Drug- induced liver injury at a tertiary care centre in Germany: Model for end- stage liver disease is the best predictor of outcome

Background & Aims: Agents most frequently inducing idiosyncratic drug- induced liver injury (DILI) differ between countries worldwide. Besides, there is no consistent data on the best model predicting mortality or the need for liver transplantation in DILI. We here analysed the DILI cohort of our centre with regard to causative drugs and clinical outcome. Methods: A retrospective analysis of 157 consecutive severe DILI patients presenting to our tertiary care centre in Hamburg, Germany, from 2008 to 2018, was performed. Results: The most frequent putatively causative drugs were phenprocoumon (n = 21), metamizole (n = 17) and flupirtine (n = 6). The mean values of ALT, bilirubin and Model for End- stage Liver Disease (MELD) score at the time of hospitalisation were 1201 U/L (SD: 1169 U/L), 6.8 mg/dL (SD: 7 mg/dL) and 17 (SD: 8). About 71% of all cases were treated with steroids or steroids combined with n - acetylcysteine. About 12.1% of all DILI cases had a poor outcome (liver transplantation and/or death). At the time of admission, MELD score performed better than Hy's law, the ratio (R) or the new ratio (nR) on their own or combined with bilirubin, regarding sensitivity or specificity for poor outcome.


| INTRODUC TI ON
Idiosyncratic drug-induced liver injury (DILI) is a heterogenic entity that can be induced by various drugs though the most frequent agents vary across the world. Amoxicillin-clavulanate is the most frequent agent responsible for idiosyncratic DILI in the western world. [1][2][3] In contrast, tuberculostatic drugs and herbal and dietary supplements (HDS) are the leading causes of DILI in Southeast Asia. 4,5 There are several reasons for the varying prevalences of agents causing DILI around the world, such as geographically different popularity or approval of single drugs or locally changing awareness for DILI-causing agents over time. As an example, the awareness for HDS as DILIcausing agents in the western world has increased only recently. 6 Severe DILI is rare, but one of the leading causes for acute liver failure (ALF). 7 Population-based studies revealed that the rate of death or liver transplantation (LTx) due to idiosyncratic DILI ranges between 1% and 6%. 2,5,8 In non-population-based studies, it reaches a rate between 7% and 15%. 1,3,[9][10][11] Therefore, markers for the prognosis of DILI progressing to ALF are urgently needed. The Model for End-stage Liver Disease (MELD) score seems to be suitable for the prognosis of poor outcome (defined as LTx and/or death) in DILI patients, with high sensitivity and good specificity. [11][12][13] Other prognostic scores with a comparable sensitivity and specificity are Hy's law, and the ratio (R) ≥5 and the new (n)R ≥ 5, each combined with total bilirubin (TBL) >2 × the upper limit of normal (ULN). R is defined as the ratio between alanine aminotransferase (ALT) and alkaline phosphatase (ALP), each related to their ULN, and nR uses the higher value of either ALT or aspartate aminotransferase (AST). 11,14,15 We herein characterised a cohort of idiosyncratic severe DILI cases at our tertiary care centre in Hamburg, Germany. Our study shows striking differences regarding the most frequent putatively causative drugs of DILI in comparison with previous cohorts from other regions of the world. Moreover, MELD score performed best for the prediction of poor outcome of DILI in our cohort.

