Penalized partial likelihood regression for right-censored data with bootstrap selection of the penalty parameter.

摘要: The Cox proportional hazards model is often used for estimating the association between covariates and a potentially censored failure time, corresponding partial likelihood estimators are estimation prediction of relative risk failure. However, unstable have large variance when collinearity exists among explanatory variables or number failures not much greater than interest. A penalized (log) proposed to give more accurate estimators. We show that asymptotically there always penalty parameter reduces mean squared error log risk, we propose resampling method choose parameter. Simulations an example bootstrap-selected can, in some instances, smaller bias errors risk. These methods illustrated with data set multiple myeloma from Eastern Cooperative Oncology Group.