EFFICIENT ESTIMATION OF THE MAXIMAL ASSOCIATION BETWEEN MULTIPLE PREDICTORS AND A SURVIVAL OUTCOME.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• High-dimensional data present challenges for survival outcome prediction.
• Post-selection inference is crucial for accurate predictor effect estimation.
• Existing methods lack reliability and scalability in high-dimensional settings.

Purpose of the Study:

• To develop a robust and computationally efficient post-selection inference method for high-dimensional survival data.
• To enable accurate identification of predictors associated with survival outcomes.
• To address confirmation bias inherent in predictor screening.

Main Methods:

• Construction of semi-parametrically efficient estimators for predictor-survival outcome associations.
• Development of a test statistic for detecting predictor-outcome associations.
• Application of a bagging-inspired stabilization technique for normal calibration and confidence interval construction.

Main Results:

• The proposed testing procedure is statistically valid even with superpolynomially increasing numbers of predictors relative to sample size.
• Simulations confirm the asymptotic guarantee at moderate sample sizes.
• The method successfully identified gene expression patterns linked to antiviral drug potency.

Conclusions:

• The new approach offers a reliable and scalable solution for post-selection inference in high-dimensional survival analysis.
• It effectively controls for confirmation bias, enhancing the validity of predictor screening.
• The method has practical applications in fields like pharmacogenomics and viral research.