Entropy-driven decision tree building for decision support in gastroenterology

Area of Science:

• Gastroenterology
• Medical Informatics
• Decision Support Systems

Background:

• Gastroesophageal reflux disease (GERD) significantly impacts patient quality of life globally.
• Accurate patient classification is crucial for effective GERD management.
• Current classification relies on endoscopic examination to assess esophageal mucosal lesions and their severity.

Purpose of the Study:

• To develop a decision support system for classifying GERD patients.
• To create a simplified decision tree model for GERD patient stratification.
• To identify characteristic patient values for each GERD class.

Main Methods:

• Utilized data from 269 patients, including healthy controls and various GERD classifications (non-erosive, erosive, complicated erosive, Barrett's disease).
• Integrated gastroscopy, pH-metry, and manometry test values into a decision tree framework.
• Employed entropy and information gain calculations to optimize the decision tree structure.

Main Results:

• A decision tree was successfully constructed to classify patients into distinct GERD categories.
• Identified specific patient data points and pathways within the tree that characterize each GERD class.
• The resulting decision tree model is relatively simple and includes paths with a significant number of patients.

Conclusions:

• The developed decision tree serves as a foundation for a diagnostic decision support tool.
• This approach can assist training physicians in classifying GERD patients based on specific characteristics.
• The method offers a structured way to interpret complex diagnostic data for improved GERD patient management.