| ME THODS
Consecutive idiosyncratic, non-acetaminophen-induced DILI cases presenting to the I. Department of Medicine, University Medical Centre Hamburg-Eppendorf, Hamburg, Germany from January 2008 till July 2018 were analysed in this retrospective study. A case series of patients with metamizole-induced liver injury, being also included into this study, has been published elsewhere. 16 Patients presented initially to our centre with symptoms of acute liver injury or were referred to our centre from another healthcare provider. For the latter cases, initial laboratory parameters of the referring centres or physicians were analysed. Diagnosis of DILI was based on the Roussel Uclaf Causality Assessment Method (RUCAM). 17,18 When more than one drug was suspected to be responsible for liver injury, the RUCAM score was applied for each drug. Our inclusion criteria for DILI cases were the following: ALT ≥5 × ULN or ALP ≥2 × ULN or ALT ≥3 × ULN and total bilirubin ≥2 × ULN. 19,20 A minimal follow-up time of 30 days was also needed for inclusion in the study, except for the cases progressing to liver transplantation or death. The calculation of the biochemical pattern of liver damage was based on the ratio (R) between ALT and ALP, each related to their respective ULN. The biochemical pattern was defined as hepatocellular when R > 5, as cholestatic when R < 2 and as mixed when R was equal to or between 2 and 5. 21 For the calculation of R, the first available blood test at the time of initial presentation was used. For the new ratio (nR), we used the higher value of either ALT or AST. 14 Five points in time were used for the analysis of blood values: the first presentation to any provider within the healthcare system, when the patient presented to a hospital for the first time and the peak values for ALT, ALP and total bilirubin. Hepatic encephalopathy (HE) was graded according to the West Haven Criteria. 22 For the quantification of liver injury in DILI cases, the DILI Severity Index by Aithal et al 19  • The most frequent putatively causative drugs were phenprocoumon and metamizole.
• About 12% of all DILI patients had a poor outcome and required liver transplantation and/or died.
• MELD score ≥ 18 at the time of admission performed best in this cohort for the prediction of poor outcome in DILI.

Lay Summary
The clinical course of patients suffering from liver injuries caused by different drugs was analysed at a (!) large medical centre in Germany. The MELD score, a score based on laboratory values, calculated at the time of admission, was the best predictor for the outcome of these patients and could predict whether acute liver failure and the need for liver transplantation impends.
MELD score was computed for each of the five points in time. Cases of phenprocoumon-induced DILI (n = 25) were excluded from MELD Autoimmune hepatitis (AIH) was excluded by clinical follow-up assuring that liver enzymes did not rise again after weaning of steroid treatment. Autoantibodies such as antinuclear antibodies (ANA), anti-smooth muscles antigen antibodies (anti-SMA), anti-liver kidney microsomal antibodies (anti-LKM) and anti-mitochondrial antibodies (AMA) were detected by immunofluorescence on rat liver, kidney and stomach tissue and on HEp2 cells. Anti-soluble liver antigen/liverpancreas antigen antibodies (anti-SLA/LP) were analysed by ELISA.
Mini-laparoscopically guided liver biopsy was performed in 71.3% of patients (n = 112) of the total cohort at our centre. IBM Corp.) was used to analyse the data. For the statistical evaluation, categorial variables were presented by frequency and tested with the χ 2 -test. Accordingly, for the statistical evaluation of continuous variables, both the mean with standard deviation (SD) or the median with interquartile range (IQR) and the Kruskal-Wallis or Mann-Whitney U test were used. A logistic regression was performed for continuous and categorial variables. A receiver-operating characteristics (ROC) curve was used to identify the concordance statistic (c-statistic) of the MELD score. Statistically significant differences were defined when the P-value was less than .05. This study was approved by the local ethics committee (WF-064/12).

| Demographic and clinical characteristics
For inclusion in the study, 177 idiosyncratic DILI cases between January 2008 and July 2018 were considered. Twenty subjects were excluded as two of them did not fulfil the biochemical definition of DILI. Eighteen cases were excluded due to a short follow-up time.
Overall, the total cohort comprised 157 DILI cases, in which 88 different drugs or HDS were identified. In 128 cases (82%), a single drug was putatively causative for DILI, and in the other 29 cases, (18%) two or more drugs had a 'possible' or better RUCAM score.
Of the total cohort, 12 cases (8.6%) were classified as 'possible', 78 cases (56.1%) as 'probable' and 49 cases (35.3%) as 'highly probable' according to the RUCAM score. In 18 cases, it was not possible to calculate the RUCAM score because the point in time of the first drug intake within the last six months before liver injury could not be exactly determined retrospectively. One hundred and fifty-three cases (98%) of the cohort were hospitalised for the diagnostic evaluation of acute icteric hepatitis, and the median time of hospital stay was 11 days (IQR: 6-19 days, total range: 1-52 days). As shown in Table 1, 101 subjects of the total cohort (64%) were female, and the mean age was 53 years (SD: 17 years, total range: 19-79 years). One hundred and twenty-two subjects (77.7%) had a hepatocellular, 18 (11.5%) a mixed and 17 (10.8%) a cholestatic biochemical pattern.
Significant differences between male and female DILI patients with regard to outcome could not be detected (Supplementary Table 1).
Male DILI patients tended to receive antimicrobials (n = 14, 25%) more often than female patients (n = 14, 14%; P = .081, data not shown). A significant difference in the latency of first drug intake until first presentation to the health care system was detected among the three groups of different biochemical patterns: DILI patients with a hepatocellular pattern showed the longest latency Other autoantibodies such as AMA (detected in seven cases, 5%) or anti-LKM (detected in one case, 0.7%) were detected infrequently.
Sixteen patients (10%) were accidentally re-challenged by the DILI-causing drug. Metamizole was the most frequent drug involved in cases of re-challenge (Supplementary Table 3). Latency of first presentation to ALT peak value, days, median (IQR)

