Potential conflict of interest: Nothing to report.
The Spanish DILI Registry is supported, in part, by research grants from the Agencia Española del Medicamento and Fondo de Investigación Sanitaria (PS 09/01384). This research was also supported, in part, by the Intramural Research Program of the National Iinstitues of Health, National Cancer Institute.
Distinguishing drug-induced liver injury (DILI) from idiopathic autoimmune hepatitis (AIH) can be challenging. We performed a standardized histologic evaluation to explore potential hallmarks to differentiate AIH versus DILI. Biopsies from patients with clinically well-characterized DILI [n = 35, including 19 hepatocellular injury (HC) and 16 cholestatic/mixed injury (CS)] and AIH (n = 28) were evaluated for Ishak scores, prominent inflammatory cell types in portal and intra-acinar areas, the presence or absence of emperipolesis, rosette formation, and cholestasis in a blinded fashion by four experienced hepatopathologists. Histologic diagnosis was concordant with clinical diagnosis in 65% of cases; but agreement on final diagnosis among the four pathologists was complete in only 46% of cases. Interface hepatitis, focal necrosis, and portal inflammation were present in all evaluated cases, but were more severe in AIH (P < 0.05) than DILI (HC). Portal and intra-acinar plasma cells, rosette formation, and emperiopolesis were features that favored AIH (P < 0.02). A model combining portal inflammation, portal plasma cells, intra-acinar lymphocytes and eosinophils, rosette formation, and canalicular cholestasis yielded an area under the receiver operating characteristic curve (AUROC) of 0.90 in predicting DILI (HC) versus AIH. All Ishak inflammation scores were more severe in AIH than DILI (CS) (P ≤ 0.05). The four AIH-favoring features listed above were consistently more prevalent in AIH, whereas portal neutrophils and intracellular (hepatocellular) cholestasis were more prevalent in DILI (CS) (P < 0.02). The combination of portal inflammation, fibrosis, portal neutrophils and plasma cells, and intracellular (hepatocellular) cholestasis yielded an AUC of 0.91 in predicting DILI (CS) versus AIH. Conclusion: Although an overlap of histologic findings exists for AIH and DILI, sufficient differences exist so that pathologists can use the pattern of injury to suggest the correct diagnosis. (Hepatology 2011;)
Clinical diagnoses of idiopathic autoimmune hepatitis (AIH) and drug-induced liver injury (DILI) are challenging, as both conditions have heterogeneous disease manifestations clinically as well as histopathologically. These conditions are mediated by immunological reactions and thus show considerable resemblance in clinical and histopathologic features (e.g., autoantibodies and plasma cells and eosinophils). Manifestations of DILI can significantly vary, even among cases caused by a single agent. For instance, some drugs (e.g., statins, minocycline, nitrofurantoin, and infliximab) can cause idiosyncratic hepatocellular or cholestatic liver injury in some patients and autoimmune hepatitis—accompanied by evident autoimmunity and/or human leukocyte antigen haplotypes—in others.1-3 Further, it is often difficult to differentiate idiopathic AIH from AIH triggered by drugs in a clinical setting. Because it is practically impossible to exclude potential drug involvement in some cases, the differential diagnosis of idiopathic AIH versus DI-AIH can be extremely difficult. On the other hand, timely diagnosis and proper management are critical in both conditions.4-7 Early immunosuppressive therapy typically contains disease activity in patients with idiopathic AIH and can lead to disease remission.6, 7 Similarly, prompt identification and discontinuation of offending drugs halts ongoing liver injury in cases with DILI.4 Failure to properly treat AIH and DILI could result in clinically devastating acute or chronic outcomes.4, 5
Although AIH has some typical histologic patterns of injury, DILI may mimic any non-DILI pattern of injury. Currently, the role of liver biopsy evaluation in differentiating AIH from DILI remains uncertain. The histologic classification of DILI has been reported recently along with detailed descriptions of histologic features and corresponding differential diagnosis for each injury pattern.8 Characteristic histologic features of AIH have been also well documented in the literature.9, 10 Interface hepatitis, lymphocytic/lymphoplasmacytic infiltrates in portal tracts extending into the lobule, emperipolesis (i.e., presence of an intact lymphocyte within the cytoplasm of a hepatocyte), and hepatocyte rosette formation are considered common findings of AIH and are used in the recently published simplified criteria for the diagnosis of AIH.11 However, there is no individual histologic feature that is absolutely indicative of either DILI or AIH; thus, the evaluation of liver biopsy in determining AIH versus DILI may be a significant challenge. Previous histologic evaluations have been independently done on either DILI or AIH, but the relative frequency and/or degree of each histologic feature between the two entities remain unexplored. A better understanding of the spectrum of histologic features and/or severity of the features might be helpful in distinguishing AIH from DILI.
