Effective Treatment Recommendations for Type 2 Diabetes Management Using Reinforcement Learning: Treatment...

Area of Science:

• Artificial Intelligence in Medicine
• Computational Health
• Personalized Medicine

Background:

• Type 2 diabetes mellitus (T2DM) and its complications pose a significant economic burden.
• A gap exists between recommended and actual patient treatment, necessitating personalized approaches.
• Individual patient variability and complex therapeutic targets challenge T2DM management.

Purpose of the Study:

• To develop and evaluate deep reinforcement learning models for T2DM treatment recommendations.
• To assess the reliability and effectiveness of AI-driven treatment strategies.
• To personalize antiglycemic, antihypertensive, and lipid-lowering therapies.

Main Methods:

• Developed hybrid knowledge- and data-driven models using deep reinforcement learning.
• Utilized a large dataset of 189,520 T2DM patients from Singapore Health Services.
• Evaluated model effectiveness by comparing outcomes of model-concordant vs. non-concordant treatments.

Main Results:

• Model recommendations showed concordance rates of 43.3% (antiglycemic), 51.3% (antihypertensive), and 58.9% (lipid-lowering).
• Model-concordant treatments correlated with improved glycemic (OR 1.73), blood pressure (OR 1.26), and lipid control (OR 1.28).
• Higher concordance with model recommendations was linked to reduced diabetes complications and mortality risk.

Conclusions:

• Combining knowledge- and data-driven models offers potential for improved T2DM clinical outcomes.
• AI-driven recommendations can enhance control of blood glucose, blood pressure, and lipids.
• Personalized treatment strategies may reduce long-term diabetes complications and mortality.