Health care delivery, economics and global health ORIGINAL ARTICLE Risk of drug-induced liver injury with the new oral anticoagulants: systematic review and meta-analysis Daniel Caldeira,1,2 Márcio Barra,1,2 Ana Teresa Santos,1,2 Daisy de Abreu,1,2 Fausto J Pinto,3 Joaquim J Ferreira,1,2 João Costa1,2,4,5 ▸ Additional material is ABSTRACT only emerged with postmarketing experience published online only. To view Objective In recent years, safety alerts have been because hepatic adverse drug reactions due to car- please visit the journal online (http://dx.doi.org/10.1136/ made warning of the risk of serious drug-induced liver diovascular drugs are relatively uncommon, but heartjnl-2013-305288). injury (DILI) caused by cardiovascular drugs. The new potentially serious, and premarketing clinical trials 1 oral anticoagulants (NOACs) have now reached the are underpowered to detect differences between Clinical Pharmacology Unit, fi Instituto de Medicina market. However, safety concerns have been raised treatment arms. These recent high pro le cases of Molecular, Lisbon, Portugal about their hepatic safety. Therefore we aimed to serious liver adverse reactions associated with car- 2Laboratory of Clinical evaluate NOAC liver-related safety. diovascular drugs have amplified the need for Pharmacology and Methods Systematic review and meta-analysis of phase careful premarketing analysis of DILI risk asso- Therapeutics, Faculty of Medicine, University of Lisbon, III randomised controlled trials (RCTs). Medline and ciated safety. Lisboa, Portugal CENTRAL were searched to September 2013. Reviews In the last 5 years, new oral anticoagulants (NOACs), 3Cardiology Department, and reference lists were also searched. Two reviewers with direct inhibition of factors IIa or Xa, have CCUL, CAML, Faculty of independently searched for studies and retrieved data granted European and US marketing authorisation Medicine, Lisbon, Portugal estimates. Primary outcome was DILI (transaminases for the prevention of thrombotic events in high-risk 4Evidence Based Medicine Centre, Faculty of Medicine, elevations >3× upper limit of normal (ULN) with total adult patients. The past history of ximelagatran University of Lisbon, Portugal bilirubin >2× ULN). NOACs were compared against any further contributed to a close surveillance and 5Portuguese Collaborating control group. Random-effects meta-analysis was reporting of hepatic adverse events during NOAC Centre of the Cochrane performed, and pooled estimates were expressed as clinical trials. In fact, with the exception of dabiga- Iberoamerican Network, Faculty relative risk (RR) and 95% CI heterogeneity was tran, NOACs are metabolised by the liver (CYP3A4 of Medicine, University of 2 Lisbon, Portugal evaluated with I test. involvement) and, according to the public assess- Results Twenty-nine RCTs evaluating 152 116 patients ment reports of these drugs they all are associated Correspondence to (mean follow-up of 16 months) were included. All RCTs with increases in transaminases and abnormal liver Dr Daniel Caldeira, Laboratório were rated as having low risk of bias. NOAC were not function, with an incidence up to 1 in 100 to 1 in de Farmacologia Clínica e 6–8 Terapêutica, Faculdade de associated with an increased risk of DILI (RR 0.90, 95% 1000 people. Furthermore, the 2013 European 2 Medicina da Universidade de CI 0.72 to 1.13, I =0%). Similar results were obtained guidance for the use of NOAC recommends yearly Lisboa, Av Prof Egas Moniz, for individual NOAC (rivaroxaban, apixaban, dabigatran, monitoring of liver function.9 Lisboa 1649-028, Portugal; darexaban, edoxaban) and considering the different In this context, we aimed to better estimate the [email protected] control groups (vitamin K antagonists, low molecular risk of hepatic adverse drug reactions associated Received 20 November 2013 weight heparin (LMWH) and placebo). The risk of with NOAC by performing a systematic review and Revised 7 January 2014 transaminases elevations (>3×ULN) was lower among meta-analysis of phase III randomised controlled Accepted 8 January 2014 NOAC-treated patients, in particular in comparison with trials (RCTs). Published Online First LMWH-treated patients (RR 0.71, 95% CI 0.59 to 0.85; 29 January 2014 I2=27%) METHODS Conclusions NOACs are not associated with an Guidelines increased risk of DILI. The unexpected ‘protective’ effect This systematic review followed the Preferred of NOAC is probably due to LMWH-associated Reporting Items for Systematic Reviews and hepatotoxicity. Meta-analyses framework guidelines.10 Studies’ eligibility criteria INTRODUCTION Phase III RCTs comparing NOACs, including direct Mostdrugsaremetabolisedintheliver.1 Drug-induced inhibitors of IIa (dabigatran) or Xa (apixaban, dar- liver injury (DILI) includes a broad clinical and exaban, edoxaban, or rivaroxaban), against any pathological spectrum of hepatotoxicity and many control group (placebo, no-treatment or standard genetical and non-genetical patient characteristics care, non-pharmacological interventions or any have been proposed as risk factors for DILI from drug). Only phase III RCTs were considered to medications.23In recent years, safety alerts have avoid bias in risk estimation due to statistical effects been made warning for the risk of DILI, including of rare events and the impact of small size under- – life-threatening liver failure, caused by cardiovascu- powered studies on meta-analysis results.11 14 lar drugs. For example, dronedarone, an antiar- Furthermore, we were interested in determining the rhythmic drug, can cause serious liver injury,4 and risk associated with the approved and commonly To cite: Caldeira D, ximelagatran, an oral direct thrombin (IIa) inhibi- used doses of the NOAC. All RCTs were considered Barra M, Santos AT, et al. tor, has been withdrawn from the market in 2004 for inclusion irrespective of patients’ disease, – Heart 2014;100:550 556. due to the risk of DILI.5 These safety warnings comorbidities, background therapy, NOAC treatment 550 Caldeira D, et al. Heart 2014;100:550–556. doi:10.1136/heartjnl-2013-305288 Health care delivery, economics and global health duration or follow-up. Only trials reporting hepatic data as a Review Manager V.5.2.6 (The Nordic Cochrane Centre, The prespecified outcome were included to avoid selective reporting. Cochrane Collaboration, 2012) was used to obtain the estimates Trials had to provide laboratory data for transaminases and of individual studies, pooled analysis and to retrieve the forest bilirubin. plots. Heterogeneity was assessed with the I2 test, which mea- Primary outcome was DILI, defined as increases in serum sures the percentage of total variation attributed to interstudy levels of transaminases above three times the upper limit of heterogeneity rather than random.18 The inverse of variance normal (ULN) and total bilirubin above two times the ULN. method with random effects model was used by default inde- According to Hy’s law, the outcome defined above is the most pendently of the existence (I2≥50%) or not of substantial het- specific predictor of potential severe hepatotoxicity.15 Secondary erogeneity between studies’ results. In case the event rates were outcomes were incidence of transaminases elevation >3× ULN, <1% in overall NOAC and control groups, we determined the and incidence of bilirubin elevation >2× ULN. OR of primary outcome through Peto’s method, because under these circumstances of relatively rare events the Peto’s ORs are a 19 Search method less biased measure. Investigators retrieved potential eligible studies through an elec- For outcome analysis, in case the study provided data for tronic search in Medline and Cochrane Library, run in both transaminases values, we considered for statistical analysis fi September 2013. Search strategy for Medline (see supplemen- the results of alanine transaminase due to its higher liver speci - 20 tary online) included free text and Medical Subjects Headings city in comparison with aspartate transaminase. Outcome data fi fi terms without language restrictions. Additionally, we checked was analysed according to prespeci ed subgroups de ned by the the references of systematic reviews and meta-analyses that eval- individual NOAC and type of control group. Differences uated NOAC, as well as the reference list of each included study. between subgroups were assessed based on random effects 21 When data for pretended outcomes were not available from model due to the lower risk of false-positive results. published articles, we looked at the available public reports of Publication bias was assessed through visual inspection of ’ ’ these drugs at the European Medicine Agency and Food and funnel plot asymmetry and with Egger s and Peters regression 22 23 Drug Administration. tests. Data extraction, evaluation and synthesis Titles and abstract of obtained records were screened independ- RESULTS ently by two authors. Doubts and disagreements were solved by Results of the search and description of the studies consensus. Selected studies were assessed in full text in order to Figure 1 shows the flow chart of studies’ selection and the determine the appropriateness for inclusion in the systematic reasons for studies’ exclusion. We were able to include 29 review. Study characteristics and results were extracted inde- studies enrolling 152 116 patients, 83 513 of them treated with – pendently into a standardised form. NOACs.24 50 The NOACs evaluated were apixaban (8 RTCs; – – Appraisal of methodological bias was done according to