In this study, we performed a detailed, blinded histologic evaluation on liver biopsies from clinically well-characterized cases of AIH and DILI in a standardized manner. Our specific aims were to (1) characterize the histologic features in comparison of AIH versus DILI, (2) explore potential histologic hallmarks (or combination of histologic features) to differentiate AIH versus DILI, and (3) evaluate a diagnostic value of liver biopsy in this differential diagnosis. We also performed a subgroup analysis to compare the histologic features of DI-AIH versus AIH to explore potential histologic hallmarks for DI-AIH versus AIH.
AIH: autoimmune hepatitis; ALP: alkaline phosphatase; ALT: alanine aminotransferase; AST, aspartate aminotransferase; AUC: area under the curve; AUROC, area under the receiver operating characteristic curve; CS, cholestatic/mixed injury; DI-AIH: drug-induced autoimmune hepatitis; DILI: drug-induced liver injury; HC, hepatocellular injury; ROC: receiver operating characteristic; SE, standard error
Patients and Methods
We evaluated 63 liver biopsies of clinically well-characterized cases of AIH and DILI. Cases of AIH (N = 28) were obtained from the Mayo Clinic (Rochester, MN). The diagnosis of AIH was made by experienced hepatologists based on the presence of autoantibodies and/or gamma globulins with compatible histology and exclusion of competing etiologies. Clinical information and histologic data were collected at the time of their first AIH diagnosis via medical chart review. Cases of DILI (N = 35) were obtained from the Spanish DILI registry (Málaga, Spain), Hospital Clinic (Barcelona, Spain), and Mayo Clinic. The diagnosis of DILI was made at each site using a standardized causality assessment, the Roussel Uclaf Causality Assessment Method (RUCAM), and/or based on clinical judgment by experienced hepatologists. When the RUCAM was used, only cases assessed as highly probable or probable were included for this study. DILI cases were then subclassified into two categories using the criteria of the International Consensus Meeting for Liver Injury applied to the laboratory data at their first visit with liver injury events12: hepatocellular injury (HC) (n = 19) and cholestatic/mixed injury (CS) (n = 16). Cases with mixed injury were included in the category of CS. The 35 DILI cases included 7 cases (6 HC and 1 CS) of DI-AIH. The diagnosis of DI-AIH was made based on the international criteria and/or the recent simplified criteria for the diagnosis of autoimmune hepatitis (probable or definite).11, 13 Relevant clinical information at time of diagnosis was collected from each site. The study was approved by the Mayo Institutional Review Board, and written informed consent was obtained from all patients for participation in medical research.
The 63 biopsy slides were evaluated concomitantly by four experienced hepatopathologists (E.M.B., D.E.K., R.M., and T.C.S.) blinded to the clinical information using a standardized histologic scoring sheet. Scoring was done by consensus, but the final diagnostic categorization (i.e., DILI, A.I.H., or indeterminate) was recorded individually. Features listed in Table 1 were scored and recorded in the standardized scoring sheets. The presence or absence of chronic hepatitic patterns and acute hepatitic patterns was recorded to characterize the cases based on their necroinflammatory patterns.8 Degree of inflammation and fibrosis was graded according to the Ishak scoring system.14 AIH typical histological findings (i.e., rosettes and emperipolesis) were also recorded.11 To assess the character of the inflammatory cell types (i.e., neutrophils, eosinophils, lymphocytes, and plasma cells), prominent infiltrates were those that were readily noted on low power examination. On higher power evaluation, the type and extent of each inflammatory cell type was confirmed and the more prominent cell types were recorded separately for portal tracts and parenchyma.
