Presented in part during Digestive Disease Week at the meeting of the American Association for the Study of Liver Diseases, May 20, 2007, Washington, DC.
Potential conflict of interest: Nothing to report.
Autoimmune hepatitis may fail to respond to corticosteroid therapy, but the frequency and bases for this outcome are uncertain. We aimed to determine the frequency and nature of treatment failure in patients with type 1 autoimmune hepatitis, define features associated with its occurrence, and assess if the model for end-stage liver disease can predict this outcome. Patients failing conventional corticosteroid regimens were compared to patients who responded to similar regimens. Fourteen of 214 patients (7%) failed corticosteroid treatment. Patients who failed therapy were younger (33 ± 3 years versus 48 ± 1 years, P = 0.0008), had higher serum levels of bilirubin at accession (4.1 ± 0.9 mg/dL versus 2.3 ± 0.2 mg/dL, P = 0.02), presented acutely more frequently (43% versus 14%, P = 0.01), and had a higher frequency of HLA (human leukocyte antigen) DRB1*03 (93% versus 53%, P = 0.004) than did patients who achieved remission. An alternative disease (fatty liver disease) emerged in only 1 patient who failed therapy (7%). Scores determined by the model of end-stage liver disease at presentation of patients who failed treatment were higher than those of who achieved remission (16 ± 1 versus 10 ± 0.3 points, P < 0.0001), and score greater than 12 points had greater sensitivity (97%) and specificity (68%) for treatment failure than did HLA DRB1*03 or other features. Conclusion: Onset at an early age, acute presentation, hyperbilirubinemia, and presence of HLA DRB1*03 characterize patients who fail corticosteroid treatment. The model for end-stage liver disease may be a useful instrument for identifying patients prone to this outcome. (HEPATOLOGY 2007.)
Corticosteroids are effective in the treatment of most patients with autoimmune hepatitis, but deterioration is still possible despite compliance with therapy.1, 2 High doses of prednisone alone or in combination with azathioprine can improve the clinical and laboratory manifestations of treatment failure, but histological resolution is uncommon.2, 3 These patients typically require indefinite therapy, and they remain at risk for drug toxicity and liver failure. The results of empiric treatment with ursodeoxycholic acid,4 cyclosporine,5, 6 tacrolimus,7, 8 mycophenolate mofetil,9–12 cyclophosphamide,13 methotrexate,14 rituximab,15 budesonide,16 intravenous globulin,17 or leukocytopheresis18 attest to the refractory nature of this condition.
The frequency of treatment failure and the basis for deterioration during corticosteroid therapy are unknown. Corticosteroid responsiveness is a defining feature of autoimmune hepatitis, and failure of the disease to improve suggests an alternative diagnosis.19 Many diseases can resemble autoimmune hepatitis, including Wilson disease, chronic hepatitis C, fatty liver disease, primary biliary cirrhosis (PBC), and primary sclerosing cholangitis (PSC). These conditions may be unrecognized at presentation and account for treatment failure.
Autoimmune hepatitis may also undergo transitions during its course, with a cholestatic syndrome emerging that might be refractory to the original treatment.20 Furthermore, another condition may be superimposed on the original process, such as PSC,21–24 viral infection,25 drug toxicity,26 or fatty liver disease,27 which could alter treatment outcome. Last, treatment failure may reflect intrinsic pathogenic mechanisms promoted by host-dependent genetic predispositions that cannot be suppressed by conventional corticosteroid regimens.28–31 HLA DRB1*03 has been the principal risk factor associated with treatment failure in white North American and northern European patients, but its sensitivity, specificity, and predictability for treatment failure are unknown.28
The model of end-stage liver disease (MELD) is an established method for predicting early mortality associated with severe liver disease.32, 33 It evolved from a model that used serum creatinine level, total serum bilirubin concentration, the international normalized ratio (INR), and the cause of the liver disease to predict early survival in patients with cirrhosis who were candidates for a transjugular intrahepatic portal-systemic shunt (TIPS).34 Its applications were subsequently extended to stratifying patients for urgent liver transplantation35 and to predicting early mortality associated with alcoholic hepatitis,36 hepatorenal syndrome,37 acute liver failure,38 sepsis in cirrhosis,39 surgical procedures in chronic liver disease,40 and acute-on-chronic hepatitis.41 These applications have been possible because survival predicted by the MELD score relates mainly to the degree of impaired liver function, not to the presence of cirrhosis or the cause of the disease. If treatment failure in patients with autoimmune hepatitis relates to intrinsic pathogenic mechanisms that compromise critical liver functions, MELD might be useful in identifying patients who are at risk to fail corticosteroid therapy at presentation.
