A Method for Analyzing Censored Survival Data with Application to Coronary Heart Disease

An objective of analyzing survival data via regression is to develop a predictive model given predictors. However, due to the censoring in response variables and the high dimensionality of predictors, information needed for an appropriate model specification is often inadequate. We propose a method for an integrated study of survival time and predictors. At first, variable selection methods are employed for finding the correct subset of predictors with significantly higher probability. This is based on the Lasso approach. Then, the dimension of the predictors is further reduced using sufficient dimension reduction methods. This is based on the Sliced inverse regression for censored data (DSIRII). In particular we use the popular Cox proportional hazards model to build a predictive model for survival data. An application to Coronary heart disease (CHD) data from the Tehran Lipid and Glucose (TGLS) study further illustrates the usefulness of the work.