Comparative analysis of robotic vs. laparoscopic right hepatectomy: propensity score matching and machine learning analysis for outcome prediction

Background Robotic right hepatectomy (RRH) is increasingly utilized as a minimally invasive liver surgery approach, yet its comparative effectiveness against laparoscopic right hepatectomy (LRH) remains debated. Complication prediction is crucial to efficiently choice the approach. This study aimed to compare RRH and LRH using a propensity score-matched (PSM) analysis and assess the predictive power of machine learning (ML), focusing on Random Forest-based feature selection. Methods We retrospectively analyzed patients undergoing RRH and LRH, performing a 1:1 PSM based on preoperative characteristics. Postoperative outcomes were compared, and significant predictors of textbook outcomes (TO) were identified through univariate and multivariate analysis. A Random Forest model was developed, selecting key variables for TO achieving prediction. Results After PSM, 30 RRH and 30 LRH patients were compared. RRH was associated with increased clamping (80% vs. 23.3%, p = 0.0001) and longer clamping duration (44 min vs. 35 min, p = 0.209) but resulted in a shorter hospital stay (6 vs. 7.5 days, p = 0.004) and fewer complications (13.3% vs. 53.3%, p = 0.001). Multivariate analysis identified Robotic surgical approach (OR 0.282, p = 0.004), Chronic hepatitis/Cirrhosis (OR 1.90, p = 0.034), and estimated blood loss (OR 2.001, p = 0.014) as significant predictors of TO. Feature selection via ML confirmed robotic surgical approach as one of the most relevant factors, alongside operative duration, estimated blood loss, and lesion size, in determining postoperative outcomes. Conclusion The choice of surgical approach is a key determinant of outcomes, as demonstrated by its significance in both multivariate analysis and ML-based feature selection. RRH shows potential advantages in reducing hospital stay and complications while requiring more intraoperative management. Machine learning, particularly Random Forest, enhances outcome prediction by identifying critical surgical factors, supporting a more tailored approach in hepatobiliary surgery.