, 54 (60.7 % of 89) were classified as definite, 23 (25.8 %) as possible cases, and 12 (13.5 %) as isolated hepatomegaly.

Fig. 1
figure 1

Flow chart of cohort and cases selection. CIOMS Council for International Organizations of Medical Sciences, HSD Health Search/CSD Longitudinal Patient Database, IPCI Integrated Primary Care Information database, ULN upper limit of normal

Full size image
Table 1 Features of idiopathic acute liver injury by countries
Full size table

3.1 Features of Idiopathic Acute Liver Injury

An overview of the clinical features of idiopathic ALI in the general population aged 18 years or younger is described in Table 1. In Italy, asymptomatic liver enzyme elevations were reported in around two-thirds of definite cases (n = 487, 66.6 %), followed by cases of hepatitis (n = 141, 19.3 %) and hepatic steatosis (n = 66, 9.0 %). Diagnoses were confirmed by diagnostic tests (i.e. laboratory data and hepatic ultrasounds) in 70.2 % and by specialists in 24.9 % of the cases. In The Netherlands, most of the definite cases were combinations of multiple signs/symptoms of liver injury (n = 26, 48.1 %) and asymptomatic liver enzyme elevations (n = 25, 46.3 %). The diagnoses were confirmed by diagnostic tests in 48.1 % of the cases.

According to liver enzyme elevations, idiopathic ALI presented hepatocellular in 52.6 % and cholestatic in 39.4 % of the cases in Italy. Around 80 % of those were considered mild or moderate. In The Netherlands, the pattern of idiopathic ALI seems to be similar, but the contribution of hepatocellular cases (76.0 %) was higher compared to Italy. Likewise, the degree of severity of ALI is likely consistent with the Italian degree of severity, because most of the Dutch cases (64.0 %) presented a mild to moderate degree of severity. In The Netherlands, the number of cases is too low to conduct a formal comparison with the Italian data.

3.2 Epidemiology of Idiopathic Acute Liver Injury

Among the pooled 785 definite cases, the crude IR of idiopathic ALI was 62.4 (95 % CI 58.1–66.8) per 100,000 PYs. Per country, the IR was 3.5 times greater in Italy (73.0/100,000 PYs; 95 % CI 67.8–78.4) than in The Netherlands (21.0/100,000 PYs; 95 % CI 16.0–27.2). Standardized to the population distribution of The Netherlands, the incidence in Italy was 81.1/100,000 PYs (95 % CI 70.7–92.7) [Table 2]. Age distribution was similar in the two countries (mean age: 12.1 ± 4.6 in Italy versus 12.9 ± 4.5 in The Netherlands, P = 0.105), while a statistically significantly higher proportion of boys with idiopathic ALI was found in Italy as compared to The Netherlands (60.9 vs. 44.4 %, P < 0.001). Males showed a higher incidence than females in Italy (84.8/100,000 vs. 59.9/100,000 PYs, respectively), whereas the opposite trend was observed in The Netherlands (18.3/100,000 vs. 23.9/100,000 PYs, respectively) [Table 2]. In both countries, the IRs of idiopathic ALI increased with age, even if a slightly increase of the rates was observed among the neonates up to 1 years. In particular, among the total cases, also including hepatomegaly, a notable peak of the incidence rate was observed in very young Italian children rather than in Dutch newborns (Table 2).

Table 2 Age and sex standardized IRs per 100,000 PYs of cases of idiopathic acute liver injury distributed by countries
Full size table

3.3 Associations between Drugs and Idiopathic Acute Liver Injury

Out of the 785 definite cases of idiopathic ALI, there were 110 cases (14 %) where the index date occurred during the time window of the prescription. Such a proportion differed between the two countries (Italy: n = 95/731, 13 %; The Netherlands: n = 15/54, 28 %). Clarithromycin was associated with the highest risk of ALI compared to non-use (n = 9, adj. RR 25.9; 95 % CI 13.4–50.0; P < 0.001), followed by amoxicillin/clavulanic acid (n = 16, adj. RR 18.6; 95 % CI 11.3–30.6; P < 0.001), and amoxicillin (n = 6, adj. RR 7.5; 95 % CI 3.4–16.8; P < 0.001). Statistically significant increases in the rates of ALI were also found during current use of rifampicin/isoniazid combination, acetaminophen (paracetamol), rokitamycin, sulfamethoxazole/trimethoprim, phenobarbital, ketoprofen, carbamazepine and valproic acid, despite a very low number of exposed cases (n < 3; Table 3).

