Analysis of additive risk model with high-dimensional covariates using partial least squares

Area of Science:

• Bioinformatics
• Biostatistics
• Cancer Research

Background:

• Predicting cancer patient survival is crucial for treatment planning.
• Microarray gene expression data offers a rich source of potential biomarkers.
• Existing methods may have limitations in handling high-dimensional censored survival data.

Purpose of the Study:

• To introduce and evaluate a Partial Additive Regression (PAR) model for cancer survival prediction.
• To assess the PAR model's performance against established methods like partial Cox regression and supervised principal component analysis.
• To demonstrate the utility of the PAR model using real-world cancer datasets.

Main Methods:

• Construction of a Partial Additive Regression (PAR) model.
• Utilizing microarray gene expression data with right-censored survival times.
• Employing the area under the time-dependent receiver operating characteristic curve (AUC) for model evaluation.
• Conducting simulation studies for comparative analysis.
• Applying the model to breast cancer and diffuse large B-cell lymphoma datasets.

Main Results:

• The proposed PAR model demonstrated superior predictive performance compared to partial Cox regression and supervised principal component analysis.
• The PAR model effectively achieved dimension reduction on the gene expression data.
• Analysis of breast cancer and diffuse large B-cell lymphoma datasets confirmed the model's practical applicability and effectiveness.

Conclusions:

• The Partial Additive Regression (PAR) model is a powerful tool for predicting cancer patient survival from gene expression data.
• The PAR model offers significant advantages in both dimension reduction and predictive accuracy.
• This approach holds promise for improving prognostic assessments in oncology.