Table 1. Histologic Features Evaluated in This Study
Prominent infiltrates were defined as those that were readily noted on low power examination.
Prominent inflammatory cell types in intra-acinar areas*
Yes or no
Yes or no
Yes or no
Yes or no
Yes or no
Yes or no
Yes or no
Yes or no
Yes or no
Yes or no
Yes or no
Bile duct injury
Yes or no
Yes or no
Yes or no
Yes or no
After the histologic evaluation, agreement/disagreement among the four pathologists was recorded for histologic diagnosis (i.e., DILI, AIH, or indeterminate), and the rate of concordance was calculated as number of pathologists in agreement divided by 4. The final histopathologic diagnosis was determined based on the diagnosis agreed upon with the highest number of pathologists.
Data are reported as mean ± standard deviation for continuous variables or proportion of patients with a condition. We described the concordance rate between clinicopathologic and histologic diagnosis and rate of agreement on histologic diagnosis among the four pathologists in three groups, AIH, DILI (HC), and DILI (CS), and one subgroup (DI-AIH). Ishak scores and other histologic features were compared between (1) AIH versus DILI-HC, (2) AIH versus DILI-CS, and (3) AIH versus DI- AIH, using Wilcoxon rank-sum, chi-square, or Fisher's exact tests.
Multiple logistic regression models to predict (1) DILI (HC) cases versus AIH and (2) DILI (CS) cases versus AIH were developed. Variables associated with P < 0.2 in the univariate analyses were considered in the models. When significant colinearity (r > 0.5) existed between variables, one of the variables was included in the model at a time. Final models were determined by backward elimination. Performance of the prediction models was assessed by an area under the receiver operating characteristic (ROC) curve (AUROC).
For analyses, we used JMP statistical software (version 7.0; SAS Institute Inc., Cary, NC), and differences were considered statistically significant when the P values were less than 0.05. All P values presented are two-sided and have not been adjusted for multiple comparisons.
Clinical characteristics of patients with AIH and DILI are summarized in Table 2A,B. Mean age and female gender of the AIH versus DILI population were 49 ± 18 versus 52 ± 18 years (P = 0.516) and 71% female versus 49% female (P = 0.07). Incriminated drugs in the 35 DILI cases are also listed in Table 2B. Median duration of exposure to the incriminated drugs was 29 days, ranging from less than 24 hours to 4 years. Among the 28 AIH cases, 15 (53.6%) had serum bilirubin levels greater than 3.0 mg/dL at baseline.
Table 2. Clinical Characteristics of AIH Cases and DILI Cases
Summary Statistics (n = 28)
Six cases were HC, whereas 1 case was CS.
Median (25th, 75th).
Reference ranges of serum ALT, AST, and total bilirubin were: 7-55 U/L (males) and 7-45 U/L (females) for ALT; 8-48 U/L (males) and 8-43 U/L (females) for AST; and 0.1-1.0 mg/dL for total bilirubin. ALP reference range is age- and gender-dependent, and not provided.
Concordance Rate of Clinicopathologic Versus Histologic Diagnosis.
Data are summarized in Table 3. Overall, 65% of the cases (41) were concordantly diagnosed: 71.4% (20) of AIH cases and 60.0% (21) of DILI cases. There was no significant difference in concordance rate for AIH cases versus that for DILI cases (chi-square test; P = 0.45). In contrast, 22 cases (35%) were diagnosed as indeterminate or were suspected of being something other than AIH or DILI. Three (15.8%) DILI (HC) and 2 (12.5%) DILI (CS) cases were diagnosed as AIH. No AIH cases were diagnosed as DILI.
Table 3. Agreement Between Clinical and Histological Diagnosis for AIH, DILI (HC), and DILI (CS)
Seven cases, 6 DILI (HC) and 1 DILI (CS), were separately analyzed as a subgroup.
