Berend R Beumer1, Stefan Buettner1, Boris Galjart1, Jeroen L A van Vugt1, Robert A de Man2, Jan N M IJzermans1, Bas Groot Koerkamp3. 1. Erasmus MC Transplant Institute, Department of Surgery Division of HPB & Transplant Surgery, Erasmus MC, University Medical Centre Rotterdam, Rotterdam, the Netherlands. 2. Erasmus MC Transplant Institute, Department of Gastroenterology and Hepatology, Erasmus MC, University Medical Centre Rotterdam, Rotterdam, the Netherlands. 3. Erasmus MC Transplant Institute, Department of Surgery Division of HPB & Transplant Surgery, Erasmus MC, University Medical Centre Rotterdam, Rotterdam, the Netherlands. Electronic address: b.grootkoerkamp@erasmusmc.nl.
Abstract
BACKGROUND: Many prognostic models for Hepatocellular Carcinoma (HCC) have been developed to inform patients and doctors about individual prognosis. Previous reviews of these models were qualitative and did not assess performance at external validation. We assessed the performance of prognostic models for HCC and set a benchmark for biomarker studies. METHODS: All externally validated models predicting survival for patients with resected HCC were systematically reviewed. After selection, we extracted descriptive statistics and aggregated c-indices using meta-analysis. RESULTS: Thirty-eight validated prognostic models were included. Models used on average 7 (IQR:4-9) prognostic factors. Tumor size, tumor number, and vascular invasion were almost always included. Alpha-fetoprotein (AFP) was commonly incorporated since 2007. Recently, the more subjective items ascites and encephalopathy have been dropped. Eight established models performed poor to moderate at external validation, with a pooled C-index below 0.7; including the Barcelona Clinic Liver Cancer (BCLC) system, the American Joint Committee on Cancer (AJCC) 7th edition, the Cancer of the Liver Italian (CLIP) Program, and the Japan Integrated Staging (JIS) score. Out of 24 prognostic models predicting OS, only 6 (25%) had good performance at external validation with pooled C-indices above 0.7; the Li-post (0.77), Li-OS (0.74), Yang-pre (0.74), Yang-post (0.76), Shanghai-score (0.70), and Wang-nomogram (0.71). Models improved over time, but overall performance and study quality remained low. CONCLUSIONS: Six validated prognostic models demonstrated good performance for predicting survival after resection of HCC. These models can guide patients and doctors and are a benchmark for future models incorporating novel biomarkers.
BACKGROUND: Many prognostic models for Hepatocellular Carcinoma (HCC) have been developed to inform patients and doctors about individual prognosis. Previous reviews of these models were qualitative and did not assess performance at external validation. We assessed the performance of prognostic models for HCC and set a benchmark for biomarker studies. METHODS: All externally validated models predicting survival for patients with resected HCC were systematically reviewed. After selection, we extracted descriptive statistics and aggregated c-indices using meta-analysis. RESULTS: Thirty-eight validated prognostic models were included. Models used on average 7 (IQR:4-9) prognostic factors. Tumor size, tumor number, and vascular invasion were almost always included. Alpha-fetoprotein (AFP) was commonly incorporated since 2007. Recently, the more subjective items ascites and encephalopathy have been dropped. Eight established models performed poor to moderate at external validation, with a pooled C-index below 0.7; including the Barcelona Clinic Liver Cancer (BCLC) system, the American Joint Committee on Cancer (AJCC) 7th edition, the Cancer of the Liver Italian (CLIP) Program, and the Japan Integrated Staging (JIS) score. Out of 24 prognostic models predicting OS, only 6 (25%) had good performance at external validation with pooled C-indices above 0.7; the Li-post (0.77), Li-OS (0.74), Yang-pre (0.74), Yang-post (0.76), Shanghai-score (0.70), and Wang-nomogram (0.71). Models improved over time, but overall performance and study quality remained low. CONCLUSIONS: Six validated prognostic models demonstrated good performance for predicting survival after resection of HCC. These models can guide patients and doctors and are a benchmark for future models incorporating novel biomarkers.