An interpretable survival benefit analytics framework for optimizing cancer treatment decision-making.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Optimal cancer treatment decisions are crucial for patient survival.
• Traditional models indirectly assess risks, limiting interpretability and performance.
• Need for interpretable frameworks to personalize cancer treatment.

Purpose of the Study:

• Introduce SurvS (survival supervision), a novel interpretable framework for survival benefit analytics.
• Integrate individual treatment effects directly impacting long-term survival.
• Develop both binary and ternary decision-making models for personalized cancer treatment.

Main Methods:

• Utilized a real-coded genetic algorithm for survival benefit analysis.
• Developed SurvS to construct personalized decision-making models.
• Integrated weighted clinical features and real-valued cutoff thresholds for model optimization.

Main Results:

• Demonstrated robust performance in nasopharyngeal and rectal cancer treatment scenarios.
• SurvS outperformed traditional decision-making methods.
• Maintained strong performance in an independent validation cohort.

Conclusions:

• SurvS provides a powerful tool for personalized cancer treatment planning.
• Enables survival benefit-supervised optimization and interpretable model construction.
• Potential to improve treatment efficacy and reduce overtreatment in oncology.