Pseudo-value regression trees.

Area of Science:

• Biostatistics
• Survival Analysis
• Machine Learning in Healthcare

Background:

• Estimating survival functions from right-censored time-to-event data is crucial in medical research.
• Standard methods like generalized estimating equations (GEE) have limitations in variable selection and interaction detection.
• Pseudo-value regression provides a framework for modeling individual survival probabilities.

Purpose of the Study:

• To introduce and evaluate a novel semi-parametric modeling technique, pseudo-value regression trees (PRT).
• To address limitations of standard GEE approaches by incorporating tree learning and additive modeling.
• To improve variable selection, identify covariate interactions, and handle time-dependent effects in survival analysis.

Main Methods:

• Developed pseudo-value regression trees (PRT) by integrating multivariate regression trees with pseudo-value outcomes.
• Employed regularized additive models within tree nodes to capture complex relationships.
• Incorporated variable selection, interaction identification, and time-dependent effects.
• Controlled tree depth for model interpretability.

Main Results:

• PRT demonstrated effectiveness in variable selection and identifying relevant covariate interactions.
• The method successfully incorporated time-dependent effects into survival models.
• Simulation studies validated the properties of PRT and its comparison to alternative techniques.
• PRT was illustrated on a primary invasive breast cancer patient dataset.

Conclusions:

• Pseudo-value regression trees (PRT) offer a powerful and interpretable semi-parametric approach for survival analysis.
• PRT overcomes limitations of standard GEE by effectively handling variable selection, interactions, and time-dependent effects.
• The method shows promise for analyzing complex time-to-event data in clinical research, as evidenced by its application to breast cancer survival data.