Alcoholic hepatitis is a form of acute liver injury with significant morbidity and mortality. Several scoring systems have been developed and used to predict mortality, including Maddrey's discriminant function (MDF), model for end-stage liver disease (MELD), and modified MELD including sodium (MELD-Na). We conducted a systematic review of Pubmed, Embase, and Cochrane Library to compare the accuracy of these scores in alcoholic hepatitis mortality prediction. The pooled sensitivity and specificity of MDF for mortality prediction in 28 days were 0.898 95% confidence interval (CI) 0.777-0.957, I2 = 0% and 0.370 (95% CI: 0.271-0.482, I2 = 78%), respectively; in 90 days, they were 0.898 (95% CI: 0.799-0.951, I2=21%) and 0.346 (95% CI: 0.246-0.463, I2 = 88%), respectively. The pooled sensitivity and specificity of MELD for mortality prediction in 28 days were 0.862 (95% CI: 0.755-0.926, I2 = 0%) and 0.697 (95% CI: 0.552-0.812, I2 = 82%), respectively; in 90 days, they were 0.884 (95% CI: 0.802-0.935, I2 = 21%) and 0.557 (95% CI: 0.300-0.787, I2 = 96%), respectively. The pooled sensitivity and specificity of MELD-Na for mortality prediction in 28 days were 0.652 (95% CI: 0.429-0.823, I2 = 79%) and 0.757 (95% CI: 0.666-0.830, I2 = 91%), respectively; in 90 days, they were 0.701 (95% CI: 0.661-0.737, I2 = 53%) and 0.816 (95% CI: 0.711-0.888, I2 = 94%), respectively. No significant statistical difference was observed when comparing area under the curve of these scores.
Tschaen et al. (Thu,) studied this question.