A Two-Stage Dimension Reduction Method for Induced Response and Its Applications

Abstract

Researchers in the biological sciences nowadays often encounter the curse of highdimensionality, which many previously developed statistical models fail to overcome. To tackle this problem, sufficient dimension reduction aims to estimate the central subspace (CS), in which all the necessary information supplied by the covariates regarding the response of interest is contained. Subsequent statistical analysis can then be made in a lower-dimensional space while preserving relevant information. Oftentimes studies are interested in a certain transformation of the response (the induced response), instead of the original one, whose corresponding CS may vary. When estimating the CS of the induced response, existing dimension reduction methods may, however, suffer the problem of inefficiency. In this article, we propose a more efficient two-stage estimation procedure to estimate the CS of an induced response. This approach is further extended to the case of censored responses. An application for combining multiple biomarkers is also illustrated. Simulation studies and two data examples provide further evidence of the usefulness of the proposed method.