Diagnostic and Prognostic Assessment of Suspected Drug-induced Liver Injury in Clinical Practice

Raúl J. Andrade; Mercedes Robles-Díaz


Liver International. 2020;40(1):6-17. 

In This Article

Different Prognostic Scores Applied to Individual Cases of Suspected DILI

Predicting Serious DILI Outcome

Prediction of severe outcome at DILI recognition remains a challenge in clinical practice; however, it is crucial for improving patient management. Early recognition of cases with potential to develop ALF can help the clinician to identify those patients who need closer observation, hospitalization or transferal to a liver transplant centre. ALF is a sudden deterioration in liver function in which the patient develops encephalopathy, jaundice and coagulopathy in the absence of underlying chronic liver disease.[61] These features are delayed, severe and irreversible in most patients,[16] with a high mortality rate (from 60% to 90%) without liver transplantation (LT). DILI is responsible of over 50% of ALF in the USA,[62] UK[63] and Sweden[64] and is the main indication for liver transplantation in ALF cases.[65]

Liver biopsy findings can help to predict a worse prognosis. In a review of biopsies performed on DILI patients in the Swedish Adverse Drug Reactions Advisory Committee (SADRAC) retrospective database, presence of necrosis was predictive of lower rate of survival, while eosinophilia was associated with more favourable DILI outcomes.[66] Likewise, liver failure and death were associated with higher degrees of necrosis, fibrosis stage, microvesicular steatosis and ductular reaction, whereas eosinophils and granulomas predominated in those with milder DILI outcome in the DILIN cohort prospective study including 249 liver biopsies.[47] Similarly, evidence of bile duct loss in patients with acute DILI (generally presenting with cholestatic pattern) indicates the development of vanishing bile duct syndrome with progressive cholestasis leading to liver failure requiring LT or death.[48]

The Food and Drug Administration endorsed many years the denominated 'Hy's law' for detecting serious liver signals during drug development. The Hy's law (Table 2) is based on the observation of Hyman Zimmerman[67] that hepatocellular DILI with jaundice, ruling out other potential aetiologies, denotes a severe reaction with a 10%-50% mortality rate from liver failure (before LT were performed).[68]

The Hy's law has been further validated in the post-marketing setting in three large DILI populations from the Spanish DILI Registry,[7] SADRAC retrospective database[69] and DILIN[8] showing 11.7%, 9.2% and 15% mortality/liver transplantation, respectively, in DILI patients with hepatocellular pattern of liver damage and jaundice.

A number of international consortia that prospectively recruit bona fide DILI cases have evaluated the performance of laboratory and clinical variables for early ALF prediction in DILI patients. The Spanish DILI Registry analysing a cohort of 771 patients proposed a new Hy's Law (Table 2). It demonstrated more accuracy compared with traditional Hy's law (similar sensitivity (90% vs 93%) and higher specificity (63% vs 43%) and also higher area under the receiver operating characteristic (AUROC) (0.77 vs 0.67)). In this study, an independent new prognostic algorithm for ALF in DILI was also developed (Table 2), which demonstrated a good balance between sensitivity and specificity with validation in an independent cohort (specificity, 82%; sensitivity, 80%; AUROC 0.80).[9]

Using a retrospective cohort of 15,353 DILI patients from the Kaiser Permanente database in California, Lo Re et al[70] developed a model including platelet count and total bilirubin (Table 2). The model had the highest discrimination (C statistic, 0.87) with decreasing platelet count and increasing total bilirubin as strong predictors of ALF. This model had a sensitivity of 0.91 and specificity of 0.76 for ALF. In an external validation of the model, high sensitivity was maintained for ALF (0.89). In the initial cohort, traditional Hy's Law criteria showed high specificity (0.92), but low sensitivity (0.68), while a Model for End-stage Liver Disease (MELD) (Table 2) Score of ≥10 showed higher sensitivity (0.84) for ALF.[70]

The performance of MELD score for early prediction of mortality in DILI patients has also been explored in several studies. Thus, in a retrospective study of DILI patients who visited emergency departments in Seoul, Korea from 2010 to 2012, a logistic regression identified MELD [Odds ratio (OR)] 1.21 and haemoglobin (OR 0.77) as independent predictors of poor outcomes.[71] Rathi et al evaluated MELD score in a prospectively collected Indian DILI cohort of 82 patients with 8 liver-related deaths. In a multivariate logistic regression analysis, jaundice, encephalopathy, MELD score and alkaline phosphatase at 1 week independently predicted mortality.[12]