| Prevalence of drug groups and individual drugs
The most frequent drug groups causing DILI with a single implicated drug (n = 128) were anti-infectious drugs with 32 cases (25%),
Another six cases with ALF (4%) were not transplanted due to different reasons and died from liver failure or liver-related complications (Supplementary Table 7   In total, 46 cases (35% of the entire cohort, DILI cases due to phenprocoumon excluded) had a MELD score ≥18 at the time of hospitalisation, including 14 cases (82%) of the poor outcome group and 32 cases (27%) of the good outcome group (Table 6). These two groups were further analysed in the following. The presence of HE of any grade was significantly more frequent in the poor outcome group (n = 14, 100%) than in the good outcome group (n = 0, 0%, P < .001).

| Prediction of outcome
A combination of the MELD score ≥18 and the presence of HE had a sensitivity of 88% and a specificity of 100% for poor outcome. An extent of confluent necrosis between 70% and 100% on liver histology was significantly more frequent among patients with poor outcome (23%) than in patients with good outcome (0%, P = .006, Abbreviations: ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; HE, hepatic encephalopathy; LDH, lactate dehydrogenase; MELD, Model for End-stage Liver Disease score; nR, new ratio; OR, odds ratio; R, ratio; TBL, total bilirubin; ULN, upper limit of normal. a Excluding INR, creatinine and TBL for multivariable analysis with backwards elimination.
b Excluding ALT for multivariable analysis with backwards elimination, additionally.
c Excluding ALT and ALP for multivariable analysis with backwards elimination, additionally.
d Including 25 patients with DILI after the intake of phenprocoumon.
Significant results with P values <.0.05 are highlighted with bold font.

