How FDA Assesses Drug-Induced Liver Injury: Hy's Law, eDISH, and Liver-Safety Monitoring Expectations

How FDA assesses drug-induced liver injury: Hy's Law, eDISH, and liver-safety monitoring expectations

Drug-induced liver injury (DILI) is one of the most consequential safety findings a review division can encounter, because the clinical trial database that supports an approval is almost never large enough to detect rare, idiosyncratic, potentially fatal hepatotoxicity directly. FDA's response has been to rely on a biochemical surrogate, Hy's Law, that flags the potential for severe injury from a handful of cases, and to build monitoring, labeling, and post-marketing frameworks around that signal. This article summarizes how FDA clinical reviewers actually apply Hy's Law, how they screen an entire program with eDISH, what liver-safety monitoring they expect, and how DILI shows up in Complete Response Letters (CRLs) and approved labeling.

What Hy's Law is, and why one case is enough to worry

FDA clinical reviewers treat Hy's Law as a conservative screen for potentially severe, hepatocellular DILI. A biochemical Hy's Law case has three components that must occur together:

Hepatocellular injury: ALT or AST greater than 3x the upper limit of normal (ULN) ²⁷²⁸³²³³⁴⁰⁴⁵.
Impaired bilirubin excretion: total bilirubin greater than 2x ULN, arising after the aminotransferase rise ²⁷²⁸³²³³⁴⁰⁴⁵.
No cholestatic explanation: no initial rise in alkaline phosphatase, and reviewers commonly exclude cases with ALP greater than 2x ULN because that pattern points to cholestasis rather than hepatocellular injury ²³²⁷²⁸⁴⁰.

Critically, DILI is a diagnosis of exclusion. A case only counts when the aminotransferase-plus-bilirubin combination is not better explained by another cause, such as viral hepatitis A/B/C, pre-existing or acute liver disease, biliary obstruction, alcohol, or another concomitant drug ²⁷²⁸³²³³⁴⁵. Reviewers work from the biology: ALT/AST reflect hepatocellular injury, while bilirubin reflects loss of liver function, so the combination signals enough hepatocellular damage to impair the liver's excretory capacity ³⁹.

The reason a small number of cases carries so much weight is the observed post-marketing mortality. Reviewers state plainly that one Hy's Law case in a clinical database is worrisome and that two are highly predictive that the drug can cause severe DILI in a broader population ²⁷. Confirmed Hy's Law injury is associated with serious post-marketing outcomes, including fatal cases and transplant, with mortality estimated at roughly 10% in one review and historically described as 10 to 50% ²⁸⁴⁵. Reviewers emphasize that this is the pattern that can progress to acute (fulminant) liver failure, which is why even a single confirmed case is read as a meaningful signal of future fatal DILI risk ²⁸⁴⁰.

How FDA screens a whole program: eDISH and Temple's Corollary

Because individual case-finding does not, by itself, tell reviewers whether a drug is hepatotoxic across a population, FDA uses the eDISH (evaluation of Drug-Induced Serious Hepatotoxicity) analysis as a program-level screening and visualization tool. eDISH plots each subject's peak aminotransferase (ALT or AST) against peak total bilirubin, typically on log scales, so reviewers can rapidly locate subjects that may represent serious hepatotoxicity ⁶⁶⁶⁷⁷⁰⁷⁶⁸⁰.

In practice, reviewers use the plot in two complementary ways:

The right-upper quadrant identifies "possible Hy's Law" subjects: aminotransferase greater than 3x ULN together with total bilirubin greater than 2x ULN, especially where ALP is not elevated (ALP less than 2x ULN) ⁶⁵⁶⁷⁷⁶⁸⁰⁴⁰⁸¹²⁵. A 30-day concurrence window is often applied so that a bilirubin rise within about a month of the aminotransferase rise can still be counted as a potential case ⁸⁰. Each flagged subject is then adjudicated individually against narratives and competing etiologies (another drug, viral hepatitis, gallstone disease, infection, muscle injury), and cases with a clear alternative explanation may be judged unlikely to be true DILI ⁶⁶⁷³⁴⁰⁸¹⁸⁵.
Temple's Corollary tracks the more common, enzyme-only pattern: ALT/AST greater than 3x ULN without accompanying bilirubin elevation. Reviewers compare the number and rate of these isolated transaminase elevations in the drug arm versus placebo or comparator, often at multiple thresholds (greater than 3x, 5x, or 10x ULN) and exposure-adjusted where available ⁶⁵⁷⁴⁸⁵⁶⁷⁶⁸.