Table 3 Drugsa associated with acute liver injury among definite idiopathic cases
Full size table

4 Discussion

To our knowledge, this is the first population-based study that estimated the incidence and the features of the idiopathic acute liver injury in a paediatric outpatient population by combining three longitudinal EHR databases from Italy and The Netherlands.

Acute liver injury of which aetiology is not identified (i.e. idiopathic) or potentially attributed to the drug is relatively rare in children and adolescents, with approximately an annual incidence of 63/100,000 PYs. There are no paediatric studies to which we can directly compare our data because of different age-range setting and/or methodology. The most similar is a study conducted in the UK, which was restricted, however, to the population aged over 15 years. Using healthcare records as the data source, Duh et al. [18] found 66 cases per 100,000 PYs of liver injury defined as liver enzyme abnormalities of whom 41 potentially drug-attributable and 25 with uncertain aetiology. In both studies, the outcome was identified by similar searching and diagnostic criteria (e.g., through ICD-9-CM referring to hepatic disorders and by a set of international CIOMS criteria based on serum liver enzymes). On the other hand, our rate is substantially higher compared that previously estimated by de Abajo et al. among UK outpatients, namely 2.4/100,000 PYs [17]. However, this lower rate could be attributed to the differences in study population (patients aged 4–65 years) and study outcome (acute and clinically relevant drug-induced liver injury) [17]. In 2003, moreover, the prospective observational study, Drug-Induced Liver Injury Network (DILIN) was established to create a registry and biosample repository for clinical and mechanistic studies of DILI among patients >2 years old [5]. However, these data, as well as the most published studies concerning clinical features of DILI, as described by Bell and Chalasani [26], only reported the frequency of disease and, thus, the absolute number of cases rather than rates; moreover, they only included the most severe cases of liver injury which makes comparison difficult [1, 3, 4, 27–29]. In our study, we identified all cases of acute liver injury, even less severe, that might be theoretically idiopathic because of exclusion of all the competing well-known aetiologies. However, as differential diagnosis of idiopathic liver injury versus liver injury trigger by drugs can be extremely complicated, even through manual case-by-case evaluation, we estimated the incidence of acute and idiopathic liver injury, including even those potentially drug exposed.

The incidence of idiopathic ALI is much higher (roughly 4 times) in Italy than in The Netherlands. In Italy, most of the liver enzyme elevations were asymptomatic with a mild/moderate degree of severity, whereas in The Netherlands were reported in combination with other typical signs/symptoms of hepatotoxicity. This discrepant pattern may be explained by differences in healthcare systems and healthcare-seeking behavior of patients and parents. In Italy children are being inspected and monitored systematically during growth by the paediatrician, whereas GPs in The Netherlands only see children with complaints. Moreover, laboratory tests are requested more frequently in Italy than in The Netherlands, which also leads to a higher chance to detect asymptomatic cases of ALI. A similar phenomenon has already been described for Wilson disease. Compared to the other European countries, many Italian children have a Wilson disease’s diagnosis when they are completely asymptomatic and present only mild hypertransaminasemia [30]. Additionally, we cannot exclude that the discrepancy is influenced by different drug-prescribing patterns between the two countries. The prevalence of the use of well-known hepatotoxic drugs, such as anti-infective drugs and drugs for musculoskeletal disorders, is indeed much higher in Italy than in The Netherlands [12].

In line with recent evidence [5], the mean age of children and adolescents with idiopathic and potential drug-induced liver injury is 12 years in both countries. Moreover, the rates progressively increase with age, confirming our previous signal detection analysis on global spontaneous reports of liver injury in children [31]. As a matter of fact, susceptibility to drug toxicity changes across different age groups and may differ largely among newborns, toddlers, adolescents and adults, because of age-dependent metabolic activities of the hepatic cytochrome P450 (CYP) iso-enzymes [32]. In our study, the rates of ALI were slightly higher in the first year of life as compared to 2–5 years. This might be a reflection of varying activity of CYP in early life. On the other hand, a determinant for the slightly high rates in the early stage of life might be represented by a better observation of the parents, medical doctors and paediatricians to the neonates. This finding is supported by the highest observed rate in the youngest Italian children, among the total cases. At this age, indeed, any extension of the liver below the right costal margin has been noted as ‘hepatomegaly’ by Italian pediatricians, despite mostly physiological.

