Buckley-James boosting for survival analysis with high-dimensional biomarker data

Area of Science:

• Bioinformatics
• Biostatistics
• Genomics

Background:

• Predicting patient survival post-therapy using gene expression microarray data is gaining interest.
• High-dimensional genomic data presents challenges for traditional regression and classification models.
• Boosting methods have shown success in building accurate predictive models and performing variable selection.

Purpose of the Study:

• To propose Buckley-James boosting for semiparametric accelerated failure time models with right-censored survival data.
• To enable prediction of future patient survival using high-dimensional genomic data.
• To incorporate twin boosting for fitting sparse models, inspired by adaptive LASSO.

Main Methods:

• Buckley-James boosting for semiparametric accelerated failure time models.
• Twin boosting for fitting sparse models.
• A unified approach for linear, non-linear, and interaction models.

Main Results:

• The proposed methods perform simultaneous variable selection and parameter estimation.
• The methods are evaluated through simulations.
• Applied to a microarray gene expression dataset for diffuse large B-cell lymphoma patients.

Conclusions:

• Buckley-James boosting offers a robust approach for survival prediction with high-dimensional genomic data.
• The twin boosting incorporation allows for parsimonious model fitting.
• The methods demonstrate utility in real-world clinical applications, such as predicting outcomes for lymphoma patients.