Autoimmune hepatitis (AIH) typically responds to treatment in 90% of patients. Early prediction of treatment outcome would be advantageous in clinical practice. We evaluated whether parameters at initiation of therapy or changes in these parameters at day 3 and day 7 following corticosteroid initiation predicted treatment failure. Treatment-naive, jaundiced patients presenting to our tertiary unit between 1999-2009 were identified and mathematical models of prognosis in liver disease scores calculated at day 0, day 3, and day 7. Overall, 72 patients were identified (48 women, 24 men). Treatment failure occurred in 18% (13/72) of patients. At diagnosis, higher median bilirubin (451 lmol/L versus 262 lmol/L, P 5 0.02), INR (1.62 versus 1.33, P 5 0.005), model for endstage liver (MELD) score (26 versus 20, P 5 0.02), MELD-sodium (Na) score (27 versus 22, P 5 0.03) and United Kingdom endstage liver disease score (UKELD) score (59 versus 57, P 5 0.01) significantly correlated with treatment failure. Analysis of area under the receiver operator characteristic curve (AUROC) values at day 7 identified change (D) bilirubin (AUROC 0.68), D creatinine (0.69), D MELD (0.79), D MELD-Na (0.83) and D UKELD (0.83) best predicted treatment failure. Specifically, a fall in UKELD of less than 2 points predicted treatment failure with a sensitivity of 85% and specificity of 68%. Of 13 treatment failures, nine required second-line immunosuppression, three required emergency transplant, and one died of sepsis. In total, four patients died in the treatment failure group compared with one in the responder group (4/13 5 31% versus 1/59 5 1.7%, P 5 0.003). Conclusion: Approximately 20% of icteric AIH presentations fail corticosteroid therapy. This is associated with significant mortality and the need for emergency transplantation. Treatment failure is best predicted by change in MELD-Na and UKELD at day 7. Early identification of nonresponders may allow timely escalation of immunosuppression to prevent clinical deterioration. (HEPATOLOGY 2011;53:926-934