In an attempt to see whether longitudinal EHRs may be a useful source for drug safety signal detection concerning ALI in paediatric outpatients, we identified several drugs as the potential underlying cause. In our study, antibiotics are associated with the highest risk of ALI in children and adolescents. Amoxicillin (with or without clavulanic acid) has already been described in the literature as an hepatotoxic drug in paediatrics [5, 31, 33, 34], whereas we failed to find evidence of clarithromycin-induced liver injury in children. Nevertheless, paediatrics cases of liver injury have also been reported for other macrolides, such as erythromycin, azythromycin and roxithromycin [5, 33], suggesting that the hepatotoxicity might be an effect of therapeutic class. Noteworthy, the combination amoxicillin/clavulanic acid displays a higher risk of ALI than amoxicillin alone. Our finding is in line with the results from two previous population-based studies in adults on drug-induced liver injury, even if the risk estimated was lower [17, 35]. In addition, a better hepatic safety profile for amoxicillin alone compared to the combination with clavulanic acid was also reported in two studies on spontaneous reporting systems, one of them specifically addressing pediatrics [31, 36]. A recent study confirmed that susceptibility to amoxicillin/clavulanate-induced liver injury is influenced by genetic multiple variability [37]. We do believe that the hepatotoxicity related to the combination of amoxicillin and clavulanic acid in children is not rare as assumed until now and, as the number of cases is low, further investigations are needed to quantify this risk.

Noteworthy, we found only two events of acute liver injury related to acetaminophen, despite its well-known hepatotoxicity in adults as well as in children [33] and its wide use in paediatric patients [38]. Low exposure and number of cases regarding acetaminophen are due to the study data source that does not collect information about over-the-counter product and drugs not reimbursed by the healthcare system.

Overall, because all of these drugs were already known to be hepatotoxic in adults, this analysis could be considered a proof-of-concept that EHRs might be useful for estimation of incidence rates and for safety signal detection in paediatrics [12]. To gain sufficient statistical power to detect drug safety signals concerning a wide range of drugs exposure in paediatric population, combining data from multiple longitudinal healthcare databases was necessary. Other initiatives such as OMOP (Observational Medical Outcomes Partnership) and EU-ADR (Exploring and Understanding Adverse Drug Reactions) may also be explored for signal detection and validation in children and adolescents [6, 39].

The strength of this population-based study is the availability of detailed information on several clinical variables and drug use for a well-defined cohort of large size from two Countries. In particular, using electronic medical record databases allowed us to identify asymptomatic and less severe liver injury, which cannot be fully captured in prospective disease registries [5]. A very careful stepwise approach for case ascertainment has been performed to ensure the accuracy of the outcome data extraction by conducting an initial broad search of potential cases of acute liver injury and, subsequently, through manual inspection of all the medical records of potential cases by expert medical doctors.

Being observational in nature, our study is also vulnerable to confounding and bias. To reduce the effect of misclassification of the outcome, the main analyses have been performed among the cases confirmed by a specialist or laboratory data/ultrasound evidence. Nevertheless, because of lack of consistent diagnostic criteria for liver injury that are specific for children, residual misclassification cannot be completely excluded. In particular, serum liver enzyme activity is a marker of liver injury but no data are available on the sensitivity and specificity of different thresholds and the normal reference range for liver enzymes has not been clearly established in children [40–43]. Part of this potential misclassification has been tackled by excluding children with enzyme values less than twice the ULN. Finally, in the second part of our research, we identified drugs that were associated with ALI. As proper drug-event causality assessment cannot be performed based on the available information, our findings should be considered as purely hypothesis-generating.

5 Conclusions

The incidence rate of idiopathic ALI is relatively low in children and adolescents and comparable with adults. The differences observed between Italy and The Netherlands are likely due to the variability in the use of healthcare resources as well as drug prescribing pattern. Data mining of EHRs allowed us to identify antibiotics as the drugs with the highest and most frequent associations with ALI with apparently unknown aetiology. Combination of several healthcare databases is necessary to gain the statistical power to investigate the association of liver injury with a much larger range of drugs in the paediatric population.