Development of therapeutic predictive biomarkers for various human diseases has become one of the central issues in current medicine to guide individual patients with their optimal treatments. In particular, such therapeutic diagnostic tests are critically demanded in advanced cancer whose patient population is found to be highly heterogeneous and only a very small proportion shows clinical benefit from each given therapeutic option. The development of such therapeutic biomarker models in oncology, however, is highly challenging since prognostic and predictive factors are often confounded in patient outcomes. It is essential to carefully examine and validate candidate biomarker models in order to confirm their predictive power for a specific treatment response. We will briefly discuss relevant statistical issues for therapeutic biomarker development in this talk.