Machine learning to predict venous thromboembolism After Colorectal Cancer Surgery: a Chinese dynamic modelling...

Area of Science:

• Medical Informatics
• Oncology
• Vascular Surgery

Background:

• Existing venous thromboembolism (VTE) prediction tools for colorectal cancer (CRC) surgery lack precision for personalized patient care.
• There is a critical need for improved methods to dynamically predict postoperative VTE risk in CRC patients.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting postoperative VTE in Chinese patients undergoing CRC surgery.
• To enhance individualized risk assessment and optimize VTE prevention strategies through accurate prediction.

Main Methods:

• Developed 162 ML models using data from 1,836 Chinese CRC surgery patients (CRC-VTE trial).
• Utilized recursive feature elimination and Boruta for feature selection; employed SHapley Additive exPlanations (SHAP) for risk interpretation.
• Evaluated models on training, validation, and external test sets, prioritizing Area Under the Receiver Operating Characteristic Curve (AUROC).

Main Results:

• The CatBoost ML model demonstrated high performance, achieving AUROCs of 0.971 (postoperative) and 0.950 (preoperative) in the validation set.
• Models using postoperative data consistently outperformed those using only preoperative data.
• Simplified models incorporating key SHAP-identified predictors maintained strong predictive performance.

Conclusions:

• Machine learning models are feasible for predicting VTE after CRC surgery.
• Integrating ML with SHAP methodology provides a clinically applicable tool for personalized VTE risk assessment.
• This approach supports optimized VTE prevention strategies in CRC surgical patients.