In the subgroup of DI-AIH cases (N = 7), 4 (57.1%) were histologically diagnosed as AIH, whereas 2 (28.6%) and 1 (14.3%) were diagnosed as DILI and indeterminate, respectively.
Agreement Among Pathologists on Histologic Diagnosis.
Overall, complete agreement on histologic diagnosis among the four pathologists was found in 46% (29 of the 63 cases). Individually, complete agreement in AIH, DILI (HC), and DILI (CS) cases were 46.4%, 42.1%, and 50.0%, respectively, and no significant difference was noted among the three groups (chi-square test; P = 0.89).
In the subgroup of DI-AIH, there was complete agreement on the histologic diagnosis in 2 of 7 cases. The rate of complete agreement in DI-AIH cases was lower than that in non-autoimmune DILI (n = 28) cases, although it did not reach statistical significance (28.6% versus 50.0%, Fisher's exact test; P = 0.42).
Comparison of Histologic Features Between AIH and DILI Cases.
Data of the comparison between AIH versus DILI (HC) and DILI (CS) are summarized in Table 4. Histologic pictures of representative cases are shown in Figures 1 and 2. Significant overlap of histologic findings was observed for AIH and DILI (HC). Interface hepatitis, focal necrosis, and portal inflammation were present in all cases of AIH and DILI (HC), but were more severe in AIH (P < 0.05). Portal and intra-acinar plasma cells, rosette formation, and emperiopolesis were features that favored AIH (P < 0.05) (Table 4). Prominent intra-acinar eosinophils tended to be more frequent in AIH (P = 0.086).
Table 4. Comparison of Histologic Features/Scores Between AIH Versus DILI (HC)
(N = 28)
(N = 19)
P Value Versus AIH
(N = 16)
P Value Versus AIH
P values are from chi-square tests (or Fisher's exact tests*) or Wilcoxon rank-sum tests.
Significant overlap of histologic findings was also noted in the comparison of AIH versus DILI (CS) (Table 4). However, AIH more frequently had an “acute hepatitic pattern” of injury and was associated with higher scores of interface hepatitis, confluent necrosis, focal necrosis, and portal inflammation (all P ≤ 0.05). Interface hepatitis (P = 0.014), rosette formation (P = 0.005), emperiopolesis (P = 0.014), portal plasma cells (P = 0.0001), and intra-acinar plasma cells (P = 0.002) were more prevalent in AIH. Portal neutrophils (P = 0.019) and hepatocellular cholestasis (P = 0.019) were more prevalent in DILI (CS).
Combinations of the Histologic Features in Predicting DILI Versus AIH.
Prediction models using combinations of the histologic features were developed as described before and are shown in Table 5A,B. A model combining portal inflammation, portal plasma cells, intra-acinar lymphocytes and eosinophils, rosette formation, and canalicular cholestasis yielded an AUROC of 0.90 in predicting DILI (HC) versus AIH. The presence of prominent intra-acinar lymphocytes and canalicular cholestasis favored the diagnosis of DILI, whereas more severe portal inflammation, portal plasma cells, intra-acinar eosinophils, and rosette formation favored the diagnosis of AIH. On the other hand, a model combining portal inflammation, focal necrosis, fibrosis, portal neutrophils, and intracellular (hepatocellular) cholestasis yielded an AUC of 0.90 in predicting DILI (CS) versus AIH, whereas a model replacing focal necrosis with the presence of prominent portal plasma cells yielded an AUC of 0.91. In these models, portal neutrophils and intracellular (hepatocellular) cholestasis favored the diagnosis of DILI (CS), whereas all other features favored the diagnosis of AIH.
Table 5. Models to Predict DILI (HC) Versus AIH and DILI (CS) Versus AIH
A. Model for DILI (HC) Versus AIH
AUROC = 0.90
Ishak scores of portal inflammation were dichotomized in the models (>=2, yes or no).
Ishak scores of fibrosis were dichotomized in the model (>=1, yes or no).
Ishak scores of focal necrosis were dichotomized in the model (>=4, yes or no).
Abbreviations: DILI, drug-induced liver injury; HC, hepatocellular injury; AIH, autoimmune hepatitis; CS, cholestatic/mixed injury; AUROC, area under the receiver operator characteristics curve; SE, standard error.
