M2060 Time-Dependent COX Regression Model Is Superior in Prediction of Prognosis in Hepatocellular Carcinoma

Abstract

BACKGROUND/AIMS; The prognostic assessment of hepatocellular carcinoma (HCC) patients using time-fixed Cox regression models have been made up to the present. The aim of this study was to develop the novel prognostic model of HCC by time-dependent Cox regression analysis. METHODS; Five hundred and sixty-three patients with HCC who had initial treatment between January 1995 and June 2005 at Okayama University Hospital were enrolled. Clinical characteristics of the patients were as follows; 396 (70%) patients corresponded to Child-Pugh grade A, 141 (25%) to grade B, and 26 (5%) to grade C. The tumor size was 3.4±2.7 cm in diameter (mean±SD). Tumors were uninodular in 306 (54%) patients. Median follow-up period was 2.5 years. Survival duration was calculated from the date of initial treatment to the date of liver-related death. In standard time-fixed Cox regression model and time-dependent Cox regression model, both 8 tumor-related and 11 host-related covariates were evaluated as predictors of survival. We assessed the goodness of fit of both prognostic model by Akaike information criterion (AIC). We defined the sum of logarithm of estimated hazard ratio of each factor as prognostic index (P.I.), based on time-dependent Cox regression model and estimated survival in each value of P.I. All statistical analysis was performed by SAS statistical software. RESULTS; Overall 5-years survival rate was 52.4%. In multivariate analysis of both model, the following factors were independently related to survival; tumor size, tumor morphology, portal vein invasion, distant metastasis, serum AFP level, serum DCP level, serum albumin level and serum total bilirubin level. Estimated survival corresponding to P.I. 0-1/2/3/4was 7.0/3.0/1.4/0.3 years at initial treatment and 6.1/2.0/0.4/0.4 years at three years after initial treatment, respectively. AIC of time-dependent model (1665.2) was lower than that of time-fixed model (2114.4), indicating that better prediction can be achieved by time-dependent model. CONCLUSIONS; Time-dependent Cox regression model is superior for the prediction of the prognosis of hepatocellular carcinoma, compared with standard time-fixed Cox regression model.