Optimized machine learning model for predicting unplanned reoperation after rectal cancer anterior resection
Yang Su1, Yanqi Li1, Wangshuo Yang1
1Department of Gastrointestinal Surgery Center, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, 430030, China; Molecular Medicine Center, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, 430030, China.
Summary
This study developed a machine learning model to predict unplanned reoperation (URO) risk in rectal cancer patients after anterior resection. The model accurately identifies high-risk patients, aiding clinical decisions and improving outcomes.
Area of Science:
- Oncology
- Medical Informatics
- Surgical Outcomes
Background:
- Unplanned reoperation (URO) after anterior resection for rectal cancer negatively impacts patient prognosis and quality of life.
- There is a critical need for accurate predictive tools to identify patients at high risk of URO.
Purpose of the Study:
- To develop and optimize a machine learning (ML) model for predicting the risk of URO in rectal cancer patients undergoing anterior resection.
- To create a user-friendly online platform for real-time URO risk assessment.
Main Methods:
- Retrospective analysis of 2384 rectal cancer patients who underwent anterior resection.
- Feature selection using LASSO regression and Boruta algorithm, followed by ML model development with grid search and cross-validation.
- Performance evaluation using accuracy, specificity, sensitivity, and AUC; model interpretation via SHAP analysis.
Main Results:
- A robust ML model was constructed using 14 selected variables.
- The optimized random forest (RF) model achieved an AUC of 0.889 and accuracy of 0.842.
- Tumor location, prior abdominal surgery, and operative time were identified as key predictors of URO risk.
Conclusions:
- An optimized ML-based online system was developed for predicting URO risk after anterior resection in rectal cancer patients.
- The system provides accurate, real-time risk assessment, supporting clinical decision-making.
- This tool has the potential to improve patient prognosis and surgical outcomes.
