Construction of a predictive model for severe hypertriglyceridemia-associated acute pancreatitis using machine...

Area of Science:

• Gastroenterology
• Medical Informatics
• Biostatistics

Background:

• Severe hypertriglyceridemia-associated acute pancreatitis (SAP-HTG) requires accurate risk stratification.
• Distinguishing SAP-HTG from non-severe forms is crucial for timely intervention.
• Conventional scoring systems may have limitations in predicting SAP-HTG severity.

Purpose of the Study:

• To identify critical risk factors for SAP-HTG using machine learning.
• To develop and compare predictive models (logistic regression, random forest, SVM, XGBoost) against traditional scoring systems (APACHE II, Ranson).
• To enhance early diagnosis and risk stratification of SAP-HTG.

Main Methods:

• Retrospective analysis of 514 acute pancreatitis patients (90 SAP-HTG, 424 N-SAP-HTG).
• Utilized LASSO and multivariate logistic regression to identify predictors.
• Developed and evaluated four machine learning models (logistic regression, RF, SVM, XGBoost) using AUC, ROC curves, and SHAP analysis.

Main Results:

• Identified nine predictors: age, diabetes history, fasting blood glucose (FBG), TyG index, amylase (AMY), triglycerides (TG), total cholesterol (TC), C-reactive protein (CRP), and calcium (Ca2+).
• CRP and TG were significant risk factors; Ca2+ and age were protective.
• XGBoost (AUC=0.959) and Random Forest (AUC=0.955) models outperformed logistic regression, SVM, and traditional scoring systems.

Conclusions:

• Machine learning models effectively identified SAP-HTG risk factors.
• XGBoost demonstrated superior performance for SAP-HTG risk stratification.
• These models offer a valuable tool for early clinical diagnosis and management of SAP-HTG.