The Drug-Induced Liver Injury Network evaluated MELD score as well as Hy's law and nHy's law as ALF predictor in 1089 DILI patients, including 107 death/LT, in 68 of which DILI had a primary role. In this multivariate analysis, Hy's law was significantly associated with poor outcome owing to DILI (hazard ratio (HR), 2.2), nR Hy's law showed a stronger association (HR, 6.2), and the predictive capacity of MELD was still stronger (1.2 per MELD point). The C statistic for a MELD cut-off of 19 was 0.83 compared to 0.73 for nR Hy's law and 0.60 for Hy's law. In this study, leukocytosis, coagulopathy, higher bilirubin and thrombocytopenia were independently associated with DILI mortality.[10] It is important to keep in mind that mortality in DILI patients can have other causes than ALF. In 22 of the 107 patients who died up to 2 years from the onset of liver damage in this study, DILI did not play a role in mortality.

Finally, in an Indian cohort of 905 DILI patients of whom 128 (14%) developed ALF, only total protein and INR were independent predictors of mortality.[72] Performance of MELD score, King's college criteria (KCC)[11] score and Acute Liver Failure Study Group (ALFSG) index[73] (described in Table 2) with regard to mortality was compared with receiver operating characteristic (ROC) curve. ROC curve for MELD and ALFSG index was 0.76, but only 0.51 for KCC. The sensitivity and specificity for MELD scores were 72% and 74%, respectively, and 41% and 51%, respectively, for KCC. The best cut-off value for MELD score was ≥28.5.[72]

The comparison of all these studies is difficult because of differences in populations included, definition of DILI, follow-up and the statistical methods used. In addition, the ROC and other quantitative measures determined in each of the studies have been ascertained in a variety of different ways such that cross-study comparisons of their performance characteristics are highly limited. It is worth noting that some of these scores highlight the presence of findings with metrics that signify advancing liver injury while this is not the case for Hy's law. Hence, the reliability of these scores is different and they cannot be strictly compared. Furthermore, Hy's law, although primarily useful for predicting liability of drugs in clinical drug development, is also being used in the post-marketing setting (Registry studies) for prediction of severe outcome. To overcome some of these limitations, the scores discussed in this review should be prospectively validated in similar cohorts of DILI patients with common diagnostic criteria to obtain comparable and more reliable results.

Predicting Chronicity

The ability to predict those DILI patients in whom the injury will peb rpetuate and become chronic is an important task for clinicians. The first difficulty, however, is to reach a consensus on what chronic DILI is. A former international consensus meeting defined chronic DILI as perpetuating liver damage after 3 months of drug withdrawal,[74] whereas the DILIN group advocated for 6 months,[8,75] and a consensus of DILI experts in 2011 recommended persistence of liver test abnormalities more than 3 and 6 months for hepatocellular and cholestatic/mixed liver damage, respectively.[16] Because these definitions were not evidence based, the Spanish DILI Registry undertook a prospective follow-up study with rigorous exclusion criteria to avoid confounding factors in 298 DILI patients. This study demonstrated that the best cut-off point for defining chronicity was the persistence of liver tests alterations 1 year after drug discontinuation independent of the type of liver injury (hepatocellular, cholestatic or mixed).[13] Reported risk factors for chronic DILI have been older age,[13] females,[13,17,75] diabetes,[13,75,76] dyslipidaemia, hypertension and use of statins.[13] Because the elderly have a higher prevalence of metabolic syndrome, it is possible that this can account for an increased risk of chronicity rather than the older age itself. Similarly, the use of statins as a risk factor for chronic DILI could be a reflection of the presence of dyslipidaemia in this group of patients or the result of an immune response triggered by these drugs as they have been associated with drug-induced AIH.[77]

Other risk factors for chronic DILI are jaundice at presentation[13] and need for hospitalization.[13,75] These findings probably reflect that more severe liver damage need longer time to resolve. Increased ALP values at DILI onset has also been identified as a risk factor for chronicity.[13,76] Furthermore, Medina-Cáliz et al developed a prognostic model for chronicity based on values of ALP and TBL, showing that persistently elevated TBL (>2.8 × ULN) and ALP (>1.1 × ULN) in the second month from DILI onset predict increased risk of chronic DILI.[13]