Structured learning in time-dependent Cox models

Area of Science:

• Biostatistics
• Survival Analysis
• High-Dimensional Data Analysis

Background:

• Time-dependent Cox models are crucial for survival analysis with evolving risk factors.
• High-dimensional data necessitates sparse regularization for variable selection.
• Existing methods lack flexibility in handling complex covariate structures in time-dependent Cox models.

Purpose of the Study:

• To propose a flexible framework for variable selection in time-dependent Cox models.
• To accommodate complex grouping structures and selection rules.
• To develop an efficient computational tool for these models.

Main Methods:

• A novel framework for flexible variable selection in time-dependent Cox models.
• Adaptability to arbitrary grouping structures (interactions, temporal, spatial, tree, DAGs).
• Implementation using a network flow algorithm within the sox package.

Main Results:

• Accurate estimation with low false alarm rates in variable selection.
• Efficient computation for models with complex covariate structures.
• Demonstrated practical application in a case study of atrial fibrillation patients.

Conclusions:

• The proposed framework offers a flexible and accurate approach to variable selection in time-dependent Cox models.
• The sox package provides an efficient and user-friendly tool for analyzing complex survival data.
• This method enhances the understanding of predictors in time-to-event data, particularly in clinical settings.