Utilizing machine learning models for predicting outcomes in acute pancreatitis: development and validation in three...

Area of Science:

• Computational medicine and data science
• Clinical informatics and predictive analytics

Background:

• Acute pancreatitis (AP) has a high readmission rate, yet lacks robust predictive models for post-discharge outcomes.
• Existing in-hospital prediction models for AP patients have significant limitations.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting prognosis in AP patients.
• To encompass in-hospital mortality, readmission rates, and post-discharge mortality.
• To compare the predictive performance of ML models against established clinical scores.

Main Methods:

• Retrospective analysis of clinical and laboratory data from 2,559 AP patients across three databases.
• Variable selection using univariate logistic regression and LASSO method.
• Construction and validation of six ML algorithms, including Logistic Regression, Random Forest, and XGBoost, using training and external validation sets.

Main Results:

• The Logistic Regression, Random Forest, and XGBoost models demonstrated superior performance in predicting mortality.
• The XGBoost model showed optimal performance in predicting in-hospital mortality for ICU-admitted AP patients.
• XGBoost exhibited stability across centers, balanced sensitivity/specificity, and effective overfitting prevention.

Conclusions:

• An XGBoost model utilizing dynamic in-hospital variables can effectively predict short-term and long-term prognoses for AP patients.
• This ML approach supports clinical decision-making for improved AP patient management.