In this retrospective analysis of prospectively acquired data, we determined the frequency and nature of treatment failure in patients with definite type 1 autoimmune hepatitis, defined the features associated with treatment failure, assessed the emergence of alternative or superimposed conditions that might alter treatment response, and determined the ability of MELD alone and in combination with HLA DRB1*03 to identify patients at risk for treatment failure.
Patients and Methods
Two hundred and fourteen patients who satisfied the international pretreatment criteria for the definite diagnosis of autoimmune hepatitis19 constituted the study population. All patients had been referred to the Autoimmune Hepatitis Treatment Program of the Mayo Clinic, and they had been evaluated by 1 investigator (A.J.C.). Each had been treated with conventional corticosteroid regimens, had fulfilled preestablished clinical, laboratory, and histological criteria for having experienced either treatment failure or remission, and had been systematically monitored according to a previously published protocol.42
One hundred seventy-four patients (81%) were female, and the mean age at presentation was 47 ± 1 years (median, 49 years; range, 13–82 years). One hundred forty-seven patients (69%) had been treated with prednisone in combination with azathioprine, and 67 patients (31%) had received a higher dose of prednisone alone. It was previously shown that the 2 treatment regimens were of comparable efficacy.43
All patients satisfied pretreatment scoring criteria for a definite diagnosis of autoimmune hepatitis (mean score, 18 ± 0.1 points; median, 18 points, range, 16–24 points) and had smooth muscle antibodies (65 patients), antinuclear antibodies (56 patients), or both (93 patients) at accession. One hundred ninety-six patients tested for antibodies to liver kidney microsome type 1 (anti-LKM1) were negative for this marker. The serological findings justified the designation of type 1 autoimmune hepatitis.44 Our investigation was approved by the Institutional Review Board of the Mayo Clinic.
Clinical and Laboratory Assessments.
Conventional laboratory tests of liver inflammation and function were assessed at each examination; these included serum aspartate aminotransferase (AST), bilirubin, γ-globulin, and immunoglobulin levels. Concurrent extrahepatic disorders of an immune nature were sought in a systematic fashion during each evaluation in a fashion reported previously.42 Patients with ulcerative colitis underwent cholangiography to assess the possibility of PSC, and only those patients with normal cholangiograms were included in the study.21
Smooth muscle antibodies were determined in all patients by indirect immunofluorescence on murine tissue sections, as described elsewhere.45 Antinuclear antibodies were assessed in 168 patients by indirect immunofluorescence on murine tissue sections46 and in 46 patients by a previously reported enzyme-linked immunosorbent assay (ELISA) based on recombinant nuclear antigens.47 Antibodies to liver kidney microsome type 1 were evaluated by indirect immunofluorescence on murine tissue sections.48
A serum titer of at least 1:40 was considered positive for SMA and ANA, and a serum titer of at least 1:10 was considered positive for anti-LKM1, both determined by indirect immunofluorescence; a serum level greater than 1.0 units was considered positive for ANA as determined by ELISA. The MELD score at presentation was calculated according to the formula 10[0.957 ln (serum creatinine) + 0.378 ln (total bilirubin) + 1.12 ln (INR) + 0.643].33
Examination of liver tissue of the 214 patients at accession showed that 49 (23%) had cirrhosis. Histological examination of liver tissue of the patients who had features of treatment failure was similarly performed. All tissue specimens were examined by a liver pathologist who was a member of the liver pathology working group at the Mayo Clinic. Histological findings of autoimmune hepatitis and features inconsistent with that diagnosis, including biliary changes and fatty infiltration, were recorded in accordance with a standardized report adopted by the liver pathology working group. In this fashion, transitions from the original diagnosis and superimposed diseases accounting for treatment failure were assessed.
Inflammatory activity (grade) and fibrosis (stage) were assessed on a scale of 0–4 in accordance with the system of Batts and Ludwig.49 Histological interpretation was done prospectively and included in our database for each patient. The findings were then analyzed retrospectively for this study. A diagnosis of cirrhosis required demonstration of fibrosis and a complete regenerative nodule.50
Our previous validation studies have indicated high consistency of pathological interpretation for grade (90%) and stage (78%) of autoimmune hepatitis using preestablished definitions for the histological findings. The reproducibility of our histological interpretation has been 94%, and intraobserver consistency in diagnosing cirrhosis has been 88%.51–53
Definitions of Treatment Response and Courses of Action.
