Emerging Tools to Support DILI Assessment in Clinical Trials with Abnormal Baseline Serum Liver Tests or Pre-existing Liver Diseases

Abstract

Based on the late Dr. Hyman Zimmerman's observation that hepatocellular drug-induced liver injury (DILI) leading to jaundice carries a ≥ 10% fatality risk (coined as Hy's law by others), evaluation of Drug-Induced Serious Hepatotoxicity (eDISH) continues to play a central role in the assessment of a study drug's liability for acute hepatocellular DILI. The eDISH identifies drugs in clinical trials with DILI fatality (death or transplant) risk that may be unacceptable in a post-market setting. As a two-dimensional graph that plots peak total bilirubin (TB) versus peak serum aminotransferase levels for each patient during study drug or comparator treatment, eDISH identifies potential cases of acute, modest, and serious hepatocellular DILI for in-depth analysis of liver tests (LT) and clinical course so that the likelihood of causal association with the study drug can be determined. Unfortunately, the generalizable utility of this tool only pertains to trials enrolling patients with normal or near normal (NNN) baseline (BL) serum LTs. The eDISH does not necessarily apply to trials of patients with abnormal baseline (ABN-BL) LTs that often coincide with underlying liver disorders. Because drug development programs being reviewed by the FDA increasingly target liver disorders, we are often challenged to evaluate DILI risk in trials of patients with ABN-BL LTs. Also, the high background prevalence of metabolic dysfunction associated steatotic liver disease (MASLD) means patients with LTs above NNN may need to be enrolled in trials treating non-liver disorders to reflect the target population. Such study populations create challenges for industry and regulators because eDISH may not reliably categorize or identify potential cases of DILI for further analysis, as it so efficiently does in NNN-BL trials. We describe the main functionalities of eDISH in NNN-BL trials to understand what should be emulated by new tools or eDISH modifications. We then discuss non-eDISH-based plots that may be useful in ABN-BL trials.

Fig. 1

Hypothetical eDISH plot of a trial with active treatment (Drug x) and placebo arms, enrolling approximately two to one. Each subject’s maximum (Max), on-treatment total bilirubin (TB) and maximum on-treatment aminotransferase (alanine aminotransferase [ALT] or aspartate aminotransferase [AST]) are plotted. Four quadrant categories are defined by TB >2× upper limit of normal (ULN) and ALT or AST ≥ 3 × ULN. Potential Hy’s Law quadrant may contain drug-induced liver injury (DILI) subjects with a 10% fatality risk. The subjects circled in red may be prioritized for in-depth review

Fig. 2

Modified eDISH (mDISH) plot based on peak total bilirubin (TB) and peak alanine aminotransferase (ALT) in times baseline (× baseline [BL]) and modified from a clinical trial with three treatment arms (red triangles, blue squares, and green circles). Each data point represents one subject’s peak on-treatment TB and ALT levels in multiples of their own baseline values. Hypothetical Subjects A and B plot to the same location because each had a peak bilirubin four times their baseline values and peak ALT five times their baseline values. However, Subject B’s jaundice and ALT of 1000 U/L suggest a more severe potential drug-induced liver injury (DILI) compared to Subject A’s bilirubin and ALT. Subject C’s deeper jaundice and ALT rise suggest a more worrisome injury compared to Subject A’s or B’s bilirubin and ALT, yet Subject C is lower and more to the left compared to Subjects A and B

Fig. 3

Sankey plots and cross table counts for a Study 1 and b Study 2. Sankey baseline categorizations are in the middle. Peak on-treatment shifts from baseline for the placebo treatment arm move left, while shifts from baseline for the active treatment arm move right. All categorizations are based on × upper limit of normal (ULN) alanine aminotransferase (ALT) and total bilirubin (TB) levels. Unfavorable shifts (potential drug-induced liver injury [DILI]) move upward and are highlighted in pink. Study 1 had fewer types of upward shifts (3 vs 5), and fewer total numbers in these shifts (13 vs 15) in the drug arm compared to placebo. Study 2 had markedly more upward (pink) and downward (green) shifts in the drug arm compared to placebo arm. Cross tables have shift counts with unfavorable (potential DILI) in red, favorable (potential efficacy) in green, and neutral shifts in grey and yellow for each arm. Near normal (NN) = ALT between ULN and < 3 × ULN

Fig. 4

Composite plot modified eDISH (mDISH) plot adapted from Tesfaldet et al [12]. Only the five subjects from Study 2 who met alanine  aminotransferase (ALT) (× upper limit of normal [ULN]) and total bilirubin (TB) (×ULN) criteria for Hy’s Law quadrant are plotted. The category (quadrant) of origin for each subject is indicated by marker color and shape. Four of the five subjects started in the normal or near normal quadrant at baseline (green squares), and one subject started in Temple’s quadrant at baseline (blue plus sign). There were no subjects in Hy’s Law quadrant based on peak values who were in that quadrant at baseline or in the cholestasis quadrant at baseline

Fig. 5

Modified waterfall plot for Study 2. Alanine aminotransferase (ALT) levels in U/L are plotted for each subject’s baseline value along the black lines and each subject’s maximum change on treatment are depicted by vertical bars, blue for the placebo arm subjects and bronze for active treatment arm subjects. Subjects who developed jaundice have green bars. The active treatment arm had more subjects with decrease in ALT seen by the large number of bronze bars dropping below baseline compared to the number of blue bars dropping below baseline for placebo subjects. However, there were more subjects in the active treatment arm with substantial increases in ALT over baseline and several developed jaundice. These cases may have had drug-induced liver injury (DILI) and can be easily identified and prioritized for in-depth analyses