Related Experiment Video
Updated: Jun 5, 2026
Laparoscopic Radical Left Pancreatectomy for Pancreatic Cancer: Surgical Strategy and Technique Video
Published on: June 6, 2020
Predicting Best Performers After Minimally Invasive Left Pancreatectomy: Insights From a National Cohort
Clément Pastier1,2, Marc-Anthony Chouillard1, Charles De Ponthaud2,3,4
1Department of HPB Surgery and Liver Transplantation, AP-HP, Beaujon Hospital, University of Paris Cité, Centre de Recherche sur l'Inflammation, INSERM Unité Mixte de Recherche 1149, Clichy, France.
Predicting ideal postoperative trajectories after minimally invasive left pancreatectomy (MILP) is challenging. A machine learning model incorporating preoperative and intraoperative factors can estimate the likelihood of an ideal outcome, aiding in patient care optimization.
Area of Science:
- Minimally invasive surgery
- Surgical outcomes research
- Predictive analytics in healthcare
Background:
- Assessing postoperative recovery after minimally invasive left pancreatectomy (MILP) is crucial for patient management.
- Currently, no validated tool exists to predict an ideal postoperative course following MILP.
Purpose of the Study:
- To identify predictors of ideal postoperative trajectories after minimally invasive left pancreatectomy (MILP).
- To develop a predictive model for optimal recovery following MILP.
Main Methods:
- Analysis of 2,092 MILP cases from 55 French centers (2010-2022).
- Defined Ideal Outcome (IO) and Best Performer (BP) based on mortality, complications, pancreatic fistula, reoperation, readmission, and length of stay.
- Utilized multivariable logistic regression and extreme gradient boosting (XGB) for predictor evaluation, with nested cross-validation and bootstrap resampling for performance assessment.
Main Results:
- The final XGBoost model, incorporating preoperative and intraoperative variables, achieved an AUC of 0.72 for predicting Best Performer.
- Key predictors included center volume, operative duration, age, BMI, conversion, blood loss, and splenectomy.
- The model demonstrated good discriminative ability with a sensitivity of 0.78 and NPV of 0.87 at the optimal threshold.
Conclusions:
- Predicting ideal postoperative recovery after MILP is complex but feasible with advanced modeling.
- Integrating preoperative and intraoperative determinants can help optimize postoperative pathways.
- An online risk calculator is available to assist clinicians in predicting patient outcomes.
