Machine learning based prediction of recurrence after curative resection for rectal cancer

Area of Science:

• Oncology
• Medical Informatics
• Surgical Oncology

Background:

• Rectal cancer treatment often involves radical surgery for non-metastatic disease.
• Tumor recurrence post-curative resection is influenced by various factors.
• Predictive modeling can aid in identifying high-risk patients for recurrence.

Purpose of the Study:

• To analyze factors associated with rectal cancer recurrence following curative resection.
• To apply and compare different machine learning techniques for predicting recurrence.
• To identify significant prognostic factors for rectal cancer recurrence.

Main Methods:

• A cohort of 961 patients undergoing curative rectal cancer surgery (2004-2018) was analyzed.
• The Synthetic Minority Oversampling Technique with Tomek link (SMOTETomek) balanced the dataset.
• Four machine learning models (logistic regression, SVM, RF, XGBoost) were employed, with feature importance assessed by permutation importance.

Main Results:

• The recurrence rate was 13.2% (127/961 patients) over a median follow-up of 60.8 months.
• Top predictors included pathologic Tumor stage (pT), sex, concurrent chemoradiotherapy, and pathologic Node stage (pN).
• Support Vector Machine (SVM) achieved the highest Area Under the Curve (AUC) of 0.831, with pathologic Tumor stage (pT) being the most influential factor across most models.

Conclusions:

• Support Vector Machine (SVM) demonstrated superior performance in predicting rectal cancer recurrence.
• Pathologic Tumor stage (pT) emerged as the most critical factor influencing recurrence risk.
• Enhanced surveillance is recommended for rectal cancer patients with advanced pT stage post-surgery.