“Treatment failure” was defined as deterioration during therapy, characterized by a progressive increase in serum AST or bilirubin at presentation.42 Its occurrence justified clinical reassessment to exclude noncompliance with therapy, alcohol consumption, or drug toxicity. Laboratory studies were performed to exclude a superimposed viral infection or a transition state to PBC or PSC, and liver tissue was examined to reconfirm the original diagnosis and to assess alternative possibilities, including fatty infiltration, PBC, or PSC. Confirmation of treatment failure associated with the original diagnosis justified increased doses of medication, either prednisone, 30 mg daily, in conjunction with azathioprine, 150 mg daily, or prednisone alone, 60 mg daily, in accordance with our previously published treatment failure protocol.3, 42 “Remission” was defined as the absence of symptoms, improvement in serum AST level to normal or near normal, and histological features of minimal or no inflammatory activity. Treatment was then withdrawn in a tapered fashion over a 6-week period.42
One hundred and ninety-four patients (91%) were evaluated for having the class II (DR locus) HLA. DNA-based HLA was determined by restriction fragment-length polymorphism in 86 patients and by polymerase chain reaction with sequence-specific primers in 108 patients using methods reported previously.54
The Fisher exact probability test was used to compare categorical variables, and the unpaired t test was used to compare differences in the means of continuous variables. Nonparametric variables in independent samples were compared with the Mann-Whitney test. A receiver-operating characteristics (ROC) curve was plotted to measure how the MELD score performed in predicting treatment failure in patients with type 1 autoimmune hepatitis. The validity of the model was measured by the area under the curve.55 Furthermore, the ROC curve was used to determine the MELD score cutoff that discriminated between patients who developed treatment failure and those who achieved remission.
The frequency of HLA DRB1*03 in patients who experienced treatment failure was compared with that in patients who did not, and the predictive value of combining MELD score with HLA DRB1*03 status was assessed. All variables for comparison had been formulated a priori and then were assessed systematically in the study population, and an unadjusted P value of less than 0.05 was considered statistically significant. Data in tables and text are presented as the mean ± standard error of the mean (SEM).
Treatment Failure and Features at Presentation.
Fourteen patients (7%) satisfied criteria for having experienced treatment failure during the first course of therapy. Worsening of laboratory indices occurred 5 ± 1 months after conventional corticosteroid treatment was begun (range, 3–12 months; median, 3 months). All patients showed an increase in serum AST level, and 11 patients (79%) also showed an increase in serum bilirubin level. Mean serum AST level at treatment failure was 1,463 ± 300 U/L (range, 180–3,814 U/L; median, 1,200 U/L), and mean serum bilirubin concentration at treatment failure was 8.6 ± 2 mg/dL (range, 0.9–27 mg/dL; median, 3.9 mg/dL). The serum AST level had increased by 70%–452% of the pretreatment value (median increase, 127%), and the serum bilirubin level had increased by 0%–210% of the pretreatment value (median increase, 79%).
Patients who failed therapy were younger (33 ± 3 years versus 48 ± 1 years, P = 0.0008), had a higher serum level of bilirubin at accession (4.1 ± 0.8 mg/dL versus 2.3 ± 0.2 mg/dL, P = 0.02), more frequently showed acute presentations (43% versus 14%, P = 0.01), and had a higher frequency of HLA DRB1*03 (93% versus 53%, P = 0.004) than patients did who achieved remission. MELD score at presentation of patients who failed treatment was also higher than that of those who achieved remission (mean score, 16 ± 1 versus 10 ± 0.3, P < 0.0001; Table 1).
Table 1. Features at Presentation Associated with Treatment Failure
Features at presentation
Treatment failure (n = 14)
Remission (n = 200)
Significantly different from each other at
P < 0.0001,
P = 0.0008,
P = 0.01,
P = 0.02. Numbers in parentheses are percentages.
Duration of symptoms ≤ 3 months before diagnosis. Abbreviations: AST, aspartate aminotransferase; IgG, immunoglobulin G; IAHG, International Autoimmune Hepatitis Group; MELD, model for end-stage liver disease.
