When Is Suspected Drug-Induced Liver Injury (DILI) Not DILI? An Analysis of Unlikely Cases From the Drug-Induced Liver Injury Network

Abstract

INTRODUCTION:

Diagnosis of drug-induced liver injury (DILI) is difficult. We reviewed cases in the DILI Network prospective study that were adjudicated to have liver injury due to other causes to discover pearls for improved diagnostic accuracy.

METHODS:

Cases were adjudicated by expert opinion and scored from 1 (definite DILI) to 5 (unlikely DILI). Confirmed cases (1–3) were compared with unlikely cases (5).

RESULTS:

One hundred thirty-four of the 1,916 cases (7%) were unlikely DILI. Alternative diagnoses were autoimmune hepatitis (20%), hepatitis C (20%), bile duct pathology (13%), and hepatitis E (8%).

DISCUSSION:

Thorough evaluation, including follow-up, is essential to minimize incorrect diagnosis of idiosyncratic DILI.

INTRODUCTION

Drug-induced liver injury (DILI) can lead to asymptomatic laboratory abnormalities or acute liver failure (1). DILI may occur because of prescription medications, over-the-counter medications, or herbal and dietary supplements. Making a correct diagnosis of DILI requires an astute clinician to perform the detective work in unveiling the suspected product and excluding alternative diagnoses (2). To advance understanding of DILI, the National Institutes of Health established the DILI Network (DILIN) in 2004 to identify the etiologies and outcomes of DILI. Despite the high pretest probability of DILI in this study, some cases are ultimately adjudicated to be from alternative etiologies, that is, unlikely DILI. Factors that lead experts to determine that a case of liver injury is not due to DILI have not previously been described. Such factors may aid clinicians in coming to more accurate diagnoses and initiating therapy for alternative causes more rapidly. Therefore, the aim of this study was to describe the most common missed diagnoses when DILI is suspected, and the clinical factors that should lead to an alternative diagnosis.

METHODS

Consecutive subjects enrolled from September 24, 2004, through December 31, 2021, were included. Study subjects provided consent before enrollment. Study participation was approved by each site’s institutional review board.

Details of the DILIN protocol have been previously published (1). Each patient is followed for at least 6 months after enrollment, and the case is scored according to the Roussel Uclaf Causality Assessment Method (RUCAM) and also given an expert opinion causality score of 1–5 (3). Comparisons were made between definite-probable DILI (score 1–3) and unlikely DILI (score 5).

Descriptive data are reported with N and frequency. Differences in bivariate comparisons are made with the χ² test and analysis of variance where appropriate. Owing to its association with autoimmune hepatitis, human leukocyte antigen (HLA) in unlikely DILI cases was compared with patients with autoimmune hepatitis (AIH), patients with true hepatocellular DILI, and population controls from NCBI Database of Genotypes and Phenotypes Electronic Medical Records and Genomics and Biome data sets. HLA allele frequency difference between comparison groups was tested using the Fisher exact test and logistic regression with Firth correction. Statistical analyses were performed by H.B., J.G., and A.D.

RESULTS

One hundred thirty-four of the 1,916 cases reviewed (7%) were unlikely DILI. There were no demographic features to distinguish DILI from unlikely DILI. Unlikely cases more often had underlying comorbidities, such as renal disease and HIV (18% vs 9% and 7% vs 2%). Latency for implicated drugs in unlikely cases was either very brief (<1 week 9% vs 5%) or very long (>24 weeks 28% vs 16%) (P = 0.002 overall). Clinical and most laboratory features of unlikely DILI and confirmed DILI cases were similar. There was a numerically higher proportion of hepatocellular injury pattern of liver enzymes among unlikely DILI cases that did not meet statistical significance. The mean and median R-values were similar as well. The international normalized ratio was higher among unlikely cases (1.8 vs 1.5, P = 0.009) (Table 1).

Table 1:

