Machine learning-based predictive model for acute pancreatitis-associated lung injury: a retrospective analysis

Area of Science:

• Medical Informatics
• Critical Care Medicine
• Machine Learning in Healthcare

Background:

• Acute Pancreatitis-Associated Lung Injury (APALI) is a severe complication of acute pancreatitis, associated with high morbidity and mortality.
• Early diagnosis and intervention are crucial for improving patient outcomes.

Purpose of the Study:

• To develop and validate machine learning models for predicting APALI.
• To identify key clinical factors associated with APALI development.

Main Methods:

• Utilized data from two major hospitals (2012-2023) with a total of 1,975 patients.
• Applied LASSO regression for feature selection and constructed six machine learning models.
• Evaluated models using AUC, accuracy, sensitivity, specificity, F1 score, and recall, with SHAP for feature importance analysis.

Main Results:

• XGBoost and Random Forest models showed superior predictive performance (AUC 0.92 and 0.914, respectively).
• Key predictors identified include CRP, BMI, neutrophil count, calcium, lactate, and neutrophil-to-albumin ratio (NAR).
• An online risk calculator was developed based on the optimal models.

Conclusions:

• Developed and validated machine learning models demonstrate strong predictive capability for APALI.
• The online risk calculator can aid in early risk stratification and clinical decision-making.
• Further prospective multicenter validation is recommended to confirm generalizability.