Variable Selection for Nonlinear Covariate Effects with Interval-Censored Failure Time Data

This paper discusses variable selection when one faces general, high-dimensional interval-censored failure time data, which commonly occur in many areas such as epidemiological, medical and public health studies. Furthermore, for the situation, it is often the case that covariates may have nonlinear effects, which makes the variable selection more challenging. For the problem, we propose a novel and robust variable selection technique under a class of semiparametric additive transformation models. In particular, Bernstein polynomials are employed to approximate unknown functions and an EM algorithm is developed with the use of Poisson-based data augmentation for the implementation of the proposed method. An extensive simulation study is conducted to assess the empirical performance of the approach and suggests that it works well in practical situations. Finally, we apply the method to a set of real data arising from an Alzheimer’s Disease study.