Subgroup Analysis of Autoimmune Hepatitis Versus DI-AIH.
We analyzed the 7 cases of DI-AIH versus AIH to characterize potential discriminating histologic features (Supporting Table 1). This small subgroup analysis showed that severity of inflammation and fibrosis (Ishak scores) and frequency of AIH-specific findings were comparable between AIH and DI-AIH. However, marked bridging fibrosis (Ishak score, ≥4) was only observed in AIH cases (14.8%). Prominent intra-acinar lymphocyte infiltration tended to be more frequently observed in DI-AIH (17 of 28 versus 7 of 7, Fisher's exact test; P = 0.07). Canalicular/intracellular (hepatocellular) cholestasis was observed in some of the AIH cases (3 of 28 and 2 of 28, respectively).
We performed a standardized, detailed histologic evaluation of well-characterized idiopathic AIH and DILI cases in a blinded manner to delineate discriminating histologic features and determine the role of histologic evaluation in this differential diagnosis. Although there is an overlap of histologic findings for AIH and DILI, we show that pathologists can utilize combinations of features to suggest the correct diagnosis, even in the absence of clinical information. Further, the results of our preliminary modeling effort suggested that, once independently validated in a larger study, models using the combined information may assist histologic differential diagnosis and provide a simple, standardized way to calculate an evidence-based probability for each disease.
We documented the daunting histologic overlap between AIH and DILI. Not surprisingly, a chronic hepatitic pattern, as opposed to an acute hepatitis pattern, was commonly seen in AIH. But, as previously suggested,8 a chronic hepatitic pattern was also more frequent than an acute pattern in both types of DILI (i.e., HC and CS). Similarly, histologic findings often cited as “typical” of AIH were also observed in a significant proportion of DILI cases, including interface hepatitis (89%), emperipolesis (34%), and rosette formation (40%). Thus, though some findings are typical of AIH,11 they are clearly not pathognomonic.
In the comparison of AIH versus DILI (HC), the main findings of chronic hepatitic pattern, interface hepatitis, and portal inflammation were present in all cases with AIH and DILI (HC), but were generally more severe in AIH. Prominent eosinophil infiltration, which has been regarded as one of the histologic findings suggesting DILI,8 is not useful in the differential diagnosis with AIH. Our standardized comparative histologic evaluation demonstrated that the prevalence of prominent eosinophil infiltrates in portal and intra-acinar areas was, in fact, higher among AIH cases versus DILI cases (both in HC and CS). Although the difference was not statistically significant in univariate analyses, a prominent intra-acinar eosinophil infiltrate was one of the predictors in our multivariable analysis that favored AIH over DILI (HC). Intra-acinar lymphocytes and canalicular cholestasis favored DILI (HC) in the same model, whereas rosette formation, portal plasma cells, and relatively severe portal inflammation favored AIH. This suggests that the combined observation of cholestasis, inflammation (i.e., types of infiltrates and area), and degree of inflammation (i.e., Ishak scores) can help discriminate AIH versus DILI (HC). Based on the models, having (1) prominent intra-acinar lymphocyte infiltrate or canalicular cholestasis without the four AIH-favorable features (i.e., portal inflammation score of >=2, rosette formation, prominent portal plasma infiltrates, and prominent intra-acinar eosinophils) or (2) the combination of prominent intra-acinar lymphocyte infiltrate, canalicular cholestasis, and less than two of the four AIH-favorable features is more likely to be DILI (HC) than AIH.
In comparing AIH to DILI (CS), a chronic hepatitic pattern was frequently seen in both conditions. The acute hepatitic pattern of injury was less common in DILI-CS, and AIH cases typically had higher scores for confluent necrosis, focal necrosis, and fibrosis. Models including fibrosis, portal inflammation, intracellular (hepatocellular) cholestasis, prominent portal neutrophil infiltrates, and focal necrosis (or portal plasma cell infiltrates) discriminated our cases of DILI (CS) from those with AIH. Histologic features favoring AIH and DILI, based on the models, are summarized in Table 6.
