1: \begin{abstract}

2: We study the Cox models with semiparametric relative risk, which can be

3: partially linear with one nonparametric component, or multiple additive

4: or nonadditive nonparametric components. A penalized partial likelihood

5: procedure is proposed to simultaneously estimate the parameters and

6: select variables for both the parametric and the nonparametric parts.

7: Two penalties are applied sequentially. The first penalty, governing

8: the smoothness of the multivariate nonlinear covariate effect function,

9: provides a smoothing spline ANOVA framework that is exploited to derive

10: an empirical model selection tool for the nonparametric part. The

11: second penalty, either the smoothly-clipped-absolute-deviation (SCAD)

12: penalty or the adaptive LASSO penalty, achieves variable selection in

13: the parametric part. We show that the resulting estimator of the

14: parametric part possesses the oracle property, and that the estimator

15: of the nonparametric part achieves the optimal rate of convergence. The

16: proposed procedures are shown to work well in simulation experiments,

17: and then applied to a real data example on sexually transmitted

18: diseases.

19: \end{abstract}