Predicting operative time in robotic ventral hernia repair using machine learning: a retrospective pilot study

Robotic ventral hernia repair (VHR) is increasingly common, yet predicting operative time remains challenging. No study has yet applied machine learning (ML) to this setting. Objective To compare ML models for predicting operative time in elective robotic VHR and identify key preoperative predictors. Methods Retrospective single-center cohort study (AZORG Hospital, Aalst, Belgium) including 208 consecutive patients undergoing elective robotic VHR (October 2020–April 2025). Three models—Random Forest, Gradient Boosting, and Ridge Regression—were trained with 5-fold cross-validation using demographics, BMI, EHS hernia size, mesh position, and surgical complexity. Performance was assessed by MAE, RMSE, and R²; feature importance by SHAP analysis. Results Median operative time was 90 min (IQR 60–130). Random Forest achieved the best performance (MAE 38.0 min, RMSE 52.8 min, R² 0.22), outperforming Gradient Boosting (R² 0.01) and Ridge Regression (R² −0.53). SHAP identified mesh position (45.0%), hernia size (30.7%), and BMI (14.2%) as the top predictors. Conclusion In this pilot study, Random Forest provided modest operative time predictions (R² 0.22). Limited explained variance likely reflects the absence of surgeon experience and intraoperative variables. Although mesh position and hernia size are clinically consistent predictors, current accuracy is insufficient for clinical implementation. Larger multicenter studies incorporating surgeon-level data are needed.