The purpose of this study was to build prognostic models capable of estimating the outcomes of individual sorafenib-treated advanced stage hepatocellular carcinoma (HCC) patients based on specific patient and tumor factors.Methods:
A parametric model for time-to-event data was used to construct scoring systems based on the intent-to-treat data set from 480 sorafenib-treated patients with advanced stage HCC: 356 for derivation and 124 for validation. Clinical parameters included in the models were based on importance variable scores generated by a random forest approach and bootstrap resampling. The model’s accuracy was internally and externally assessed using the time-dependent C-index of discrimination and a Hosmer-Lemeshow type test for calibration.Results:
The models generated for time-to-progression and overall survival based on Child-Pugh score, serum α-fetoprotein, tumor morphology, and vascular invasion and/or extrahepatic involvement had good calibration and discrimination abilities, with C-indexes of 0.669 (3 mo progression) and 0.809 (6 mo survival), respectively. External validation results also showed that these models performed well in terms of goodness-of-fit and discrimination (C-index: 0.746 for 3 mo progression and 0.875 for 6 mo survival). Receiver operating characteristic curve analysis in the validation patients indicated that these models have better predictive power than Child-Pugh scores (C-index: 0.686 for 3 mo progression and 0.777 for 6 mo survival).Conclusions:
The prognostic tools developed to quantify the potential outcomes for progression and survival expected from sorafenib treatment can serve as useful clinical aids in personalized decision making regarding treatment in advanced stage HCC patients.