Combining data from liver disease scoring systems better predicts outcomes of patients with alcoholic hepatitis

Abstract

Background & Aims Several models have been used to determine prognoses of patients with alcoholic hepatitis. These include static systems (the Maddrey discriminant function; age, bilirubin, international normalized ratio, creatinine [ABIC] score; and model for end-stage liver disease [MELD] score) and dynamic models (the Lille model). We aimed to combine features of all of these models to develop a better method to predict outcomes of patients with alcoholic hepatitis. Methods We collected data from several databases of patients with severe alcoholic hepatitis treated with corticosteroids in France and the United Kingdom to create a model to predict patient survival (derivation cohort, n = 538 patients). We compared the performances of 3 joint-effect models (Maddrey+Lille, MELD+Lille, and ABIC+Lille) to determine which combination had the best prognostic value, based on known patient outcomes. The model was validated using data from trials of the effects of corticosteroids in patients in the United States, France, Korea, and Belgium (n = 604 patients). Results We created a joint-effect model to predict patient survival after 2 and 6 months; in the derivation and validation cohorts it predicted outcome significantly better than either static or dynamic models alone (P