Optimising test intervals for individuals with type 2 diabetes: A machine learning approach.

Area of Science:

• Biomedical Informatics
• Clinical Decision Support Systems
• Personalized Medicine

Background:

• Current chronic disease monitoring often uses a one-size-fits-all approach, failing to account for individual patient needs and risk levels.
• Machine learning (ML) presents an opportunity to advance personalized medicine in clinical practice for better chronic disease management.

Purpose of the Study:

• To explore the application of ML in optimizing resource allocation for chronic disease management.
• To predict individualized HbA1c testing intervals for patients with Type 2 Diabetes (T2D).
• To assess the fairness of ML models across different socioeconomic factors and evaluate risks of prediction errors.

Main Methods:

• Utilized Danish administrative and clinical data from over 57,000 T2D patients (2015-2018).
• Employed logistic regression, random forest, and XGBoost to classify HbA1c test intervals (3, 6, 9, 12 months).
• Assessed model performance using ROC-AUC and feature importance with SHAP; determined optimal intervals based on 'time-to-next-positive-test' concept.

Main Results:

• ML models demonstrated varying predictive accuracy (AUC 0.53-0.89) for optimal HbA1c testing intervals.
• Identified potential for resource optimization by extending testing intervals for well-controlled T2D patients.
• Fairness metrics indicated good performance across income and education levels, but highlighted a notable risk of false negatives.

Conclusions:

• ML holds significant potential for personalized diabetes management by aiding physicians in setting appropriate patient testing frequencies.
• Clinical validation across diverse patient populations is essential to confirm the real-world effectiveness and safety of ML-driven testing interval recommendations.