The odds ratio (OR) of patients with HLA DRB1*03 experiencing treatment failure was 11 (95% confidence interval [CI], 1.5–88.8, P = 0.01; Table 2). The sensitivity, specificity, positive predictive value, and negative predictive value of HLA DRB1*03 for treatment failure were 93% (95% CI, 69%–99%), 47% (95% CI, 40%–54%), 12% (95% CI, 7%–19%), and 99% (95% CI, 94%–99%), respectively (Table 3).
Table 2. Absolute Frequencies and Odds Ratios of Features Associated with Treatment Failure
Feature associated with treatment failure
Treatment failure (n)
Significantly different from each other at *
P < 0.0001
P = 0.01. Abbreviations: HLA, human leukocyte antigen; MELD, model of end-stage liver disease; OR, odds ratio; CI, confidence interval.
Table 3. Performance Parameters of Features Associated with Treatment Failure
Prognostic parameter for treatment failure
Positive predictability (%)
Negative predictability (%)
Abbreviations: HLA, human leukocyte antigen; MELD, model of end-stage liver disease. Numbers in parentheses are 95% confidence intervals.
MELD ≥ 12 points
MELD ≥ 12 points and cirrhosis
MELD ≥ 12 points and no cirrhosis
MELD ≥ 12 points and HLA DRB1*03
MELD ≥ 12 points, HLA DRB1*03 and cirrhosis
MELD and Treatment Failure.
The MELD score as a predictor of treatment failure was determined by assessing the area under the ROC curve (Fig. 1). The area under the curve (AUC) was 0.87 (standard error [SE], 0.03; P < 0.0001; 95% CI, 0.78–0.94). The best cutoff for the MELD score was 12 points. The 14 patients who subsequently had treatment fail had a MELD score of at least 12 points at accession, compared to 63 of the 200 patients who achieved remission (100% versus 32%, P < 0.0001). Patients with a MELD score of at least 12 points had an OR of 63 (95% CI, 3.7–1,069; P < 0.0001) of experiencing treatment failure (Table 2). The sensitivity of a MELD score greater than 12 points for treatment failure was 97% (95% CI, 75%–99%), the specificity was 68% (95% CI, 62%–74%), the positive predictive value was 19% (95% CI, 12%–29%), and the negative predictive value was 99% (95% CI, 97%–100%), shown in Table 3.
Individual Components of the MELD Score and Treatment Failure.
The AUC for creatinine was 0.57 (SE, 0.08; P = 0.4; 95% CI, 0.41–0.73), and the best cutoff for creatinine was 1.1 mg/dL. Six of the 14 patients (43%) who subsequently failed therapy had a creatinine level greater than 1.1 mg/dL at accession, compared to 45 of the 200 patients (22%) who achieved remission. The sensitivity of a serum creatinine level greater than 1.1 mg/dL for treatment failure was 43% (95% CI, 21%–67%), its specificity was 78% (95% CI, 71%–83%), its PPV was 12% (95% CI, 6%–23%), and its NPV was 95% (95% CI, 91%–98%), shown in Table 4.
Table 4. Performance Parameters of Individual Components of the MELD Score
Prognostic parameter for treatment Failure
Positive predictability (%)
Negative predictability (%)
Abbreviations: MELD, model of end-stage liver disease; INR, international normalized ratio. Numbers in parentheses are 95% confidence intervals.
Creatinine ≥ 1.1 mg/dL
INR ≥ 1.2
Bilirubin ≥ 1.8 mg/dL
MELD ≥ 12 points
The AUC for the INR was 0.74 (SE, 0.07; P = 0.003; 95% CI, 0.61–0.87), and the best cutoff value was 1.2. Twelve of the 14 patients (86%) who subsequently developed treatment failure had an INR greater than 1.2, compared to 78 of the 200 patients (39%) who achieved remission. The sensitivity of an INR greater than 1.2 for treatment failure was 86% (95% CI, 60%–96%), its specificity was 61% (95% CI, 54%–68%), its PPV was 13% (95% CI, 8%–22%), and its NPV was 98% (95% CI, 94%–99%), shown in Table 4.
