Non-Linear Dose-Response Relationship for Metformin in Japanese Patients With Type 2 Diabetes: Analysis of Irregular...

Area of Science:

• Pharmacology and Endocrinology
• Data Science in Medicine

Background:

• The dose-response of metformin for HbA1c reduction in Japanese patients with type 2 diabetes (T2D) is not well-established.
• Previous studies in the U.S. suggest a maximum effect at 1500-2000 mg/day.
• Metformin dosage guidelines in Japan were updated in 2010, necessitating updated research.

Purpose of the Study:

• To investigate the saturation of the metformin dose-response relationship in Japanese T2D patients.
• To determine the optimal metformin dosage for HbA1c reduction in this population.
• To apply advanced machine learning for analyzing clinical data.

Main Methods:

• Extracted longitudinal clinical data from electronic medical records of T2D patients.
• Constructed supervised machine learning models, including GLMM and GPBoost, to predict HbA1c change.
• Interpreted the GPBoost model using SHAP and partial dependence for understanding the dose-response relationship.

Main Results:

• GPBoost demonstrated superior predictive performance compared to GLMM (RMSE 0.602 vs. 0.698/0.678).
• The relationship between metformin dose and HbA1c change was non-linear, indicating saturation.
• The HbA1c-lowering effect of metformin appeared to plateau at approximately 1500 mg/day.

Conclusions:

• Metformin's dose-response relationship in Japanese T2D patients exhibits saturation.
• The optimal dose for HbA1c reduction in this population is around 1500 mg/day.
• Machine learning interpretation supports informed pharmacotherapy decisions for T2D management.