Machine Learning-Based Prediction of Postoperative Deep Vein Thrombosis Following Tibial Fracture Surgery

Area of Science:

• Orthopedic Surgery
• Biomedical Engineering
• Data Science

Background:

• Postoperative deep vein thrombosis (DVT) is a significant complication following tibial fracture surgery.
• Accurate prediction of DVT is crucial for effective preventive strategies.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting DVT after tibia fracture surgery.
• To identify optimal ML algorithms and feature selection methods for DVT risk prediction.

Main Methods:

• Retrospective analysis of 471 patients undergoing isolated tibial fracture surgery.
• Development of 42 predictive models using six ML algorithms and seven feature selection techniques.
• Performance evaluation based on area under the receiver operating characteristic curve (AUC-ROC) and Brier scores, with internal validation.

Main Results:

• 17.0% of patients developed postoperative DVT.
• Multiple ML models achieved high discrimination (AUC ~0.97-0.99).
• Support vector machine (SVM) models with Boruta or LASSO feature selection demonstrated superior calibration and high specificity (≥95%).

Conclusions:

• ML models, particularly SVM-based ones, show high accuracy in predicting postoperative DVT after tibial fracture surgery.
• These models offer potential for risk stratification to guide individualized prophylaxis.
• Further prospective validation is recommended for clinical implementation.