HemoGraph: A Machine Learning Method for Type II Diabetes Pre-screening Using Social and Behavioral Determinants of...

Area of Science:

• Computational biology and bioinformatics
• Artificial intelligence in healthcare
• Public health and epidemiology

Background:

• Type II diabetes presents a growing global health challenge affecting diverse age groups.
• Effective pre-screening tools are crucial for early detection and management of diabetes.
• Existing screening methods may not fully leverage comprehensive health determinant data.

Purpose of the Study:

• To introduce HemoGraph, a novel deep learning approach for type II diabetes pre-screening.
• To utilize social and behavioral determinants of health alongside graph neural networks for enhanced prediction.
• To evaluate HemoGraph's performance using real-world data, comparing it against established screening tools.

Main Methods:

• Development of HemoGraph, a deep learning model employing graph neural networks.
• Integration of social and behavioral determinants of health as input features.
• Application of the Learning Under Privileged Information (LUPI) framework, using lab data during training only.
• Model training and evaluation on the National Health and Nutrition Examination Survey (NHANES) dataset (1999-2018).

Main Results:

• HemoGraph achieved a recall of 71.56% in diabetes pre-screening.
• Demonstrated a 22% improvement in recall compared to the American Diabetes Association self-test tool.
• Exhibited superior overall performance metrics in identifying individuals at risk for type II diabetes.

Conclusions:

• HemoGraph offers a promising advancement in type II diabetes pre-screening by integrating complex health determinants.
• The deep learning approach, particularly with graph neural networks and LUPI, enhances predictive accuracy.
• This method has the potential to improve early detection rates and inform public health strategies for diabetes management.