A multivariate Bayesian biomarker screening algorithm that incorporates clinical covariates.

Advanced hepatocellular carcinoma (HCC) has limited treatment options and poor survival. Early detection of HCC improves survival, but current ultrasound screening is inconsistent and poorly adhered to. Biomarker-based blood tests offer a more accurate, accessible, and reliable alternative. Recent developments in methods for early detection of HCC have focused on the trajectories of multiple biomarkers, assuming that biomarkers follow a joint hierarchical mixture model with random changepoints. We extend this model, allowing covariates to influence three aspects of the model: biomarker trajectories, diagnosis times, and subject-specific prior risks. We evaluated whether covariates could capture unexplained heterogeneity in biomarker levels to improve early detection. A Markov chain Monte Carlo algorithm was implemented to derive posterior distributions used in risk calculations among future patients and to determine whether a patient has a positive screen. The screening algorithm was evaluated through simulation studies under a range of possible covariate extensions. The Hepatitis C Antiviral Long-term Treatment against Cirrhosis (HALT-C) trial was used to further study the proposed biomarker screening algorithms for HCC using cross-validation.