Clinical features of tumor-induced acute pancreatitis and construction of a machine learning prediction model
- VernacularTitle:肿瘤相关急性胰腺炎的临床特征及机器学习预测模型构建
- Author:
Chenhui DU
1
;
Yuqian GAO
1
;
Shuo ZHANG
2
;
Tieying HE
1
;
Xinling CAO
2
Author Information
- Publication Type:Journal Article
- Keywords: Pancreatitis; Digestive System Tumors; Machine Learning; Models， Statistical
- From: Journal of Clinical Hepatology 2026;42(7):1661-1669
- CountryChina
- Language:Chinese
- Abstract: ObjectiveTo investigate the clinical features of tumor-induced acute pancreatitis （TIAP）， to construct and validate a predictive model for TIAP， and to provide help for early identification in clinical practice. MethodsA retrospective analysis was performed for the clinical data of 3 051 patients with acute pancreatitis （AP） who were admitted to The First Affiliated Hospital of Xinjiang Medical University from January 2020 to January 2026， among whom there were 72 patients with TIAP. To reduce class imbalance， 216 patients with non-tumor-related AP （2 979 patients） were randomly selected as conventional group using sex-stratified sampling， resulting in a cohort of 288 patients， and this cohort was randomly divided into a training set with 201 patients and a test set with 87 patients at a ratio of 7∶3. The two groups were compared in terms of general information and laboratory markers. The independent-samples t test was used for comparison of normally distributed continuous data between two groups， and the Mann-Whitney U test was used for comparison of non-normally distributed continuous data between two groups； the chi-square test was used for comparison of categorical data between two groups. Recursive feature elimination and least absolute shrinkage and selection operator regression were used for screening of characteristic variables， and five machine learning models were constructed， i.e.， logistic regression model， random forest model， support vector machine model， extreme gradient boosting model， and light gradient boosting machine model. The receiver operating characteristic curve and the precision-recall curve were used to assess model performance； the calibration curve was used to assess goodness of fit； decision curve analysis was used to evaluate clinical applicability and practicality； Shapley additive explanations were used to assess model interpretability. ResultsIn the training set of 201 patients， there were 52 patients （25.9%） in the TIAP group， and in the test set of 87 patients， there were 20 patients （23.0%） in the TIAP group. Compared with the conventional group， the TIAP group had a significantly higher proportion of patients with pancreatic duct dilatation （χ2=79.474， P<0.05）， a significantly higher age （Z=-5.838， P<0.05）， and significantly lower levels of white blood cell count （Z=5.630， P<0.05）， amylase （Z=2.606， P<0.05）， hemoglobin （Z=5.038， P<0.05）， and neutrophil percentage （Z=5.269， P<0.05）， as well as a lower level of direct bilirubin （Z=0.936， P>0.05）. The five machine learning models constructed based on these seven variables had a certain predictive ability， among which the logistic regression model had the best performance in the test set， with an area under the curve of 0.893 （95% confidence interval： 0.821 — 0.953）， an average precision of 0.654 in the precision-recall curve， good calibration， and good clinical benefits based on the decision curve analysis. ConclusionThe predictive model for TIAP based on pancreatic duct dilatation， age， white blood cell count， amylase， hemoglobin， neutrophil percentage， and direct bilirubin shows good predictive performance and can provide important guidance for the early diagnosis of TIAP.
