Different model for end-stage liver disease score block distributions may have a variable ability for outcome prediction.

BACKGROUND The model for end-stage liver disease (MELD) scoring system has become the prevailing criteria for organ allocation in liver transplantation. However, it is not clear if the predictive accuracy of MELD is equally homogeneous in different distribution of MELD score blocks. METHODS We investigated 472 cirrhotic patients (mean MELD, 14.3+/-5.5), and compared the predictive accuracy of MELD and the corresponding Child-Turcotte-Pugh (CTP) scores in patients with low (<16), intermediate (10-20) and high (>14) MELD score range by using c-statistic for area under the receiver operating characteristic curve (AUC) at different time frames. RESULTS The MELD scores well correlated with CTP scores at baseline (rho=0.492, P<0.001). Overall, MELD was significantly better than the CTP system to predict the risk of mortality. However, in stratified analysis there were no significant differences between MELD and CTP for the c-statistic in patients with low and intermediate range MELD scores at 3-, 6-, 9-, and 12-month (p values all > 0 1). Among patients with high MELD scores, MELD was consistently more accurate than the CTP system in predicting the mortality at 3- (AUC, 0.715 vs. 0.543, P=0.020), 6- (0.705 vs. 0.536, P=0.003), 9- (0.737 vs. 0.507, P<0.001) and 12-month (0.716 vs. 0.526, P<0.001), respectively. CONCLUSIONS MELD has a better performance only in a subset of patients with higher MELD scores. The outcome in patients with lower range MELD scores cannot be reliably predicted solely with their MELD scores, and alternative prognostic markers should be used in conjunction to enhance the predictive accuracy.