eDISH: Evaluating Drug-Induced Serious Hepatotoxicity in Clinical Trials

eDISH—Evaluation of Drug-Induced Serious Hepatotoxicity—is a graphical method for organising liver-test data across a clinical-trial population. Its purpose is to help reviewers identify subjects with combinations of aminotransferase and bilirubin abnormalities that deserve focused medical evaluation, particularly potential Hy's Law cases.

Purpose and Scientific Context

Severe idiosyncratic drug-induced liver injury (DILI) is uncommon. A development programme may expose several thousand participants without observing acute liver failure, yet still contain biochemical patterns that suggest a drug has the potential to cause serious hepatotoxicity.

FDA's final guidance on premarketing clinical evaluation of DILI addresses this problem by emphasising systematic evaluation of lesser liver-test abnormalities that may signal a drug's potential for severe injury. eDISH emerged as a practical way to visualise such data across a study population.

The central idea is simple: a single laboratory listing is good at showing individual values, but poor at revealing how hepatocellular injury markers and bilirubin behave together across hundreds or thousands of participants. eDISH places those dimensions on the same graph.

What an eDISH Plot Represents

A conventional eDISH plot places a participant's peak or otherwise prespecified post-baseline aminotransferase value—usually ALT and sometimes AST—on one axis and total bilirubin on the other. Values are expressed as multiples of the relevant laboratory upper limit of normal (ULN), commonly on logarithmic scales.

Each point therefore represents a participant, not an adverse-event diagnosis.

The plot may answer questions such as:

• Which participants developed substantial ALT or AST elevations?
• Which developed substantial bilirubin elevations?
• Which developed both?
• Are concerning points concentrated in the investigational-treatment arm?
• Is there a dose or exposure pattern?
• Which subjects require individual longitudinal review?

It cannot independently determine whether the investigational medicine caused liver injury.

The Major Plot Regions

Traditional eDISH interpretation uses reference lines at approximately 3× ULN for ALT/AST and 2× ULN for total bilirubin. These boundaries divide the graph into regions that help prioritise review.

Hy's Law Region

The upper-right region contains subjects with substantial aminotransferase elevation together with substantial bilirubin elevation. These participants warrant careful evaluation for a potential Hy's Law pattern.

Location in this region is a screening flag, not a final diagnosis. The reviewer still needs to examine alkaline phosphatase, injury pattern, baseline values, timing, competing causes, concomitant medicines and clinical outcome.

Temple's Corollary Region

The lower-right region contains substantial aminotransferase elevations without corresponding bilirubin elevation. Such findings can be important because a treatment arm with an excess of marked transaminase elevations may provide evidence of hepatocellular liability even when frank jaundice is absent.

Hyperbilirubinaemia / Cholestatic Region

The upper-left region contains bilirubin elevation without marked aminotransferase elevation. Possible explanations include cholestatic disease, biliary obstruction, haemolysis, Gilbert syndrome and other disorders of bilirubin metabolism. These cases remain medically important even though they do not fit the classic Hy's Law pattern.

Lower-Left Region

Most participants usually remain in the lower-left region, without substantial elevations in either plotted variable. The analytical value of eDISH lies in making the smaller number of outlying or clustered cases visible against this background.

eDISH and Hy's Law Are Related but Different

Hy's Law is a clinical and drug-safety concept concerning hepatocellular injury with bilirubin elevation in the absence of a more plausible alternative explanation. eDISH is a population-level visualisation method that helps identify subjects who may warrant evaluation for that pattern.

The relationship is therefore:

population liver-test data → eDISH outlier → longitudinal case review → injury-pattern assessment → competing-cause evaluation → causality judgement → drug-level liver-safety interpretation.

The plot is an entry point into medical reasoning, not the endpoint.

Building an Interpretable eDISH Dataset

The usefulness of an eDISH plot depends on how the underlying dataset is defined. The analysis specification should make clear which aminotransferase is used, what constitutes baseline and post-baseline, how the treatment-emergent period is defined, how ULN is assigned, and whether independent peak ALT and bilirubin values are plotted.

This matters because independently selected maxima may occur on different dates. A participant can therefore appear in a concerning quadrant even when the ALT and bilirubin peaks were temporally separated. The graphical position must always be followed by review of the laboratory time course.

Baseline Abnormalities

Many trial populations enter treatment with abnormal liver tests. Chronic liver disease, metabolic dysfunction-associated steatotic liver disease, viral hepatitis, malignancy involving the liver, alcohol-related disease and other conditions can produce elevated baseline aminotransferases or bilirubin.

A peak expressed only as a multiple of ULN can therefore be misleading.

For example, a rise from 1× to 6× ULN is not clinically equivalent to a fluctuation from a chronic baseline of 4× to 6× ULN. The reviewer should consider:

• several pretreatment measurements where available;
• the subject's underlying liver disease;
• change from baseline as well as change from ULN;
• disease-related fluctuations; and
• whether the abnormality is genuinely treatment emergent.

In populations with chronically abnormal liver tests, baseline-adjusted visualisations or complementary analyses may be needed.

Comparator and Treatment-Group Context

An eDISH plot becomes more informative when treatment arms can be compared. Background liver-test abnormalities occur in many clinical populations, so a pattern seen equally in active and comparator groups may carry a different interpretation from one concentrated in the investigational arm.

Relevant context includes:

• placebo or active comparator findings;
• number of subjects exposed;
• exposure duration;
• dose level;
• baseline liver status;
• concomitant hepatotoxic therapies; and
• differences in laboratory monitoring.

