SHapley Additive exPlanations (SHAP)-based, multivariate machine-learning techniques with external validation: Construction of a preoperative interpretable predictive model for intestinal resection of incarcerated inguinal hernia.

Abstract

Currently, there are a lack of effective tools for preoperative risk assessment of intestinal resection in patients with inguinal incarcerated hernia. The purpose of this study is to investigate the variable characteristics associated with intestinal resection and develop an interpretable preoperative prediction model, aiming to assist clinicians in preoperative risk for intestinal resection in patients with inguinal incarcerated hernia. The data from 2 medical institutions were retrospectively collected, and they were grouped according to whether intestinal resection was performed intraoperatively and the pathologic results. Lasso and multifactor logistic regression screened variables, and 10 machine-learning algorithms built and validated the model, with evaluation using the confusion matrix and SHapley Additive exPlanations value. Lasso regression and multifactorial logistic regression analyses showed that peritonitis, intestinal obstruction, neutrophil count, C-reactive protein, and preoperative total protein were the key characteristic variables. The area under curve of models constructed by 10 algorithms in the external validation set were all above 0.8, and the k-nearest neighbor algorithm had the most comprehensive model performance. The constructed model exhibits good predictive performance on the external validation set. Accurate preoperative prediction of intraoperative intestinal ischemia in patients with incarcerated inguinal hernia is crucial. This study identified peritonitis, intestinal obstruction, neutrophil count, C-reactive protein, and preoperative total protein as characteristic variables for predicting intraoperative intestinal ischemia in these patients. The constructed prediction model can assist clinicians in more accurately assessing intestinal viability during surgery, offering valuable insights for evaluating intestinal resection risk.