Progress in our understanding of severe drug-induced liver injury

Abstract

Despite its low incidence, drug-induced liver injury (DILI) continues to have a high impact on the medical community, pharmaceutical industry, and general public. DILI is often difficult to diagnose, and few specific treatments are available. Severe DILI is the primary reason for regulatory actions on drugs, including failure to approve, “Dear Doctor” letters, and removal from the market. Such actions often deprive those who would otherwise benefit from these medications and increase drug development costs. At the individual level, drug-induced acute liver failure (DIALF) is often a devastating, sudden, and completely unexpected event for patients, loved ones, and physicians. The low incidence, lack of definitive diagnostic criteria for DILI, and varied array of medications that can lead to DIALF make clinical research difficult. DIALF, drug-induced acute liver failure; DILI, drug-induced liver injury; UNOS, United Network for Organ Sharing. The study by Mindikoglu et al.1 in this issue of Liver Transplantation expands on a previous examination of United Network for Organ Sharing (UNOS) data regarding causative agents for DILI resulting in liver transplantation.2 The list of culprits and their distribution recorded by UNOS are similar to those reported by the Acute Liver Failure Study Group,3 which includes DIALF patients who may or may not require liver transplantation, and by the Drug Induced Liver Injury Network, which enrolls patients with moderate and severe liver injury.4 In aggregate, these provide a clearer picture of the more prevalent agents leading to severe DILI in the United States. DILI has been clearly established as the leading cause of acute liver failure in the United States, with acetaminophen leading the list of implicated agents.3 Even without acetaminophen, DILI remains the leading cause of acute liver failure among cases for which an etiology can be determined. Moreover, a proportion of indeterminate cases may be due to occult acetaminophen.5 Other implicated drugs have been identified through UNOS data, with antibiotics (including isonia) and antiepileptics being most commonly. The current study examined appropriate donor and recipient covariates as predictors of posttransplant survival. Besides life support and elevated creatinine, the authors make the novel observation that DIALF due to antiepileptic drugs is a strong independent risk for poor survival in children but not in adult patients. In comparison with acetaminophen, the hazard rate for death in this pediatric group was over 4.1 (95% confidence interval: 2.1-8.1). The reasons underlying the poorer prognosis in these children are not revealed by this study. The authors point out that these patients on average had a lower frequency of listing as status 1 and a longer wait list time prior to transplantation. The drug implicated in almost three-quarters of the pediatric antiepileptic cases (73%) was valproate, whereas this drug was implicated in only 17% of the adult antiepileptic cases. DILI due to valproate typically presents in children as hyperammonemic coma with only modest elevations in serum transaminases in comparison with typical DIALF cases. It is possible that this unusual presentation leads to an underappreciation of the seriousness of the liver injury and partially accounts for the longer wait for transplantation. Another possibility that might account for the poor survival is that some children in the antiepileptic group might have had underlying metabolic disorders that both predisposed them to seizures and adversely affected their long-term prognosis. For example, Alpers-Huttenlocher syndrome is an inherited mitochondrial disease that results in mental retardation, liver disease, and seizures in childhood. Interestingly, it is believed that these children may be particularly susceptible to liver injury from valproate.6 The current study is the first to propose a mathematical model to predict survival after transplantation. With a stepwise Cox regression, creatinine, life support, and antiepileptic use in pediatric patients were the only variables incorporated into the model. The predicted survivals were matched to the actual survivals in those with low, medium, and high risk scores. The curves matched very well. However, without a true validation set, one might expect the curves to be close. The authors make an admirable attempt to get around this problem by using a rotating 75% of cohort training set and rotating 25% validation set. However, as they point out in their discussion, testing the model on a truly separate data set would be desirable. Moreover, the clinical utility of this particular model, if validated, may be limited. Even in their highest risk group (the upper third), the predicted and actual survivals post-transplant are still over 50% at 5 years. Presumably, many of the patients with highest risk scores would be children on antiepileptics because this category carries significant weight in their equation. It is unlikely that a decision to not transplant would be made on the basis of a high score when the predicted survival is still in an acceptable range. No data were given on whether the model could discern those with <50% 5-year survival (eg, the upper 10%-20% of scores). Lastly, it would have been informative to include all DIALF patients listed as status 1 and a transplant-free survival analysis in an attempt to predict those that may survive without a transplant. Overall, the authors must be commended for adding valuable information regarding DIALF. Their attempt at modeling falls short of immediate usefulness, but the identification of poorer outcome for children with antiepileptic DIALF is intriguing and points out the need for more focused research on DILI in pediatric populations. DILI, though rare, has wide implications for all of us who take and prescribe medications. In the coming years, we anticipate great advances in our understanding of DILI and DIALF. Large registries such as the UNOS database provide valuable population-based data to form hypotheses, but they lack all desired phenotypic information about the patients. Multicenter studies such as the Acute Liver Failure Study Group and Drug Induced Liver Injury Network can provide detailed individual patient data, sera, and genomic DNA, which can be used to investigate these new hypotheses. The combined efforts will hopefully shed better light on preventive factors, including genetic predisposition, and move us beyond simply reporting cases and case series.