Hierarchical False Discovery Rate Control for High-dimensional Survival Analysis with Interactions

Area of Science:

• Biostatistics
• Genomics
• Survival Analysis

Background:

• High-dimensional data and interaction models are increasingly common in survival analysis.
• Gene-environment (G-E) interaction analysis presents unique challenges due to high-dimensional genetic data and low-dimensional environmental factors.
• Existing inference methods for high-dimensional data are insufficient for interaction models, lacking robust false discovery rate (FDR) control.

Purpose of the Study:

• To develop a statistically rigorous method for inference in high-dimensional survival data with interactions.
• To establish a hierarchical false discovery rate (FDR) control procedure that respects the structure of main effects and interactions.
• To address the limitations of existing high-dimensional inference tools in the context of interaction models.

Main Methods:

• Utilized the Accelerated Failure Time (AFT) model for survival data analysis.
• Employed a "weighted least squares + debiased Lasso" strategy for parameter estimation and variable selection.
• Developed a novel hierarchical FDR control approach tailored for interaction effects.

Main Results:

• Rigorously established the asymptotic distribution properties of the debiased Lasso estimators.
• Demonstrated satisfactory performance through simulations, validating the proposed approach.
• Confirmed the practical utility of the method via analysis of a real-world breast cancer dataset.

Conclusions:

• The proposed "weighted least squares + debiased Lasso" approach with hierarchical FDR control is effective for high-dimensional survival analysis involving interactions.
• This method provides a reliable framework for statistical inference in complex genetic studies, such as G-E interactions.
• The findings offer a valuable tool for researchers analyzing large-scale genomic and survival datasets.