| D ISCUSS I ON
Idiosyncratic DILI can be induced by diverse drugs, whereas the most frequent drugs causing DILI vary between countries. Phenprocoumon was the most frequently implicated drug in DILI cases at our tertiary centre in Hamburg, Germany. Other reports from Germany support that phenprocoumon frequently causes DILI. 26 The two other most frequent agents causing DILI at our centre were metamizole and flupirtine. We have recently published a case series of metamizoleinduced liver injuries elsewhere. 16 Flupirtine is a non-opioid analgetic drug and was used to treat neuronal hyperexcitability. 27 It was withdrawn from the European market in 2018 due to hepatoxicity, but it is still available in other countries. 28 In contrast to our study, lies in varying local approval and popularity, with phenprocoumon, metamizole and flupirtine being very popular on the German market in the study period. The two vitamin K antagonists on the German market are phenprocoumon and warfarin, whereas phenprocoumon is more popular than warfarin in Germany. In comparison to phenprocoumon, warfarin is associated less, or even not at all, with liver injury. [29][30][31] Prediction of clinical outcome for DILI is important since it represents one of the leading causes of ALF. In our study, the rate of poor outcome (LTx and/or death) was 12%. This corresponds well with the previous data from non-population-based studies (7%-15%). 1,3,[9][10][11] Several clinical, biochemical and histological predictors for poor outcome of DILI have been proposed, such as HE, histologically proven confluent necrosis, levels of AST, INR, serum creatinine, albumin, MELD score, fibrosis or steatosis on liver histology. [11][12][13][14][32][33][34] In previous studies, Hy's law showed a sensitivity of 82%-90% and a specificity of 44%-58% for poor outcome in DILI. 4,14 Our study is in line with these previous results (sensitivity of 95% and specificity of 57%). The newer prognostic scores for DILI R > 5 or nR >5 and TBL >2 × ULN had a lower sensitivity (both 74%) in our study than in the original publication by Robles-Diaz et al 14 (83% and 90%). In our study, but also in previous studies, MELD score showed the highest sensitivity and specificity for poor outcome in DILI (MELD score ≥19-20.5: sensitivity 89%-93%, specificity 74%-84%, c-statistic = 0.83-0.93). 11,12,15,32 In our study, the MELD score with a cut-off value of 18 at the time of hospitalisation had a c-statistic of 0.85 and a sensitivity and specificity of 88% and 72%, respectively (  12 In a study on acetaminophen-induced liver injury, changes of MELD score over time were analysed with regard to outcome. 35 The delta MELD score from day 0 to day 1 after the onset of HE had a c-statistic of 0.78 for poor outcome. In our study, changes of MELD score between different time points did not perform better than MELD at the time of hospitalisation. Only the combination of MELD score ≥18 at the time of hospitalisation and the development of HE had a higher sensitivity and specificity for poor outcome than MELD alone (88% and 100%, respectively) with a high positive and negative predictive value of 1.0 and 0.98 (Table 5). However, HE as a predictor for poor outcome implicates a delay of management due to the latency until its clinical presentation.
Most of our patients received liver biopsy. We did not find an association between liver fibrosis or steatosis with poor outcome (Table 1). However, a positive association between poor outcome and prevalence and severity of confluent necrosis on liver biopsy was detected in our study. In DILI cases with poor outcome, a high extent of necrosis (70%-100% of liver tissue) was significantly more frequent than in patients with good outcome (38% in the poor outcome group vs 1% in the good outcome group, P < .001).
Previous studies have reported on a subtype of DILI termed 'autoimmune-like DILI' which is characterised by a dominating immune cell infiltration of the liver and positivity for autoantibodies. 36 TA B L E 6 (Continued) after a very long period of spontaneous remission, we cannot totally exclude that some AIH cases were falsely defined as DILI patients in our study.
Steroids as a treatment of DILI to improve outcome is controversial, and only very few studies have partly addressed this issue. 36,[44][45][46][47] Therefore, no substantiated recommendation for the use of steroids is given by current clinical practice guidelines for DILI. 20,48 One of the reasons to withhold steroids from DILI patients is the potential risk of serious infections. 49 However, a significant increase of serious infections under steroid therapy has not been detected in the setting of ALF in previous studies. 47,50 As a member of the European Reference Network on Hepatological Diseases (ERN RARE-LIVER), our centre gets in contact with many suspected AIH cases. Acute presentation of AIH and DILI cannot be differentiated in most cases, even after obtaining liver histology. Therefore, we treat most cases of suspected AIH or DILI with prednisolone and differentiate between both entities by the clinical course: In AIH, liver enzymes will increase again after weaning of steroids, whereas, in DILI, they will not. Since most patients in our DILI cohort (70%) were treated with steroids (Table 1), we could not analyse thoroughly whether steroids had a positive effect on the outcome of DILI. The rate of infections was not significantly different between DILI cases treated with steroids (6.3%) and those who were not (4.4%, P = .66). However, it seems that patients under steroids who developed infectious complications were at higher risk of a poor outcome, although patient numbers in our study are very small (Supplementary Table 8). Future prospective studies are needed to address the risks and benefits of steroids for DILI patients.
The strength of our study is the size of the DILI cohort at a single tertiary care centre. Our study complements previous international studies on DILI, reflecting the special situation in Germany with a high rate of patients treated with phenprocoumon, metamizole and, until it was banished from the European market, flupirtine.
Furthermore, the percentage of performed liver biopsies in DILI patients is relatively high at our centre. The histological analysis supported the prognostic value of confluent necrosis on liver histology.
However, current guidelines are reluctant with the indication of liver biopsy for DILI patients. 20,48 Future studies on DILI need to address to what extent histological characteristics such as necrosis can improve prognostic scores based on just laboratory parameters. Our study has several limitations, which are mainly due to the retrospective, single-centre, non-population-based study design.
In summary, our study underlines the need to identify the frequency of DILI-inducing drugs in each country separately. At our centre in Germany, phenprocoumon and metamizole were by far the most frequent causative drugs. Of note, these drugs could be replaced with lower risk drugs. As a prognostic score, MELD score ≥18 at the time of hospitalisation performed best in our study to predict outcome in DILI patients. It seems advisable to refer these patients to a transplant centre immediately and to proceed to liver transplantation if HE is noticed.