Selected Patient Demographics and Clinical Characteristics

Characteristic	Unlikely DILI (score 5) N=134	Definite, highly likely or probable DILI (Score1–3) N=1782	P-value
Demographics
Age Mean (SD)	50.4 (17.7)	49.9 (17.2)	0.96
Gender Female	50.0%	58.1%	0.19
Self-reported race
White or Caucasian	78.9%	77.8%	0.99
Black or African American	11.3%	12.7%
Asian	4.5%	3.7%
Other/Multiracial	5.3%	5.9%
Latino	11.2%	10.7%	0.99
BMI (kg/m2) Mean (SD)	27.5 (6.6)	27.5 (6.4)	0.99
Pre-existing medical conditions
Prior drug allergies	52.2%	42.6%	0.10
Alcohol use	43.2%	47.9%	0.58
Past medical history
Diabetes/endocrine disorder	29.1%	23.0%	0.28
Neurological disease	14.2%	11.9%	0.74
Heart disease	22.4%	14.8%	0.07
Renal disease	17.9%	9.2%	<0.01
Pulmonary disease	20.9%	14.8%	0.17
Gastrointestinal disease	38.1%	29.6%	0.12
Malignancy	15.7%	11.6%	0.37
Congestive heart failure	0%	1.2%	0.43
HIV positive at baseline	6.8%	1.7%	<0.01
Past medical history - liver
HCV	11.9%	3.3%	<0.01
HBV	4.5%	0.8%	<0.01
Alcohol-related liver disease	0%	0.5%	0.71
Non-alcoholic fatty liver disease	0.7%	1.9%	0.63
Days from primary drug start to DILI onset
<= 1 week	9.3%	4.8%	<0.01
2 to 4 weeks	19.4%	28.0%
5 to 12 weeks	24.8%	36.7%
13 to 24 weeks	18.6%	14.6%
> 24 weeks	27.9%	15.8%
Pattern of liver injury			0.57
Cholestatic	23.4%	22.9%
Hepatocellular	60.5%	54.5%
Mixed	16.1%	22.6%
Median R-Value (IQR)	7.9 (2.1, 15.4)	5.8(2.2, 14.9)	0.82
Initial RUCAM score			<0.01
Highly probable	3.0%	18.6%
Probable	34.1%	62.9%
Possible	38.6%	16.4%
Unlikely/Excluded	24.2%	2.0%

RUCAM scoring did show a statistically significant shift toward unlikely/excluded categories among unlikely cases (24.2% vs 2.0% P < 0.01). Although fewer unlikely cases were scored highly probable or probable (3.0% vs 18.6% and 34.2% vs 62.9% P < 0.01), there were some outliers (Table 1).

The most common alternative diagnoses for unlikely cases were autoimmune hepatitis (AIH, n = 27, 20.1%), hepatitis C (HCV, n = 27, 20.1%), and gallstones/biliary obstruction (n = 18, 13.4%). There were 44 other cases attributed to >20 diverse etiologies. Nearly all AIH cases underwent liver biopsies, but findings were not diagnostic for DILI or AIH (Table 2).

Table 2:

Alternative Diagnoses in Unlikely Drug Induced Liver Injury Cases

Diagnosis	n	%	Comments
Autoimmune hepatitis	27	20.1	• The most common incorrectly implicated class of medications were antibiotics (9/27) followed by herbal and dietary supplement products (8/27). • Nearly all had a liver biopsy, but the findings were not diagnostic for drug induced liver injury or autoimmune hepatitis. • Other common themes among these cases were 1) prevalent underlying immune mediated disease (systemic lupus erythematous, autoimmune thyroiditis, inflammatory bowel disease, etc.) in 8/27 cases; 2) negative or relatively low antinuclear antibody titers (<1:80) which increased on follow up; 3) a response to steroids in 14/27 with six of these having a flare in liver injury after immunosuppression taper or withdrawal.
Hepatitis C	27	20.1	• 24 were acute hepatitis C and three chronic. • Of the 24 acute cases, 17 were enrolled prior to 2011 and none since 2017. Most of these cases had negative hepatitis C antibodies and were discovered to have a positive hepatitis C RNA later in their clinical course. • Three patients had prevalent detectable hepatitis C virus which later cleared on follow-up.
Gallstone/biliary obstruction	18	13.4	• Eight cases had biliary tract malignancy. • Three cases had primary sclerosing cholangitis. • Normal imaging in some patients who, it was thought, had acutely passed gall stones. • No differences in the presence of stones (3.3% vs. 6.7%, p=0.32), or ductal dilation (3.3% vs. 2.5%, p=0.89) between unlikely DILI and DILI cases.
Hepatitis E	11	8.2	• Diagnosis often delayed as tests for HEB IgM and RNA are send-outs in most centers.
Sepsis/shock liver	7	5.2	• Frequently in Intensive Care Unit on multiple drugs and hypotensive.
Other (>20 diverse etiologies)	44	32.8	• Etiologies included alcohol, myopathy, Epstein Barr virus, cytomegalovirus, nonalcoholic steatohepatitis, metastatic cancer, granulomatous liver disease and other/unknown.

HLA allele frequency was measured on 87 unlikely cases compared with 17,842 population controls. White subjects with AIH (n = 15), rather than DILI, had increased carriage of the AIH haplotype—HLA B*08:01/C*07/DRB1*03:01 (allele frequency = 0.4 v dbGap White controls = 0.16, P = 0.023). No other HLA associations were discovered to reveal associations with alternative diagnoses.

