Predicting Treatment Failure With Sodium-Glucose Cotransporter-2 Inhibitors in People With Type 2 Diabetes: Novel...

Area of Science:

• Diabetes Mellitus Research
• Artificial Intelligence in Healthcare
• Pharmacovigilance

Background:

• High rates of treatment failure with sodium-glucose cotransporter-2 inhibitors (SGLT2i) impact type 2 diabetes (T2D) management.
• Accurate prediction of SGLT2i treatment failure is crucial for optimizing clinical outcomes in T2D patients.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting SGLT2i treatment failure in T2D.
• To identify key factors contributing to SGLT2i treatment failure.

Main Methods:

• Retrospective observational cohort study of adults with T2D treated with SGLT2i (2016-2024).
• Utilized logistic regression, multilayer perceptron, XGBoost, and Transformer models to predict overall and subtype treatment failure.
• Employed Shapley Additive Explanations to identify significant predictors.

Main Results:

• 71% of patients experienced SGLT2i treatment failure, with 'failure with action' being the most common subtype.
• ML models demonstrated moderate predictive performance, with slight improvements from advanced models over logistic regression.
• Transformer models using key features performed comparably to those using the full feature set.

Conclusions:

• Current ML models offer moderate accuracy in predicting SGLT2i treatment failure, necessitating further advancements and larger datasets.
• Logistic regression coefficients from key features may guide the development of a risk score for predicting SGLT2i treatment failure.
• Improved prediction accuracy holds potential for personalized treatment strategies in T2D management.