Adverse events were graded according to v. 3.0 of the CTCAE of the National Cancer Institute, during treatment and 30 days after the last dose. Categorical variables are described as frequencies and percentages and continuous variables as median and percentiles 25 and 75 (P25-P75). Times to event data were estimated by Kaplan-Meier with plots and median (95% confidence interval [95% CI]). Fisher’s exact test was used to compare categorical variables and the Cochran-Armitage HSP inhibitor test

to assess trends. The Mann-Whitney method was used to compare ordinal and continuous variables. To define the predictors of OS we took into account the following baseline parameters: PS (0/1), Child-Pugh score (A/B 7 points), BCLC (B/C), extrahepatic spread (yes/no), total bilirubin, albumin, alpha-fetoprotein (AFP) (continued and categorized using median, tertiles, and three predefined different cutoffs [20, 200, 400]) and prior treatment (PEI/RFA/surgery). Moreover, we also assessed the impact of registering the

transition from Child-Pugh A (used as reference) into Child-Pugh B or C. Using this approach, the analysis introduces registration of Child-Pugh B or Child-Pugh C at a timepoint as one of the different time-dependent events that have been tested. These also include a change in PS (using PS 0 as reference), sorafenib dose modification (full dose as reference), presentation of encephalopathy and/or untreatable ascites, decrease in prothrombin time below 50%, albumin below 2.8 mg/dL, and AFP. Analysis of AFP was done using the same cutoffs (median, tertiles, 20, 200, www.selleckchem.com/products/Belinostat.html 400) as for the baseline. All statistics involving evolutionary events were done by means of time-dependent covariate analyses.[9] The inferential analysis for time to event data was conducted using the Cox univariate and multivariate

regression model with time-dependent covariates to estimate hazard ratios (HR) and 95% CI.[9] Statistically significant variables from the univariate Cox analysis, G protein-coupled receptor kinase progression pattern, and relevant variables from a clinical point of view were consistently included in the multivariate models, while also ensuring that the multivariate HR estimators did not change significantly when excluding those variables with P > 0.1. When specified, adjusted survival functions from that Cox model were used to draw survival plots. The analysis was performed using SAS v. 9.2 software (SAS Institute, Cary, NC), SPSS v. 18 (SPSS, Chicago, IL), and significance was established at the 0.05 level (two-sided). Between March 2008 and July 2011, 229 patients were assessed for sorafenib treatment. In all, 147 patients were enrolled and 82 patients were excluded as per inclusion and exclusion criteria (Fig. 1). At the time of database lock (May 2012), the median follow-up was 11.6 months (range: 0.4-51.8): 111 died, 28 out of 147 patients were still alive (with seven continuing sorafenib), and eight were lost to follow-up.