Predicting Colorectal Cancer Recurrence and Patient Survival Using Supervised Machine Learning Approach: A South...

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• South Africa has the highest colorectal cancer (CRC) incidence in Sub-Saharan Africa.
• Limited research exists on CRC recurrence and survival within South Africa.
• Predicting patient risk is crucial for managing clinical expectations and decisions.

Purpose of the Study:

• To explore the integration of statistical and machine learning (ML) algorithms for predicting CRC recurrence and survival.
• To achieve higher predictive performance and interpretability in findings for South African CRC patients.

Main Methods:

• Compared six algorithms: logistic regression, naïve Bayes, C5.0, random forest, support vector machine, and artificial neural network (ANN).
• Utilized 10-fold cross-validation and selected features based on OneR and information gain.
• Assessed model validity and stability using simulated datasets.

Main Results:

• All algorithms demonstrated high discriminative accuracies (AUC-ROC).
• ANN achieved the highest AUC-ROC for recurrence (87.0%) and survival (82.0%), with comparable performance from other models.
• Radiological stage, age, histology, and race were identified as significant risk factors for CRC recurrence and survival.

Conclusions:

• Rigorous procedures affirmed key predictive factors for recurrence and survival in colorectal cancer.
• The study's outcomes are generalizable to CRC patient populations in South Africa and other Sub-Saharan African countries with similar profiles.