Risk stratification at prediabetes onset and association with diabetes outcomes using EHR data.

Area of Science:

• Endocrinology
• Computational Medicine
• Public Health

Background:

• Prediabetes (PD) progression to type 2 diabetes (T2D) risk varies significantly among individuals.
• Limited research exists on characterizing subgroups at PD onset for precise risk stratification.
• Early identification of high-risk individuals is crucial for timely intervention.

Purpose of the Study:

• To develop and validate a machine learning model for stratifying prediabetes patients based on T2D progression risk.
• To identify distinct subgroups within the prediabetes population using real-world electronic health record (EHR) data.
• To enable personalized interventions and enhance diabetes risk prediction.

Main Methods:

• A high-fidelity cohort of 14,436 patients with prediabetes onset was curated from EHR data (2017-2023).
• An XGBoost machine learning model was trained using routine clinical features (HbA1c, BMI, lipids, etc.).
• Model performance was evaluated using AUC, and risk scores were used for patient subtyping.

Main Results:

• The XGBoost model achieved an AUC of 81.6% in predicting T2D progression risk.
• Patients were successfully stratified into high-, medium-, and low-risk groups with distinct progression trajectories.
• Stratification patterns demonstrated consistency across different time cohorts.

Conclusions:

• Machine learning applied to EHR data can effectively stratify prediabetes patients by T2D risk.
• This approach facilitates personalized risk assessment and supports early, targeted interventions.
• The developed model offers a valuable tool for diabetes risk prediction in clinical practice.