Development and Validation of an Early Severity Prediction Model for Hypertriglyceridemia-Associated Acute...

Hypertriglyceridemia-associated acute pancreatitis (HTG-AP) has become the second leading cause of acute pancreatitis (AP) in China. Compared with other etiologies, patients with HTG-AP are more likely to develop severe acute pancreatitis (SAP). This study aimed to develop and validate a prediction model for severe HTG-AP.

The derivation cohort consisted of 478 HTG-AP patients collected in a multicenter, prospective observational study (PERFORM study, 2020-2023, involving 36 tertiary hospitals in China). The external validation cohort included 145 prospectively enrolled HTG-AP patients from the General Hospital of Ningxia Medical University (from January 2024 to May 2025). Clinical variables were collected within 24 hours of enrollment. After excluding variables with more than 20% missing data, least absolute shrinkage and selection operator (LASSO) regression was used to select predictors. An XGBoost-based prediction model was constructed. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA), and compared with traditional scoring systems. SHapley Additive exPlanations (SHAP) analysis was employed to assess model interpretability.

A total of 113 patients (23.6%) in the derivation cohort and 23 patients (15.9%) in the validation cohort developed SAP, respectively. LASSO regression identified seven predictors: serum calcium (Ca²⁺), heart rate (HR), C-reactive protein (CRP), D-dimer (D-D), respiratory rate (RR), serum creatinine (SCr), and pleural effusion. The XGBoost model achieved an AUC of 0.873 in both the derivation and external validation cohorts, thereby significantly outperforming APACHE II (0.708, 0.701), SOFA (0.699, 0.685), SIRS (0.656, 0.649), and CTSI (0.661, 0.658) (all P < 0.05). The model showed good calibration (Hosmer-Lemeshow test P > 0.05) and provided a superior net clinical benefit across a wide range of threshold probabilities in DCA. SHAP analysis revealed that Ca²⁺ was the most influential predictor, followed by HR and CRP. To enhance clinical usability, we developed an interactive web-based calculator using the R Shiny framework.

This study developed and validated an XGBoost-based prediction model that uses seven easily obtained clinical variables for early identification of severe HTG-AP. The model demonstrated favorable discrimination, good calibration, and meaningful clinical utility, and outperformed traditional scoring systems. It offers a promising tool to improve risk stratification in HTG-AP.