In patients with alcoholic hepatitis, assessment of mortality risk is a key factor in the clinical management. Within the last decade, several prognostic scores have emerged. We aimed to evaluate and compare the predictive performances of the Glasgow alcoholic hepatitis score (GAHS); the model of end-stage liver disease (MELD); the MELD-Na; the Lille model; and the age, bilirubin, international normalized ratio, and creatinine (ABIC) score in a population-based unselected cohort of patients with alcoholic hepatitis.
We applied the scoring systems to all 274 patients hospitalized and diagnosed with alcoholic hepatitis (1999-2008) in a Danish region of 1.1 million inhabitants. The ability of each score to predict mortality was evaluated using receiver operating characteristics curves, and the area under the receiver operating characteristics curves (AUROCs) were used to compare the scores.
The 28-day mortality was 16%, 84-day 27%, and 180-day 40%. The models had similar predictive properties with AUROC = 0.74-0.78 for 28-day mortality, 0.69-0.77 for 84-day mortality, and 0.65-0.75 for 180-day mortality. There were no statistically significant differences in the models' performances (p > 0.9). Re-scoring on day seven improved the AUROCs of the models (AUROC for 28-day, 0.75-0.83; 84-day, 0.75-0.79; and 180-day, 0.72-0.74).
The MELD, the MELD-Na, the GAHS, the Lille-model, and the ABIC scores each predicted the 28-, 84-, and 180-day mortality of our patients with alcoholic hepatitis and to the same degree, although not as well as in the original cohorts. Re-scoring on day seven improved the models' predictions.