Classification Tree-Based Machine Learning to Visualize and Validate a Decision Tool for Identifying Malnutrition in...

Area of Science:

• Oncology
• Clinical Nutrition
• Medical Informatics

Background:

• The Global Leadership Initiative on Malnutrition (GLIM) framework offers a standardized approach to diagnosing malnutrition.
• The utility of machine learning (ML) in applying GLIM criteria in clinical settings is not well-established.

Purpose of the Study:

• To develop and validate a machine learning-based decision tool for malnutrition assessment in cancer patients using GLIM criteria.
• To evaluate the performance of this tool in a large, multicenter cohort.

Main Methods:

• A multicenter observational cohort study of 3998 cancer patients was conducted.
• Malnutrition was diagnosed using GLIM criteria and patients were divided into derivation and validation groups.
• A Classification and Regression Trees (CART) algorithm was employed to create a decision tree model.

Main Results:

• The GLIM criteria identified 14.7% moderate and 13.3% severe malnutrition. The CART model, using predictors like weight loss and BMI, achieved high accuracy (0.955) in the validation group.
• The decision tree demonstrated excellent performance with an AUC of 0.964.
• The model showed consistent performance across various cancer types.

Conclusions:

• Machine learning can effectively visualize and validate a decision tool based on GLIM criteria.
• This tool facilitates the accelerated pretreatment identification of malnutrition in cancer patients.