Lasso tree for cancer staging with survival data

Area of Science:

• Biostatistics
• Medical Informatics
• Oncology

Background:

• The tumor-node-metastasis (TNM) staging system is crucial for cancer diagnosis, treatment, and prognosis.
• Meaningful clinical application requires orderly grouping of T and N categories based on time-to-event outcomes.

Purpose of the Study:

• To propose a penalized regression method for selecting optimal groupings within the TNM staging system.
• To develop a data-driven approach for refining cancer staging based on survival data.

Main Methods:

• A penalized regression method, termed "lasso tree," is proposed, utilizing L1 constraints on differences between neighboring coefficients to enforce stage grouping.
• The method incorporates partial ordering constraints for ordinal T and N categories.
• The approach models stage selection as a problem on a partially ordered two-way grid.

Main Results:

• The lasso tree method successfully generates a series of optimal groupings with varying numbers of stages by adjusting a tuning parameter.
• Application to colorectal cancer staging demonstrates the method's utility.
• Simulation studies confirm the lasso tree's ability to achieve correct groupings with moderate sample sizes and its stability.

Conclusions:

• The lasso tree provides a flexible and data-driven approach to optimize cancer staging systems.
• It offers a visual, tree-like structure to illustrate progressive grouping, aiding clinical interpretation.
• The method is robust to data variations and random censoring, enhancing its reliability in survival analysis.