An imbalance in isolated transaminase elevations does not establish Hy's Law on its own, but a higher rate in the drug arm is treated as a liver-safety signal that warrants closer review ⁶⁵⁷⁴⁸⁵. In short, eDISH finds the rare "enzyme plus bilirubin" cases, Temple's Corollary tracks the more frequent enzyme-only elevations, and reviewers combine both with case-level adjudication to decide whether any true Hy's Law/DILI cases exist ⁶⁶⁶⁷⁷⁶⁴⁰.

Liver-safety monitoring expectations: baseline, periodic testing, and stopping rules

When a hepatic signal is present or plausible, FDA reviews and the resulting labeling impose structured monitoring. The recurring framework is baseline testing, scheduled periodic testing, and threshold-based rules to interrupt or discontinue the drug, followed by close follow-up to normalization and cautious or absent rechallenge.

Baseline and periodic monitoring examples from FDA-reviewed products:

Ambrisentan (Letairis): serum aminotransferases and bilirubin measured before initiation and monthly thereafter ⁴⁸⁵⁴.
Bosentan (Tracleer): serum aminotransferases measured prior to initiation and monthly, with strict adherence emphasized for the treatment duration ⁶³.
Simeprevir (Olysio) combination therapy: monitor liver chemistry tests before and as clinically indicated during treatment ⁵⁷.

The clearest worked example of stepwise stopping rules is ambrisentan labeling ⁴⁸:

ALT/AST greater than 3x and less than 5x ULN: re-measure; if confirmed, reduce dose or interrupt and monitor every 2 weeks until below 3x ULN.
ALT/AST greater than 5x and less than 8x ULN: discontinue; monitor until below 3x ULN, then may re-initiate with more frequent checks.
ALT/AST greater than 8x ULN: stop and do not consider re-initiation.
Any aminotransferase elevation accompanied by symptoms of liver injury or bilirubin greater than 2x ULN: stop treatment.

Bosentan applies the same bilirubin-plus-symptom stopping logic (stop if aminotransferase elevation is accompanied by symptoms or bilirubin greater than or equal to 2x ULN) ⁶³. Trial-level stopping rules can be even more granular: the rivaroxaban RECORD program discontinued for jaundice or clinical liver-injury symptoms with abnormal labs, ALT greater than 4x ULN for more than 4 weeks, or ALT greater than 3x ULN plus total bilirubin greater than 2x ULN with a direct/indirect bilirubin ratio criterion, and escalated monitoring cadence (from retesting within days to weekly or daily) depending on the trajectory ⁴⁷.

Two other expectations recur across reviews:

Avoid initiation when baseline aminotransferases are already abnormal, because monitoring becomes difficult. Both ambrisentan and bosentan should generally be avoided when baseline aminotransferases exceed 3x ULN ⁴⁸⁵⁴⁶³, and several programs (rivaroxaban, ozanimod) excluded patients with significant liver disease or baseline transaminase/bilirubin above prespecified thresholds ⁴⁷⁵⁹.
Rechallenge is approached with great caution. Ambrisentan may be re-initiated only after enzymes fall below 3x ULN and with more frequent monitoring, and there is explicitly no experience with re-introduction after stopping for symptoms or bilirubin elevation greater than 2x ULN ⁴⁸⁵⁴.

FDA reviews also show reviewers building bespoke Hy's-Law case-finding strategies during review. For anidulafungin, FDA requested collation of all liver chemistries and expert-hepatologist graph review for subjects with more than 2-fold elevations, using a search rule of ALT greater than 2x ULN with concomitant or up to one-month-delayed bilirubin greater than 1.5x ULN ⁴⁹⁵⁰⁵¹. For ozanimod, follow-up of ALT/AST elevations of 2 to 3x ULN was conducted per FDA DILI guidance ⁵⁹.