Table 6. Histologic Features Favoring AIH Versus DILI
Given the observed concordance rate between histologic and clinical diagnosis (65%) and the complete agreement rate among the pathologists (46%), diagnostic models (AUROC of 0.90-0.91 in this study) may be beneficial in the clinicopathologic diagnosis of AIH versus DILI by providing objective probabilities for each condition based on the combinations of histologic features.
We identified no histologic features discriminating AIH versus DI-AIH in the small subgroup analysis. Previous larger studies that compared AIH versus DI-AIH reported that histologic findings in AIH and DI-AIH were similar.1, 15 However, it still might be possible that detailed standardized histologic evaluation, using a larger cohort of DI-AIH versus AIH, could identify histologic features that could be helpful in the differential diagnosis. Interestingly, in the previous study, cirrhosis was observed in 21% of AIH cases, whereas no cirrhosis was present among DI-AIH cases.1 Our findings were consistent with this, in that advanced fibrosis (i.e., marked bridging fibrosis) was observed only among AIH cases, but not DI-AIH. As observed in our study, histologic diagnosis of DI-AIH versus AIH is particularly difficult; the concordance rate with clinical diagnosis was 28.6% (versus 65% in the overall cohort), and the complete agreement rate among the pathologists was also 28.6% (versus 46% in the overall cohort and 50% in non-autoimmune DILI). As a correct diagnosis of DILI-AIH is essential for proper clinical decision making,1 we would suggest that the histology of DI-AIH should continue to be systematically evaluated to indentify characteristics that distinguish DI-AIH from AIH.
Our findings suggest that different inflammatory cells may be enhanced in DILI versus AIH cases; prominent portal neutrophil infiltrate was favoring for DILI (CS), whereas prominent intra-acinar eosinophil infiltrate and prominent portal plasma cell infiltrate were favoring for AIH. Implication and underlying mechanisms of these findings remain unknown. It is intriguing to note that different subsets of lymphocytes and chemokines are involved in eosinophil and plasma cell versus neutrophil recruitment (Th2 versus Th17 response).16, 17 Further studies are warranted to investigate differential cytokine/chemokine profiles in AIH versus DILI.
Our study has limitations. First, our sample size was rather small, and its statistical power was not sufficient in the subgroup analysis. Because we were primarily interested in histological changes, we did not evaluate interobserver variation on each histologic feature. Second, there is a possibility that some of the observed histologic features may have been influenced by a set of the drugs incriminated in the included DILI cases and/or clinical presentation of AIH (i.e., acute vs. chronic presentation). Last, our preliminary modeling efforts were only based on statistical results of this small sample without cross-validation.
In summary, this study showed that no single feature was indicative of AIH or DILI, but rather the combination of distinct findings, such as the types of inflammatory cells in different areas, severity of injury/inflammation, and presence of cholestasis, was very helpful in differentiating DILI versus AIH. The fact that there was more concordance than discordance between clinical and histologic diagnoses suggests that the pathologists were already using the pattern of injury to direct their diagnosis. Further investigation is warranted to validate our findings and to best incorporate the knowledge into current clinicopathologic diagnosis.
On behalf of the Spanish Group for the Study of Drug-Induced Liver Disease, we would like to thank the following participating clinical centers: from the Hospital Universitario Virgen de la Victoria, Málaga (centro coordinador): R.J. Andrade, M.I. Lucena, M. García-Cortés, C. Stephens, M. Robles-Diaz, A. Fernandez-Castañer, Y. Borraz, E. Ulzurrun, B. Garcia-Muñoz, I. Moreno, and L. Vicioso; from the Hospital Torrecárdenas, Almería: M.C. Fernández, G. Peláez, M.D. Muñoz Sánchez-Reyes, R. Daza, M. Casado, J.L. Vega, F. Suárez, M. Torres, and M. González-Sánchez; from the Hospital de Mendaro, Guipuzkoa: A. Castiella and E.M. Zapata; from the Hospital de Basurto, Bilbao: P. Martínez-Odriozola; from the Hospital de Sagunto, Valencia: J. Primo; and from the Hospital de Laredo, Cantabria: M. Carrascosa.