Machine learning for temporary stoma after intestinal resection in surgical decision-making of Crohn's disease.

Abstract

Background: Crohn's disease (CD) often necessitates surgical intervention, with temporary stoma creation after intestinal resection (IR) being a crucial decision. This study aimed to construct novel models based on machine learning (ML) to predict temporary stoma formation after IR for CD.

Methods: Patient data who underwent IR for CD at our center between July 2017 and March 2023 were collected for inclusion in this retrospective study. Eligible CD patients were randomly divided into training and validation cohorts. Feature selection was executed using the least absolute shrinkage and selection operator. We employed three ML algorithms including traditional logistic regression, novel random forest and XG-Boost to create prediction models. The area under the curve (AUC), accuracy, sensitivity, specificity, precision, recall, and F1 score were used to evaluate these models. SHapley Additive exPlanation (SHAP) approach was used to assess feature importance.

Results: A total of 252 patients with CD were included in the study, 150 of whom underwent temporary stoma creation after IR. Eight independent predictors emerged as the most valuable features. An AUC between 0.886 and 0.998 was noted among the three ML algorithms. The random forest (RF) algorithms demonstrated the most optimal performance (0.998 in the training cohort and 0.780 in the validation cohort). By employing the SHAP method, we identified the variables that contributed to the model and their correlation with temporary stoma formation after IR for CD.

Conclusions: The proposed RF model showed a good predictive ability for identifying patients at high risk for temporary stoma formation after IR for CD, which can assist in surgical decision-making in CD management, provide personalized guidance for temporary stoma formation, and improve patient outcomes.