Developing a Simple Non-Laboratory-Based Machine Learning Tool for Prediabetes Screening in a Target Population: A...

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning in Healthcare

Background:

• Early detection of prediabetes is crucial for delaying progression to type 2 diabetes (T2D).
• Current screening methods rely on costly blood glucose tests, limiting accessibility.
• Existing machine learning models for prediabetes lack validation in target populations.

Purpose of the Study:

• To develop and validate a non-laboratory-based machine learning tool for prediabetes detection.
• To create a simple, implementable screening method for prediabetes.
• To address the limitations of current costly and inaccessible screening methods.

Main Methods:

• A decision tree model was developed using data from 501 adults.
• Twelve non-laboratory features were extracted, with forward feature selection identifying key predictors.
• The dataset was split into 70% for training and 30% for validation.

Main Results:

• 88 out of 501 participants were identified with prediabetes.
• Age and waist circumference were identified as the most important features for the model.
• The model achieved an ROC AUC of 0.8297 (training) and 0.7961 (validation).

Conclusions:

• A machine learning tool utilizing age and waist circumference demonstrates promising results for prediabetes screening.
• The tool's simplicity, requiring only two non-laboratory features, facilitates easy implementation.
• Further evaluation with additional data is necessary to confirm generalizability and external validity.