Development and validation of machine learning tools for predicting postoperative complications in acute calculous cholecystitis
- Autori: Cicerone, O.; Frassini, S.; Gallotti, A.; Vanoli, A.; Ansaloni, L.; Maestri, M.; Fugazzola, P.; Augustin, G.; Morić, T.; Awad, S.; Alzahrani, A.M.; Elbahnasawy, M.; Massalou, D.; De Simone, B.; Demetrashvili, Z.; Kimpizi, A.D.; Schizas, D.; Balalis, D.; Tasis, N.; Papadoliopoulou, M.; Georgios, P.; Lasithiotakis, K.; Ioannidis, O.; Bains, L.; Magnoli, M.; Cianci, P.; Conversano, N.I.; Pasculli, A.; Andreuccetti, J.; Arici, E.; Pignata, G.; Tiberio, G.A.M.; Podda, M.; Murru, C.; Veroux, M.; Distefano, C.; Centonze, D.; Catena, F.; Favi, F.; Bova, R.; Convertini, G.; Agnoletti, V.; Balla, A.; Sasia, D.; Giraudo, G.; Gabriele, A.; Tartaglia, N.; Pavone, G.; D'Acapito, F.; Fabbri, N.; Ferrara, F.; Cimbanassi, S.; Ferrario, L.; Cioffi, S.; Ceresoli, M.; Fumagalli, C.; Degrate, L.; Degiuli, M.; Sofia, S.; Licari, L.; Anderloni, A.; Ghaly, A.; Dagnoni, A.; Tassi, C.; Dominioni, T.; Tomasoni, M.; Cobianchi, L.; Cicuttin, E.; Viganò, J.; Farè, C.N.; Sargenti, B.; Improta, M.; Musella, V.; Gambini, G.; Patriti, A.; Coletta, D.; Conti, L.; Malerba, M.; Andrea, M.; Calabrò, M.; De Zolt, B.; Bellio, G.; Giordano, A.; Luppi, D.; Corbellini, C.; Sampietro, G.M.; Marafante, C.; Rossi, S.; Mingoli, A.; Lapolla, P.; Cicerchia, P.M.; Siragusa, L.; Grande, M.; Arcudi, C.; Antonelli, A.; Vinci, D.; De Martino, C.; Armellino, M.F.; Bisogno, E.; Visconti, D.
- Anno di pubblicazione: 2025
- Tipologia: Articolo in rivista
- OA Link: http://hdl.handle.net/10447/710787
Abstract
Background Acute calculous cholecystitis (ACC) is a common gastrointestinal emergency, with early laparoscopic cholecystectomy (ELC) as the standard of care. However, the risk of postoperative complications remains significant. This study developed and validated two machine learning models-CholeSurgRisk I (preoperative) and CholeSurgRisk II (comprehensive)-to predict major postoperative complications in ACC. Methods A prospectively collected, multicenter database of 1,253 patients was retrospectively analyzed. Lasso regression identified key predictive variables among demographic, clinical, and perioperative factors. Three machine learning algorithms (Random Forest, XGBoost, Decision Tree) were trained and tested, comparing their performance via AUC-ROC. Results CholeSurgRisk I achieved an AUC-ROC of 0.8456, while incorporating intraoperative variables (CholeSurgRisk II) improved performance to 0.8903. To facilitate clinical use, a web-based tool - "CholeSurgRisk I" - was developed based on the preoperative model, providing real-time, patient-specific risk estimations. Conclusion Machine learning enhances perioperative risk stratification in ACC. CholeSurgRisk I facilitates early preoperative assessment, whereas CholeSurgRisk II refines predictions by integrating intraoperative factors. The user-friendly application "CholeSurgRisk I" offers individualized complication risk forecasts, potentially aiding clinical decision-making and optimizing outcomes.