Selecting a Short-term Prognostic Model for Hepatocellular Carcinoma

Abstract

The model for end-stage liver disease (MELD) and MELD-sodium (MELD-Na) are prognostic models for cirrhotic patients with or without hepatocellular carcinoma (HCC). This study compared the predictive accuracy between the MELD, MELD-Na, TNM (tumor, node, metastasis), Cancer of the Liver Italian Program (CLIP), Barcelona Clinic Liver Cancer (BCLC), Japan Integrated Scoring (JIS), and Tokyo score for 3-month and 6-month mortality in HCC patients. A total of 953 patients were prospectively enrolled. The predictive accuracy was compared between different models using the area under receiver operating characteristic curve (AUC). The CLIP system had the highest AUC (0.875) for predicting 3-month mortality, followed by the Tokyo score (0.874), JIS (0.868), BCLC (0.855), MELD-Na (0.829), MELD (0.803), and finally, TNM (0.795) system. At 6 months, the top 3 models and their AUCs were the CLIP (0.882), Tokyo (0.861), and JIS (0.85). MELD-Na consistently had significantly better predictive accuracy than the MELD at 3 and 6 months. The MELD and MELD-Na were better prognostic models in predicting the mortality for surgical patients (AUC, 0.719 to 0.740), whereas the CLIP and Tokyo systems were the 2 better models in staging nonsurgical (AUC, 0.849 to 0.884) and high-risk patients (AUC, 0.790 to 0.846), defined as having at least 2 independent risk factors of mortality, at 3 and 6 months. The MELD-Na may improve the prognostic ability of the MELD system for patients with HCC. Both the CLIP and Tokyo systems are better short-term prognostic models. These findings are helpful in designing future clinical trials for HCC.