An apparent dose relationship can strengthen concern, but its absence does not exclude idiosyncratic DILI.

From Plot to Patient

The most important action after identifying an outlier is to reconstruct the individual subject's clinical course.

A useful chronology includes:

1. treatment start and dose changes;
2. baseline ALT, AST, ALP and bilirubin;
3. onset and trajectory of laboratory abnormalities;
4. symptoms such as fatigue, nausea, abdominal pain, pruritus or jaundice;
5. concomitant medicines and relevant exposures;
6. viral, autoimmune, vascular and biliary investigations;
7. treatment interruption or discontinuation;
8. dechallenge and, if it occurred unintentionally or was clinically justified, rechallenge information; and
9. clinical outcome.

The purpose is to determine whether the point on the graph represents a coherent episode of liver injury and, if so, what injury pattern and causal explanation are most plausible.

The Role of Alkaline Phosphatase and Injury Pattern

Traditional eDISH axes do not display ALP, yet ALP is essential to interpretation of potential Hy's Law cases because substantial cholestasis changes the clinical meaning of bilirubin elevation.

The reviewer should therefore characterise the liver-injury phenotype as hepatocellular, cholestatic or mixed using the available laboratory and clinical evidence. Where an R-value is calculated, the exact timing and laboratory values used should be clear.

The eDISH position alone should never substitute for this injury-pattern assessment.

Population-Level Interpretation

After individual cases are reviewed, the safety team returns to the drug level.

Questions include:

• Is there an excess of marked aminotransferase elevations in the investigational arm?
• Are there convincing Hy's Law cases?
• Do cases share a latency, phenotype or dose/exposure pattern?
• Are there supportive nonclinical or mechanistic findings?
• Is the pattern reproducible across studies?
• Does comparator experience provide a plausible background explanation?

This step is crucial because eDISH is not merely a case-finding tool; it also helps the reviewer understand whether isolated observations form a coherent drug-level safety signal.

Practical Review Method

A disciplined eDISH review can follow this sequence:

1. verify the dataset, treatment-emergent window and ULN definitions;
2. inspect treatment and comparator distributions;
3. identify subjects in the Hy's Law and Temple's Corollary regions;
4. review relevant baseline abnormalities;
5. examine longitudinal ALT, AST, ALP and bilirubin values;
6. establish whether peaks were temporally related;
7. characterise the injury pattern;
8. investigate competing causes;
9. assess individual-case causality; and
10. integrate the adjudicated cases into a drug-level liver-safety conclusion.

This is a scientific review workflow, not a regulator-prescribed eDISH SOP.

Potential Failure Modes

The following are illustrative failure modes rather than reported regulatory findings.

Relationship With Pharmacovigilance

Although eDISH was developed primarily for clinical-development datasets, its underlying reasoning is relevant to pharmacovigilance more broadly: screen large data, identify medically important subjects, return to the clinical record, evaluate causality, and then interpret the pattern at product level.

After marketing, however, spontaneous-report data rarely provide the complete serial laboratory information available in trials. Post-authorisation DILI evaluation therefore relies more heavily on case follow-up, literature, observational evidence, exposure context and aggregate signal assessment.

Practical Checklist

1. Is the plotted dataset defined and reproducible?
2. Are treatment-emergent and baseline periods clear?
3. Are ALT/AST and bilirubin peaks interpreted with their actual dates?
4. Are ALP and the liver-injury phenotype reviewed separately?
5. Are comparator and dose/exposure patterns considered?
6. Are important subjects medically adjudicated?
7. Are alternative causes actively investigated?
8. Is the drug-level conclusion based on adjudicated evidence rather than quadrant counts alone?

Relationship With Other QPPV.com Articles

For the broader clinical concept see [[hys-law]]. For the underlying disease entity see [[drug-induced-liver-injury-dili]]. These articles should be read together: DILI defines the clinical problem, Hy's Law identifies a particularly important severe-hepatotoxicity signal pattern, and eDISH organises the trial data so that relevant cases can be found and evaluated.

Key Takeaways

eDISH is a graphical clinical-trial liver-safety tool, not a diagnostic test.

Its main value is to make combinations of aminotransferase and bilirubin abnormalities visible across a study population and to support focused review of potential Hy's Law cases.

The meaning of an outlier depends on baseline status, timing, ALP, injury pattern, competing causes, comparator experience and individual clinical course.

The strongest use of eDISH connects population screening to longitudinal subject review and then returns to the population level for a defensible drug-safety conclusion.

References

1. U.S. Food and Drug Administration. Drug-Induced Liver Injury: Premarketing Clinical Evaluation. Guidance for Industry. July 2009.
2. Watkins PB, Desai M, Berkowitz SD, et al. Evaluation of drug-induced serious hepatotoxicity (eDISH): application of this data organization approach to phase III clinical trials of rivaroxaban after total hip or knee replacement surgery. Drug Safety. 2011;34(3):243-252.
3. Buchanan J, et al. A New Paradigm for Safety Data Signal Detection and Evaluation Using Open-Source Software Created by an Interdisciplinary Working Group. Drug Safety. 2021.

Regulatory Note

FDA's July 2009 DILI guidance remains final guidance and is non-binding. eDISH is a scientifically useful method for organising and reviewing clinical-trial liver-safety data; neither use of a particular software implementation nor a specific graphical workflow should be presented as a universal legal requirement.