A machine learning-based model for predicting survival in patients with Rectosigmoid Cancer

Area of Science:

• Oncology
• Surgical Oncology
• Machine Learning in Medicine

Background:

• The rectosigmoid junction has unique anatomical and vascular properties influencing its function and surgical approaches.
• Limited research exists on rectosigmoid junction cancer (RSC) prognosis, and effective clinical prediction models are needed.

Purpose of the Study:

• To identify independent risk factors for rectosigmoid junction cancer (RSC) survival.
• To develop and evaluate machine learning-based prediction models for RSC prognosis.
• To determine the optimal model for clinical decision support in RSC management.

Main Methods:

• Retrospective analysis of 524 RSC patients (2017-2019).
• Univariate and multivariate Cox regression to identify risk factors.
• Construction and evaluation of six machine learning models, including XGBoost.
• Assessment of model discrimination, calibration, and clinical utility using AUC and Brier scores.

Main Results:

• Seven independent risk factors for RSC survival were identified: age, gender, diabetes, tumor differentiation, N stage, distant metastasis, and anastomotic leakage.
• The XGBoost-based prediction model demonstrated superior performance with AUCs of 0.7856 (1-year), 0.8484 (3-year), and 0.796 (5-year).
• The XGBoost model exhibited the lowest Brier scores and superior clinical decision benefits compared to other models.

Conclusions:

• An optimal XGBoost-based prediction model for rectosigmoid junction cancer (RSC) prognosis was developed.
• This model can aid clinical decision-making for RSC patients.
• The model has the potential to improve survival outcomes for individuals diagnosed with RSC.