Retrospective pilot study compares machine learning models for operative time prediction in robotic VHR, suggesting key predictors.
Key Points
The research aims to predict operative time for robotic ventral hernia repair using machine learning models and to identify crucial preoperative predictors.
Retrospective single-center cohort study at AZORG Hospital including 208 patients undergoing robotic VHR.
Three machine learning models (Random Forest, Gradient Boosting, Ridge Regression) were used to predict operative time.
Model performance was assessed using MAE, RMSE, and R² with 5-fold cross-validation and feature importance analyzed via SHAP.
Random Forest provided the best performance with MAE of 38.0 min, RMSE of 52.8 min, and R² of 0.22.
Hernia size, mesh position, and BMI were identified as the top predictors for operative time with respective importance scores of 30.7%, 45.0%, and 14.2%.
Current predictive accuracy is insufficient for clinical implementation, indicating the need for larger studies.