When DILI blocks approval: the signal in Complete Response Letters

DILI recurs as a stated basis for non-approval. The CRLs below show how FDA frames the hepatic signal and what additional characterization it demands:

Tolebrutinib (Genzyme, NDA 219624): FDA cited a serious risk of severe, including fatal, DILI, describing the risk as substantial and unusually high and idiosyncratic. The program had 6 Hy's Law cases among roughly 2,700 subjects, including one death after liver transplant, and 3.6% of treated subjects versus 1.9% on placebo had aminotransferases greater than 3x ULN in the first 6 months ³. FDA concluded that weekly liver monitoring under a REMS did not adequately mitigate the risk, because additional DILI cases (including a Hy's Law case and elevations of 10 to 60x ULN, some with hyperbilirubinemia) occurred even after weekly monitoring began ¹, and it asked for further analyses of the completed and extension studies to characterize the effect of weekly monitoring on severe DILI incidence ⁷.
Vadadustat (Akebia, NDA 215192): FDA found a clinically significant risk of DILI, with 1 Hy's Law case (meeting Hy's Law under the newer R-value criterion, with at least 10% mortality risk) and at least 7 probable DILI cases with significant ALT elevation without jaundice, plus an ALT imbalance versus darbepoetin alfa ⁸. FDA was concerned that trial monitoring could underestimate real-world risk with oral administration and that an effective hepatotoxicity mitigation strategy would be needed ⁸¹³.
Relacorilant (Corcept, NDA 219398): FDA concluded the drug is associated with DILI. Although no subject met traditional Hy's Law criteria, there were 4 probable DILI cases, including one with ALT of 1,952 U/L (more than 50x ULN); FDA noted that ALT greater than 1,000 U/L raises the risk for jaundice/Hy's Law and liver-related fatality, and that a continued enzyme rise after discontinuation did not exclude drug causality ².
Dapagliflozin (Bristol-Myers Squibb, NDA 202293): FDA identified a biochemical Hy's Law case in the phase 2b/3 pool, deemed a probable DILI case because no other cause was found, and requested an updated hepatic safety review, narratives for each biochemical Hy's Law case, and transaminase-elevation incidence at 3x, 5x, 10x, and 20x ULN across treatment and comparator groups ¹⁸¹⁹.
Glepaglutide (Zealand Pharma, NDA 218828): FDA listed potential DILI among multiple safety concerns in a limited safety database, though without a detailed Hy's Law count in the cited material ⁹.

A consistent theme across these letters is that FDA does not treat a monitoring plan as automatically sufficient. Where injury appears idiosyncratic and potentially severe (tolebrutinib, vadadustat), FDA questioned whether monitoring or a REMS could realistically prevent acute liver failure, transplant, or death, and it asked applicants to quantify the residual risk rather than assert that monitoring resolves it ³⁷⁸¹³.

When DILI shapes the label: boxed warnings, REMS, registries, and post-marketing studies

For products that are approved despite a hepatic signal, FDA's toolkit is graded to the severity and manageability of the injury. Reviews show a spectrum from labeling-only mitigation to restricted-distribution REMS with mandatory monitoring and registries.

Higher-intensity actions, driven by Hy's Law cases, transplant, or death:

Tolvaptan (Jynarque): with 3 Hy's Law cases and one severe injury requiring liver transplant, FDA required a REMS (restricted distribution, certified prescribers/pharmacies/patients, scheduled liver tests, and a REMS registry), a boxed warning, a contraindication in patients with signs or symptoms of liver injury/impairment, and a post-marketing registry to characterize incidence and irreversible injury; monthly liver testing for 18 months, then periodic testing, was part of the REMS ⁹²⁹⁸¹⁰⁹¹⁰⁶.
Pexidartinib (Turalio): with 2 irreversible liver injuries among 768 patients, including one transplant and one death involving liver failure, plus biopsy evidence of bile-duct injury/ductopenia, FDA required a boxed warning for hepatotoxicity, a restricted REMS, a patient registry, and a post-marketing study of long-term risk, even while noting the risk was manageable with dose interruption/reduction in trials ¹⁰¹¹⁰²¹⁰⁴¹¹⁴.
Sparsentan (Filspari): reviewers found 6 probable/possible DILI cases (most with positive rechallenge) and one near-miss Hy's Law case, leading to a boxed warning, inclusion of hepatotoxicity in the REMS with monthly liver testing, and a post-market study. A later supplement, reporting 27 patients with transaminase elevations or serious liver events but no Hy's Law cases, supported moving monitoring to quarterly while keeping the boxed warning and REMS safeguards ⁹³⁹⁵⁹⁷.
Tocilizumab (Actemra): one Hy's Law case in roughly 4,700 RA-program patients (confounded by concomitant methotrexate) supported a REMS with liver-enzyme monitoring and dose-modification recommendations at 4 to 8 week intervals ¹⁰⁵.
Daclizumab (Zinbryta): reviewers concluded serious hepatotoxicity would occur post-marketing and would require a boxed warning, stringent monitoring, and a REMS, with some experts arguing the risk could not be minimized by labeling and REMS because serious toxicity occurred despite strict monitoring and stopping criteria ⁹⁶.

Historic examples that shaped current practice through post-marketing hepatic failure:

Nefazodone (Serzone): post-marketing life-threatening hepatic failure led to a boxed warning, advice not to start in patients with active liver disease or elevated baseline transaminases, discontinuation if AST/ALT rose above 3x ULN or liver failure was suspected, and a contraindication to re-treatment in patients with evidence of liver injury ¹⁰⁸¹¹⁵.
Pemoline (Cylert): post-marketing acute liver failure, many cases resulting in death or transplant, prompted a boxed warning and labeling that the drug should not ordinarily be first-line therapy ¹¹⁰¹¹¹.

Lower-intensity actions, where FDA judged the signal manageable by labeling:

Alogliptin (Nesina/Oseni): FDA recommended baseline liver tests and monthly monitoring for at least 6 months and maximized labeling for hepatotoxicity, but did not recommend a REMS at that time ¹⁰⁰¹⁰³.
Pirfenidone (Esbriet): FDA recognized possible liver injury with reversible enzyme elevations but no liver failure or deaths, and concluded the risk could be managed with labeling and pharmacovigilance rather than a REMS ⁹⁹.

Taken together, when a hepatic signal is present FDA's recurring interventions are a boxed warning for hepatotoxicity, a REMS with restricted distribution or ETASU, mandatory liver-test monitoring schedules, patient registries, post-marketing requirements to define incidence and long-term risk, and contraindications or avoidance in patients with existing liver disease ⁹²¹⁰¹¹⁰⁵¹⁰⁸¹⁰⁹¹¹⁰¹¹⁵. Reliance on labeling alone is reserved for signals that appear reversible and manageable ⁹⁹¹⁰⁰¹⁰³.

Practical implications for sponsors

Three practical points follow from how FDA reads these signals. First, Hy's Law is a program-level judgment, not a checkbox: reviewers adjudicate each candidate case against cholestasis (ALP) and competing etiologies, so clean, prospectively collected liver chemistries, baseline characterization, and full narratives materially affect whether a case is counted ²³²⁷²⁸³⁹⁴⁵. Second, a monitoring plan is not a guaranteed mitigation. The tolebrutinib and vadadustat CRLs show FDA will discount monitoring or a REMS when injury is idiosyncratic and can progress faster than testing intervals detect, and will instead ask the sponsor to quantify residual severe-DILI risk ¹³⁸¹³. Third, the response is calibrated to severity: reversible, enzyme-only signals may be handled in labeling, whereas confirmed Hy's Law cases, transplant, or death tend to drive boxed warnings, REMS with defined monitoring cadence, and post-marketing registries ⁹²⁹³⁹⁹¹⁰⁰¹⁰¹. For any program with an emerging hepatic signal, building the eDISH analysis, Temple's Corollary comparison, and a defensible stopping-rule and rechallenge strategy into the development plan early is the most direct way to align with how the review division will evaluate liver safety ⁴⁸⁶³⁶⁵⁶⁶⁷⁴.

How FDA Assesses Drug-Induced Liver Injury: Hy's Law, eDISH, and Liver-Safety Monitoring Expectations