Analysis of additive risk model with high-dimensional covariates using partial least squares

In this paper, we construct a partial additive regression (PAR) model to predict the survival times of cancer patients based on microarray gene expression data with right censoring. The area under time-dependent receiver operating characteristic curve is used as a model evaluation criterion. We conduct a simulation study to compare the proposed method with other methods, i.e. partial Cox regression and supervised principal component analysis. Two data sets of breast cancer and diffuse large B-cell lymphoma are analyzed to illustrate our procedure. The outcome indicates great predictive performance on both dimension reduction and predictive performance of the proposed PAR model.