Pursuing sparsity and homogeneity for multi-source high-dimensional current status data

Abstract

Nowadays, current status data with high-dimensional predictors are prevalent in observational studies. However, for a single study, the high dimensionality and the presence of censoring pose substantial challenges to statistical analysis with limited sample size. Although integrative analysis has been widely regarded as an effective strategy to improve the estimation, the source-level heterogeneity has to be carefully addressed. In this paper, we propose an integrative analysis method for multi-source high-dimensional current status data, which can simultaneously identify the homogeneity/heterogeneity structure and select important variables. We prove that the proposed approach attains consistency in estimation, sparsity recovery, and the pursuit of homogeneity. Extensive simulation studies have been carried out to assess the finite sample performance of the proposed method. A real data analysis of multi-source ovarian cancer recurrence studies further demonstrates its practical applicability.