Biomarker-based prognosis in hepatocellular carcinoma: validation and extension of the BALAD model

Background: The Japanese 'BALAD' model offers the first objective, biomarker-based, tool for assessment of prognosis in hepatocellular carcinoma, but relies on dichotomisation of the constituent data, has not been externally validated, and cannot be applied to the individual patients. Methods: In this Japanese/UK collaboration, we replicated the original BALAD model on a UK cohort and then built a new model, BALAD-2, on the original raw Japanese data using variables in their continuous form. Regression analyses using flexible parametric models with fractional polynomials enabled fitting of appropriate baseline hazard functions and functional form of covariates. The resulting models were validated in the respective cohorts to measure the predictive performance. Results: The key prognostic features were confirmed to be Bilirubin and Albumin together with the serological cancer biomarkers, AFP-L3, AFP, and DCP. With appropriate recalibration, the model offered clinically relevant discrimination of prognosis in both the Japanese and UK data sets and accurately predicted patient-level survival. Conclusions: The original BALAD model has been validated in an international setting. The refined BALAD-2 model permits estimation of patient-level survival in UK and Japanese cohorts.