A web-based dynamic nomogram integrating inflammatory and metabolic indicators for predicting prolonged hospital stay in patients with hypertriglyceridemia-induced acute pancreatitis

Hypertriglyceridemia-induced acute pancreatitis (HTG-AP) is a common cause of acute pancreatitis and is associated with worse clinical outcomes. Early prediction of prolonged hospitalization may facilitate clinical management and resource allocation. We aimed to develop and validate a web-based dynamic nomogram to predict prolonged length of stay (LOS) in patients with HTG-AP. We retrospectively analyzed 608 patients with HTG-AP admitted between 2014 and 2024. Patients were randomly divided into a training cohort (n = 487) and an internal validation cohort (n = 121). An independent external cohort of 39 patients (2021–2023) was used for external validation, and temporal validation was performed using a time-based split within the development cohort. Predictor variables were selected using LASSO regression and SHAP analysis. Independent predictors were identified by multivariate logistic regression and incorporated into a web-based nomogram. Model performance was assessed by discrimination, calibration, and decision curve analysis. Five independent predictors—systemic inflammatory response syndrome (SIRS), blood glucose, serum calcium, D-dimer, and APACHE II score—were included. Using LOS > 14 days as the endpoint, prolonged hospitalization occurred in 60.6%, 60.3%, and 69.2% of patients in the training, internal validation, and external validation cohorts, respectively. The nomogram achieved AUCs of 0.839, 0.812, and 0.691 in the training, internal validation, and external validation cohorts, respectively. Calibration curves demonstrated good agreement between predicted and observed outcomes, with mean absolute error (MAE) values of 0.012, 0.020, and 0.042 in the training, internal validation, and external validation cohorts, respectively. Decision curve analysis demonstrated net clinical benefit across a range of clinically relevant threshold probabilities. We developed and validated an admission-based prediction model presented as a web-based dynamic nomogram integrating inflammatory and metabolic indicators to predict prolonged hospitalization (LOS > 14 days) in patients with HTG-AP. The model showed good discrimination in internal validation but attenuated performance in external validation. Prospective multicenter validation and model recalibration are warranted before routine clinical implementation.