ABSTRACT
The R package ‘bhm’ or biomarker threshold models evaluate the treatment effect, biomarker effect, and treatment by biomarker interaction using hierarchical priors in Bayesian models with Markov chain Monte Carlo. The method of median split is simply dividing the overall population into two halves at the median value of the biomarker. The quartile split produces a categorical biomarker. The optimal split method is popular for finding the threshold of a continuous biomarker in the predictive setting. There are quite a few procedures and R packages for automating cut point selection in the prognostic setting, which are very helpful in developing diagnostic biomarkers. More advanced methods are required for more complex scenarios, like nonlinear interactions, or other shapes of biomarker treatment effect function over biomarker values in the two treatment groups.
