A Weibull mixture cure frailty model for high-dimensional covariates.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Mixture cure models are valuable for survival data with a cured fraction but are underexplored, especially regarding frailty structures and high-dimensional data.
• Existing methods often fail with high-dimensional datasets where predictors outnumber observations, limiting their application in fields like omics research.

Purpose of the Study:

• To introduce a novel Weibull mixture cure frailty model capable of handling censored survival data and high-dimensional covariates.
• To develop a robust statistical framework for analyzing complex survival data, particularly in the context of high-dimensional omics datasets.
• To identify and validate prognostic biomarkers for breast cancer using RNAseq data.

Main Methods:

• Developed an extended Weibull mixture cure model incorporating a frailty component to account for latent heterogeneity.
• Integrated high-dimensional covariates into both cure rate and survival components, utilizing adaptive elastic-net penalization for variable selection.
• Employed a novel expectation-maximization (EM) algorithm for model inference and applied the approach to TCGA RNAseq breast cancer data.

Main Results:

• The proposed mixture cure frailty model demonstrated superior performance compared to existing methods in extensive simulation studies.
• Identified a set of prognostic biomarkers from RNAseq data, validated through functional enrichment analysis and literature comparison.
• Developed and validated a prognostic risk score index based on the identified biomarkers for breast cancer patients.

Conclusions:

• The novel mixture cure frailty model offers a powerful and comprehensive approach for analyzing high-dimensional censored survival data, particularly in biomedical research.
• The identified biomarkers and risk score provide valuable tools for predicting breast cancer patient outcomes.
• This methodology advances the application of statistical models in high-dimensional omics data analysis for biomarker discovery and prognostic assessment.