A Bayesian Single-Index Model with fixed intercept for multiple biomarkers subset effects in clinical trials

Abstract

Predictive biomarkers like protein expressions have the potential of becoming the aid that predict treatment benefits for individual patients. Clinical trial data can be used in secondary analysis to determine the subset of patients who are the beneficiaries of a treatment regimen and the subset of patients who are not, using predictive biomarkers and a threshold. In this report, we propose a Bayesian single-index model that combines multiple biomarkers linearly and defines treatment sensitivity subsets. Markov Chain Monte Carlo methods are used for estimation and statistical inference. Compared to previous works, our method uses a fixed intercept to solve the identifiability problem, and is computationally efficient using assumed normal priors for the linear predictors. It can directly estimate and conduct inference with noncontinuous functions. The proposed method is evaluated with simulation studies and applied to biomarkers from clinical trial data of patients with locally advanced or metastatic pancreatic cancer.