The AUC for serum bilirubin level was 0.75 (SE, 0.05; P = 0.002; 95% CI, 0.64–0.85), and the best cutoff value was 1.8 mg/dL. Twelve of the 14 patients (86%) whose treatment subsequently failed had a serum bilirubin level greater than 1.8 mg/dL, compared to 79 of the 200 patients (40%) who achieved remission. The sensitivity of a serum bilirubin level greater than 1.8 mg/dL for treatment failure was 86% (95% CI, 60%–96%), its specificity was 61% (95% CI, 54%–67%), its PPV was 13% (95% CI, 8%–22%), and its NPV was 98% (95% CI, 94%–99%), shown in Table 4.
The performance parameters of the individual components of the MELD score were less than those of the composite score with only 1 exception. A serum creatinine level greater than 1.1 mg/dL at presentation was more specific for treatment failure than were the other components and the composite score (Table 4).
MELD and Cirrhosis.
Three of the 14 patients who failed treatment (21%) had cirrhosis at presentation, and this frequency was similar to the 23% of patients who achieved remission (Table 1). The performance parameters of the MELD score for predicting treatment failure of patients with cirrhosis at presentation were similar to those of patients without cirrhosis at presentation (Table 3).
MELD and HLA DRB1*03 Status.
One hundred and nine of the 194 patients who were typed for class II HLA (56%) had HLA DRB1*03. Thirteen of the 14 patients who developed treatment failure had HLA DRB1*03, each of whom had a MELD score of at least 12 points (range, 12–26 points; median score, 15 points). Patients whose treatment failed had a significantly greater frequency of HLA DRB1*03 and of a MELD score of at least 12 points than did those who entered remission (93% versus 19%, P < 0.0001). The combination of HLA DRB1*03 and a MELD score of at least 12 points had a sensitivity for treatment failure of 93% (95% CI, 69%–99%), a specificity of 81% (95% CI, 74%–86%), a positive predictive value of 27% (95% CI, 17%–41%), and a negative predictive value of 99% (95% CI, 96%–100%), shown in Table 3.
Investigation of the histological features at treatment failure disclosed 13 of the 14 patients had moderate to severe interface hepatitis, findings consistent with the original diagnosis of autoimmune hepatitis (Table 5). Only 1 patient who failed therapy had developed another condition (fatty liver disease) that may have affected the treatment response. Most patients who failed therapy (93%) did not have alternative diagnoses, and 8 of the 10 patients with ulcerative colitis and normal cholangiograms at presentation responded to treatment. Fibrosis had progressed or developed in 7 of the 11 patients without cirrhosis at presentation (Table 5).
Table 5. Histological Features Associated with Treatment Failure
Findings at presentation
Findings at treatment failure
Numbers in parentheses are percentages.
Outcomes after Treatment Failure.
Ten patients (71%) had resolution of the abnormalities found in laboratory tests after treatment with high-dose prednisone alone (2 patients) or in combination with high-dose azathioprine (8 patients). Eight subsequently deteriorated when the doses of medication were reduced. Four patients never improved despite high-dose therapy with prednisone alone (2 patients) or in combination with azathioprine (2 patients). Nine of the 11 patients without cirrhosis at accession (82%) developed it; 3 of the 14 patients (21%) subsequently required liver transplantation; and 4 patients (29%) died, 2 from liver failure, 1 from variceal hemorrhage, and 1 from hepatocellular carcinoma, during the 95 ± 20 months of observation (median, 64 months; range, 23–276 months). Only 2 of the 14 patients (14%) who failed the initial treatment entered remission after high-dose therapy.
Our study indicates that 7% of patients with type 1 autoimmune hepatitis deteriorate during corticosteroid therapy and that treatment failure is associated with onset at an early age, acute presentation, hyperbilirubinemia, HLA DRB1*03, and a MELD score of at last 12 points. We could not ascribe the treatment failure to transition of the autoimmune hepatitis to another condition, and only 1 patient developed fatty liver disease, in whom treatment failure might have related to a superimposed or concurrent disease. Treatment failure was associated with a poor prognosis despite high-dose corticosteroid treatment, 7 patients (50%) either died of liver-related causes or required transplantation, and only 2 patients (14%) achieved remission.
Previous studies have indicated the importance of using disease severity, shown by histological pattern and laboratory features, to identify patients who have a poor prognosis if they are untreated.56 These findings, however, have not reliably identified those who will have a poor outcome after treatment.57 The prognosis for autoimmune hepatitis has been related mainly to the response to corticosteroid treatment, and failure of laboratory results to improve during therapy has been the principal indication of a poor outcome.57 The MELD score now provides an important new dimension to evaluating prognosis, as it allows the identification of patients at risk for treatment failure before treatment.
