Enhancing prediabetes and diabetes detection through a machine learning-enabled self-assessment approach.

Area of Science:

• Machine learning applications in public health
• Diabetes and prediabetes risk prediction
• Non-invasive health screening technologies

Background:

• Lack of reliable, accessible, non-invasive self-assessment screening for prediabetes/diabetes hinders early intervention.
• Machine learning (ML) offers potential for developing novel risk prediction tools.

Purpose of the Study:

• To develop and externally validate an ML-derived self-assessment system (MEDWACS) for predicting prediabetes/diabetes likelihood.
• To identify easily accessible health parameters for self-assessment screening.

Main Methods:

• Analysis of 30 years of NHANES data (N=17,458) using multimodal data for ML model development.
• Identification of key predictors using the Boruta algorithm and selection of 7 accessible parameters for the MEDWACS model.
• External validation using NHANES 2021-2023 and Korea NHANES 2023 data.

Main Results:

• The 7-parameter MEDWACS model includes age, waist circumference, systolic blood pressure, gender, upper leg length, arm circumference, and BMI.
• Achieved strong internal (ROCAUC 0.804) and external validation performance (US ROCAUC 0.773, Korea ROCAUC 0.780).
• Demonstrated superior clinical utility compared to established screening guidelines, with an online tool developed for accessibility.

Conclusions:

• MEDWACS is a validated, non-invasive ML tool for risk stratification of prediabetes/diabetes using 7 accessible parameters.
• Facilitates timely clinical evaluations, potentially mitigating the public health impact of diabetes.
• Enables home-based self-assessment and supports clinical decision-making for early intervention.