Patients with unlikely DILI had greater all-cause (16% vs 7%, P < 0.001) and liver-related (10% vs 3%, P < 0.001) mortality than patients with DILI at 6 months. Patients with unlikely DILI died within 6 months at a higher frequency as well (14% vs 6%, P = 0.004). Overall hospitalization rates, durations of illness, and receipt of liver transplants were similar.

DISCUSSION

In the DILIN, 7% cases of suspected DILI were adjudicated by a panel of experts to be unlikely DILI. Overall, this was a negative study. Patient demographics, suspect drugs, clinical features, and presenting symptoms did not differ between patients with confirmed and unlikely DILI. Patients with unlikely DILI did have more comorbid conditions. This finding is consistent with previous work showing an inverse relationship between DILI causality scores and significant comorbidities (4). At enrollment, the international normalized ratio (1.8 vs 1.5) was higher among unlikely cases, suggesting more severe liver injury; accordingly, mortality was also higher among the unlikely group, likely because of sepsis and malignancies as alternative diagnoses.

Among the subjects with HCV, 24 of the 27 were adjudicated to have acute HCV. Most of these cases (17/24) were enrolled before 2010, and there have been none since 2017. We suspect that there are fewer missed diagnoses of HCV in recent years because of overall greater awareness of disease and more widely available testing. In addition, there were 11 cases of acute hepatitis E virus infection, but the incidence of this in the DILIN cohort has been declining in recent years most likely because of more frequent testing before study referral (5). There were rare cases of hepatitis B virus flares among the unlikely DILI cases. Owing to the small N, these are part of the other category. It is critically important to consider hepatitis B virus reactivation in patients who receive immunosuppression.

AIH is particularly challenging because of the overlapping continuum of de novo AIH, drug-induced autoimmune-like hepatitis (DI-ALH), and DILI (6). Differentiating AIH from DI-ALH is critical because patients with DI-ALH usually do not require long-term immunosuppression. However, a diagnosis of AIH vs DI-AIH is difficult because of a lack of diagnostic markers for DI-ALH and no single feature that is pathognomonic for AIH. An expert opinion meeting that recently reviewed this topic concluded that a revision of diagnosis after long-term follow-up may be necessary in some cases (7). In this cohort, nearly every patient adjudicated as unlikely DILI underwent liver biopsy and most had a response to corticosteroids. Several with true AIH had a flare of liver enzymes with steroid withdrawal. A third had prevalent immune-mediated disease in nonhepatic organs, which is present in 14%–44% of patients with AIH (8,9). These features, especially steroid response, which can only be appreciated with a longitudinal follow-up, were essential for accurate diagnosis. Demonstration of HLA types known to be associated with AIH or other autoimmune disorders might be helpful in difficult cases; however, even in DILIN, these are only used for research purposes and not case adjudication. Other alternative diagnoses are even more challenging to make prospectively and may be appreciated only in hindsight.

RUCAM testing for DILI adjudication is helpful but has limitations. In this study, overall RUCAM scores trended lower for unlikely cases, yet 37.1% still scored highly probable or probable. This discrepancy with the expert opinion adjudication may reflect RUCAM points given for improvement in liver enzymes within 180 days, which would be expected in several alternative etiologies such as shock/sepsis, acute passage of a gallstone, or acute viral hepatitis. In addition, although RUCAM uses quantitative scoring, some of the inputs may be ambiguous and RUCAM has poor interuser and intrauser reliability (10,11).

This study has limitations. The DILIN study is an observational study of patients with a high suspicion of DILI; thus, there is selection bias toward DILI, with 85% of cases being confirmed (score 1–3) as such, and there is no true control group. Adjudication is performed by RUCAM scoring and by consensus among a panel of hepatologists with special expertise and interest in DILI. In earlier work from DILIN, our consensus approach was assessed as superior to RUCAM (3).

In summary, establishing a diagnosis of DILI is challenging. Viral hepatitis, AIH, biliary obstruction, and sepsis must be considered as potential alternative diagnoses because these causes of liver injury are more common than idiosyncratic DILI in the general population (12). Tips to avoid misdiagnosis of DILI include (i) distinguishing acute HCV from DILI with an HCV RNA test, (ii) cross-sectional imaging and consideration of a passed bile duct stone, and (iii) an AIH diagnosis should include a liver biopsy and a trial of steroids. Overall, a longitudinal follow-up of patients with presumed DILI is essential to continuously re-examine the diagnosis, treat an alternative disease, and potentially absolve a medication presumed to have caused DILI.