The MELD score was originally developed to predict early mortality in patients with end-stage liver disease, and the score was adjusted according to the etiology of the liver disease (cholestatic or alcoholic versus another cause).34 Because the score was intended to estimate survival of patients undergoing the TIPS procedure, it was initially applied to patients with cirrhosis and portal hypertension. Subsequently, the MELD score was found to be useful in estimating survival of patients with chronic liver disease of diverse causes and of patients who do not have cirrhosis and portal hypertension.36–41
In autoimmune hepatitis, a MELD score of at least 12 points at presentation identified 97% of patients who failed corticosteroid treatment (Table 3). The specificity of this score was 68%, and the positive predictive value was 19%. Other MELD scores had greater specificity for treatment failure but lower sensitivity (Fig. 1). Our desire to capture the patients at risk for treatment failure at presentation justified selecting a MELD score with greater sensitivity than specificity. Patients so identified could then be monitored more closely or could be considered for salvage therapies at the earliest sign of deterioration. Such therapies have consisted mainly of high-dose corticosteroid regimens, but their lack of efficacy in this study underscores the need for better salvage therapies for autoimmune hepatitis.58
The performance parameters of the individual components of the MELD score were uniformly lower than those of the composite score except for serum creatinine level (Table 4). A serum creatinine level greater than 1.1 mg/dL at presentation was more specific for treatment failure than the other components (78% versus 61%) and than the composite score (78% versus 68%). However, its low sensitivity for treatment failure compared to the other components (43% versus 86%) and the composite score (43% versus 97%) diminished its clinical value. Serum bilirubin level, which was higher at presentation in the patients who subsequently failed treatment, did not have performance parameters that distinguished it from the INR or composite score (Table 4).
The association of the MELD score with poor outcome in alcoholic hepatitis, acute liver failure, and acute-on-chronic hepatitis and its ability to identify patients in need of liver transplantation suggest that it reflects disease severity or other factors responsible for rapid disease progression.32–41 Cirrhosis at presentation was not associated with treatment failure, and the MELD score was similarly predictive of failure of treatment in patients regardless of whether they had cirrhosis at accession (Table 3). The associations of onset at an early age, acute presentation, and HLA DRB1*03 with treatment failure suggest that host-related genetic factors influence immediate treatment outcome and that the MELD score can predict this response better than conventional clinical assessments can.
HLA DRB1*03 is the principal susceptibility risk factor for type 1 autoimmune hepatitis in white North American and northern European adults, and it has been associated with treatment failure.28 The sensitivity of HLA DRB1*03 for treatment failure was 93%, but its specificity for this outcome was only 47% (Table 3). Nineteen percent of normal white North American adults have HLA DRB1*03. This high frequency in a normal population weakens its specificity for treatment failure.28 The combination of a MELD score of at least 12 points and HLA DRB1*03 improved the specificity of HLA DRB1*03 for treatment failure to 81% but lessened its sensitivity to 93% (Table 3). The performance parameters associated with the MELD score alone were superior to those of HLA DRB1*03 alone or in combination with the MELD score (Table 3). Furthermore, the MELD score is easier to obtain and less expensive than HLA typing.
The reason that the MELD score has predictive value in autoimmune hepatitis is unknown. The score is based on key functions that can be affected by severe inflammatory activity or by indolent but progressive inflammation that erodes these functions.32–34 In this fashion, MELD may be an assessment that accounts for multiple deficiencies in hepatic performance that cannot be reliably assessed by individual conventional tests. These composite deficiencies may influence corticosteroid responsiveness by reflecting certain aggressive pathogenic mechanisms that are incapable of corticosteroid suppression. Future studies that clarify the pathogenic mechanisms of autoimmune hepatitis may be able to distinguish corticosteroid-responsive from corticosteroid-nonresponsive patients in a more objective fashion.
In summary, deterioration during conventional corticosteroid therapy occurs infrequently in patients with definite type 1 autoimmune hepatitis at presentation. Treatment failure is associated with onset at an early age, acute presentation, hyperbilirubinemia, and HLA DRB1*03. Treatment failure does not represent a transition to another disease and is rarely associated with a superimposed alternative condition. A MELD score of at least 12 points identifies most patients at risk of treatment failure, and it justifies closely monitoring patients scoring more than 12 points at presentation. New salvage therapies are warranted, as high-dose corticosteroid regimens have poor immediate and long-term results.