Screening diabetes mellitus 2 based on electronic health records using temporal features

Area of Science:

• Medical Informatics
• Machine Learning
• Public Health

Background:

• Type 2 diabetes mellitus (T2DM) prevalence is rising globally.
• Current T2DM treatments and screening methods are inadequate.
• Non-invasive prognostic approaches are needed to reduce disease burden.

Purpose of the Study:

• To develop a prognostic model for T2DM using electronic health records (EHRs).
• To predict T2DM without relying on invasive measurements like glucose levels or HbA1c.
• To leverage machine learning and temporal features for enhanced prediction accuracy.

Main Methods:

• Utilized machine learning frameworks, specifically random forest classifiers.
• Enriched EHR data with temporal features to capture patient progression.
• Applied feature selection techniques to optimize the predictive model.

Main Results:

• Achieved an area under the receiver operating characteristics curve (AUC) of 84.22% for predicting T2DM in 2012 using 2009-2011 data.
• AUC reached 83.19% when incorporating temporal features.
• AUC was 83.72% after applying both temporal features and feature selection.

Conclusions:

• Predicting T2DM pathology is feasible and effective using longitudinal EHR data.
• The proposed method offers a non-invasive alternative to current invasive classification techniques.
• Patient historical data and progression patterns are valuable for disease prediction.