Background: Operative duration in emergency incisional ventral hernia repair may reflect the combined burden of patient frailty, emergency presentation, defect complexity, and intraoperative decision-making. Machine learning can capture these nonlinear interactions, but prediction of long-term recurrence remains challenging due to low event rates and follow-up dependence. Methods: This was a secondary analysis of the multicenter ACTIVE cohort. Operative time was analyzed as a binary endpoint (> 90 vs ≤90 min), and recurrence as a binary long-term outcome. Predictors were restricted to preoperative and intraoperative variables. Raw operative time was used only to define the duration endpoint and excluded from its predictor set, but retained for recurrence prediction as an intraoperative variable. Models were developed using an 80/20 stratified train-test split. Preprocessing, imputation, scaling, encoding, and any oversampling were performed within training pipelines. Operative duration models were selected by cross-validated ROC-AUC, and recurrence models by cross-validated PR-AUC. Results: A total of 556 patients were included. Procedures lasting >90 min were associated with larger defect width, higher BMI, sepsis, bowel obstruction, open approach, additional procedures, and mesh repair. The selected operative duration model (SVC RBF) achieved held-out ROC-AUC 0.807 and PR-AUC 0.875. Logistic regression showed similar performance but was not selected due to prespecified criteria. The recurrence model (ExtraTrees Balanced) achieved ROC-AUC 0.764 and PR-AUC 0.314. At the F1-optimized threshold, sensitivity was 0.429 and specificity 0.816. Conclusion: Leakage-aware machine learning predicted prolonged operative duration with useful discrimination, supporting operative time as a surrogate of intraoperative complexity. Recurrence prediction showed moderate ROC discrimination but limited precision-recall performance, reflecting the low number of events. These findings should be considered exploratory and require external validation before clinical application.

Predicting prolonged operative duration and recurrence after emergency incisional ventral hernia repair: a leakage-aware machine learning analysis of the ACTIVE cohort / Riberti, N., Marzola, F., Crepaz, L., Marom, G., Dayan, D., Lezoche, G., Podda, M., Mastronardi, M., Arezzo, A., Ortenzi, M., Null, N., Di Pietrantonio, D., Ercolani, G., Balla, A., Sartori, A., Renzi, F., Falsetti, E., Verdi, D., Mondi, I., Cianci, P., et al.. - In: SURGICAL ENDOSCOPY. - ISSN 0930-2794. - (2026). [10.1007/s00464-026-13431-y]

Predicting prolonged operative duration and recurrence after emergency incisional ventral hernia repair: a leakage-aware machine learning analysis of the ACTIVE cohort

Lezoche, Giovanni;Ortenzi, Monica;Patriti, Alberto;
2026-01-01

Abstract

Background: Operative duration in emergency incisional ventral hernia repair may reflect the combined burden of patient frailty, emergency presentation, defect complexity, and intraoperative decision-making. Machine learning can capture these nonlinear interactions, but prediction of long-term recurrence remains challenging due to low event rates and follow-up dependence. Methods: This was a secondary analysis of the multicenter ACTIVE cohort. Operative time was analyzed as a binary endpoint (> 90 vs ≤90 min), and recurrence as a binary long-term outcome. Predictors were restricted to preoperative and intraoperative variables. Raw operative time was used only to define the duration endpoint and excluded from its predictor set, but retained for recurrence prediction as an intraoperative variable. Models were developed using an 80/20 stratified train-test split. Preprocessing, imputation, scaling, encoding, and any oversampling were performed within training pipelines. Operative duration models were selected by cross-validated ROC-AUC, and recurrence models by cross-validated PR-AUC. Results: A total of 556 patients were included. Procedures lasting >90 min were associated with larger defect width, higher BMI, sepsis, bowel obstruction, open approach, additional procedures, and mesh repair. The selected operative duration model (SVC RBF) achieved held-out ROC-AUC 0.807 and PR-AUC 0.875. Logistic regression showed similar performance but was not selected due to prespecified criteria. The recurrence model (ExtraTrees Balanced) achieved ROC-AUC 0.764 and PR-AUC 0.314. At the F1-optimized threshold, sensitivity was 0.429 and specificity 0.816. Conclusion: Leakage-aware machine learning predicted prolonged operative duration with useful discrimination, supporting operative time as a surrogate of intraoperative complexity. Recurrence prediction showed moderate ROC discrimination but limited precision-recall performance, reflecting the low number of events. These findings should be considered exploratory and require external validation before clinical application.
2026
Emergency surgery; Hernia surgery; Machine learning; Ventral hernia
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11566/367274
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact