Dimension reduction methods for microarrays with application to censored survival data

Area of Science:

• Genomics
• Bioinformatics
• Biostatistics

Background:

• Gene expression profiles show potential for predicting clinical phenotypes, including survival time.
• Predicting patient survival time, especially with censored data, remains a challenge.
• Limited research has focused on predicting survival time using gene expression compared to tumor classification.

Purpose of the Study:

• To develop and evaluate a novel method for analyzing censored survival time data using microarray gene expression profiles.
• To identify linear combinations of genes that capture both expression variability and phenotypic information relevant to survival.
• To apply the developed method to a real-world cancer dataset for survival prediction.

Main Methods:

• A dimension reduction strategy combining Principal Component Analysis (PCA) and Sliced Inverse Regression (SIR).
• Identification of gene combinations that account for gene expression variability and preserve phenotypic information.
• Application of extracted gene combinations as covariates in a Cox proportional hazards survival model.

Main Results:

• The proposed method successfully identified predictive gene expression components.
• A Cox proportional hazards model built on these components demonstrated good predictive performance for patient survival.
• Significant survival differences were observed between predicted risk groups, validated by receiver operator characteristics analysis.

Conclusions:

• The combined PCA and SIR approach is effective for dimension reduction in gene expression data for survival analysis.
• The method provides a robust framework for predicting patient survival time using high-dimensional gene expression data.
• This approach holds promise for improving prognostic accuracy in cancer research and clinical applications.