Use of Automated Machine Learning to Detect Undiagnosed Diabetes in US Adults: Development and Validation Study

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning in Healthcare

Background:

• Early diabetes diagnosis is crucial for interventions to slow disease progression.
• A significant portion of individuals with diabetes remain undiagnosed, hindering timely treatment.

Purpose of the Study:

• To explore the utility of automated machine learning (AutoML) models for detecting undiagnosed diabetes in US adults.
• To evaluate the performance of AutoML models using self-reported data and biochemical markers.

Main Methods:

• Utilized the National Health and Nutrition Examination Survey (1999-2020) data from 11,815 participants.
• Employed the H2O AutoML framework for automated model training and hyperparameter tuning.
• Compared AutoML performance against traditional machine learning models (logistic regression, SVM, random forest, XGBoost).

Main Results:

• The AutoML model achieved a superior Area Under the Receiver Operating Characteristic Curve (AUC) of 0.909.
• Demonstrated high accuracy with 70.26% sensitivity and 90.46% specificity in identifying undiagnosed diabetes.
• Achieved a negative predictive value of 92.61%, indicating strong reliability in ruling out diabetes.

Conclusions:

• This study is the first to apply AutoML for undiagnosed diabetes detection in US adults.
• The AutoML model presents a powerful and scalable tool for diabetes screening in the general population.
• Findings support the integration of advanced machine learning for proactive public health initiatives in diabetes management.