Abstract
PURPOSE: Intestinal obstruction surgery is a high-risk procedure associated with postoperative sepsis. In this multicenter retrospective study, we aimed to employ machine-learning methods to predict sepsis after intestinal obstruction surgery and visualize its driving mechanism to assist clinical decision-making. METHODS: The study included patients who underwent surgery for intestinal obstruction at three medical centers between 2013 and 2021, as well as patients with intestinal obstruction from the publicly available perioperative dataset INSPIRE. Various algorithms, including logistic regression, support vector classification (SVC), the Gaussian Bayesian classifier, gradient boosting machine, random forest, adaptive boosting, and multilayer perceptron, were tested. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), and the SHapley Additive exPlanation (SHAP) values were utilized for model interpretation. RESULTS: Postoperative sepsis was observed in 60 patients (16.17%), 10 patients (20%), and 10 patients (15.15%) in three medical centers, as well as 7 patients (12.73%) in INSPIRE dataset. In the internal validation set, the SVC algorithm demonstrated favorable performance, with an AUC value of 0.876 (95%CI 0.799-0.939). The model was well-calibrated, with a Brier score of 0.108 and a non-significant Hosmer-Lemeshow test (p = 0.183). The SHAP analysis identified six influential features in the model: age, POSSUM physiology score, ASA classification, shock index at the end of surgery, intraoperative total volume of infusion, and intraoperative dose of furosemide. External validation sets showed AUC values of 0.873, 0.896, 0.926, and 0.879, respectively. CONCLUSION: We have developed a highly interoperable sepsis risk calculator based on the SVC model to assist postoperative clinical decision-making in patients with intestinal obstruction.