Wang Jing, Xie Yao, Chen Li, Zhao Runpeng, Yu Hui, Ye Lingfeng, Peng Qisong, Wang Sheng, Han Yafei
Objective To examine the association between inflammatory markers and complicated appendicitis, develop machine‑learning prediction models using selected biomarkers and demographic data, and evaluate their predictive performance to support early diagnosis and management. Methods We performed a retrospective study of 330 consecutive patients with acute appendicitis admitted to the Hepatobiliary and Pancreatic Surgery Department of Nanjing Jiangning Hospital from January 2024 to June 2025. Patients were classified as non‑complicated appendicitis（n=151） or complicated appendicitis（n=179） based on intraoperative findings and pathology. Demographic and laboratory inflammatory variables were collected. Using a hold‑out approach, patients were randomly split 7∶3 into a training set（n=230） and a test set（n=100）. Training and test set demographics： training set, including 116 males, 114 females,125 complicated cases; test set, including 55 males, 45 females, 54 complicated cases.Feature selection was performed by least absolute shrinkage and selection operator（LASSO） regression followed by multivariable logistic regression to identify variables significantly associated with complicated appendicitis. Seven machine‑learning models were then trained on the selected features： logistic regression（LR）, decision tree（DT）, random forest（RF）, extreme gradient boosting（XGBoost）,light gradient boosting machine（LightGBM）, support vector machine （SVM）, and artificial neural network （ANN）. Model performance on the test set was assessed by area under the curve （AUC）, F1 score, sensitivity, specificity and accuracy. Model interpretability and variable importance were examined with SHAP （Shapley additive explanations）. Results LASSO and multivariable logistic regression identified five predictive features： systemic immune‑inflammation index （SII）,C‑reactive protein/albumin ratio（CAR）, neutrophil percentage/albumin ratio（NPAR）,platelet‑to‑lymphocyte ratio（PLR）,and age.Among the seven models evaluated on the independent test set,XGBoost achieved the best performance with an AUC of 0.786（95%CI：0.686-0.876）, specificity 91.11%, accuracy 75.52%, and F1 score 0.733. SHAP analysis indicated that CAR, age, and NPAR contributed most to model predictions. Conclusion Models based on SII, CAR, NPAR, PLR and age can discriminate complicated from non‑complicated appendicitis; among them, the XGBoost model demonstrated the best predictive performance. This approach shows promise as an early, laboratory‑based decision support tool to aid triage and management of patients with suspected complicated appendicitis.
