Abstract:
|
While the conventional approach to recommending cancer treatments has been through expert-driven guidelines, these recommendations can be confirmed or improved by the available large amounts of clinical and genomics data. The use of data and computational methods for cancer treatment recommendation, however, is challenging. Here a model-free method and a visualization tool are developed to aid treatment recommendation with high-dimensional genomic-clinical data. A global hypothesis testing method, named Cauchy combination test, is applied to first evaluate an overall effect of a big data set for treatment response. A dimension reduction technique, namely Sliced Inverse Regression, is then used to predict treatment response. The dimension reduction method is model-free and thus works broadly for any treatment response models. It offers simple visualization that can be directly used to compare different treatment options and display the optimal one. The proposed method is applied to a real-data example for the treatment of multiple myeloma, showing biological implications of the treatment mechanism. Simulation studies further demonstrate the advantages of the proposed method.
|