A Machine Learning Framework for Prognostic Modeling in Stage III Colon Cancer

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Stage III colon cancer presents a significant challenge in predicting patient outcomes.
• Identifying reliable prognostic factors is crucial for effective treatment planning and patient management.

Purpose of the Study:

• To evaluate overall survival in stage III colon cancer patients.
• To identify clinical, pathological, and demographic factors associated with survival.
• To compare the predictive performance of machine learning algorithms against traditional statistical methods.

Main Methods:

• Retrospective analysis of 452 stage III colon cancer patients.
• Survival analysis using Kaplan-Meier and log-rank tests.
• Prognostic factor identification and prediction using machine learning (coarse trees, bagged trees, SVM, KNN) and explainable AI (SHAP).

Main Results:

• Older age, ECOG performance score ≥ 2, stage IIIC, N2 lymph node metastasis, and comorbidities (especially diabetes) were linked to poorer survival (p < 0.05).
• Machine learning identified positive surgical margins, rash, mucositis, thrombocytopenia, chemotherapy cycles, tumor subtype, diarrhea, age, and anemia as key prognostic factors.
• Ensemble machine learning models (coarse tree, bagged trees) achieved higher accuracy (87%) in predicting mortality and recurrence compared to traditional methods.

Conclusions:

• Key prognostic factors influencing survival in stage III colon cancer were identified.
• Machine learning approaches, integrating clinical and treatment data, enhance prognostic accuracy and support clinical decision-making.
• Findings support the utility of AI in risk stratification for personalized colon cancer treatment.