Predicting Treatment Failure With Sodium-Glucose Cotransporter-2 Inhibitors in People With Type 2 Diabetes: Novel Artificial Intelligence and Machine Learning Approach
A machine learning score identifies patients likely to struggle with a common diabetes medication, supporting more tailored prescribing decisions in routine practice.
In the largest ML study of SGLT2i treatment outcomes to date (n=62,222), the majority of patients experienced treatment failure, and ML models achieved moderate but not excellent prediction accuracy. While more advanced models (XGBoost, Transformer) only marginally outperformed logistic regression, the 9-feature SHAP-derived risk score provides a framework for clinical decision support in SGLT2i prescribing.
What the study was
- Study design
- Retrospective observational cohort study with ML model development
- Population
- Adults with type 2 diabetes initiating SGLT2i therapy (2016-2024)
- Sample size
- 62222
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- JMIR Diabetes
Why it surfaced
Large real-world cohort (n=62,222); direct clinical relevance given SGLT2i prescribing volumes; transparent SHAP methodology; null finding for advanced ML vs LR is itself informative for the field.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.