Development and validation of a machine learning model for predicting gangrenous cholecystitis: A retrospective cohort study
Ting Yu, Yanyu Chen, Chaoyang Wang, Yadong Zhou, Jianping GongObjective
This study aims to build a new machine learning model based on some inflammatory markers to predict the occurrence of gangrenous cholecystitis (GC) in patients with acute cholecystitis.
Materials and Methods
This retrospective cohort study included patients with acute cholecystitis who underwent cholecystectomy at our medical center between January 2018 and December 2024. The variables included in this study covered the demographic characteristics of the patients, laboratory test indicators, clinical data, and imaging data. Boruta algorithm was used to identify the key variables related to GC. After the feature selection is completed, the selected variables will be used to build 8 different machine learning prediction models. To evaluate the performance of each prediction model, this study employed receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA).
Results
The Boruta algorithm identified 7 variables that consistently held high importance in the model: Age, PNI, SII, SIRI, Gallbladder length, Gallbladder width, and Gallbladder wall thickness. Based on the key features selected, various machine learning models was constructed to predict the risk of GC. Among them, the Extra Trees model performed the best on the test set, with an AUC of 0.847. In the training set, the Extra Trees model also maintained stable and excellent performance (with an average AUC of 0.84). The calibration curves indicates that the predicted probability is in the best agreement with the actual result, demonstrating excellent calibration capabilities. DCA shows that within the prediction threshold range of 0–0.4, the Extra Trees model achieved the highest net benefit, indicating its potential application value in specific clinical scenarios.
Conclusions
This study developed a machine learning model integrating routinely available clinical, inflammatory, and imaging features for predicting gangrenous cholecystitis in patients with acute cholecystitis. Further external validation in independent multicenter cohorts is required before routine clinical implementation.