Machine learning tool may help personalize type 2 diabetes treatment

A machine learning (ML)-based tool may help clinicians choose between two commonly prescribed diabetes medications and improve blood sugar control in patients with poorly controlled type 2 diabetes (T2D), according to a study published in Nature.

Juan Shi, MD, of the Shanghai Institute of Endocrine and Metabolic Diseases at Ruijin Hospital and Shanghai Jiao Tong University School of Medicine, both in China, and colleagues who participated equally in the study, developed the “TiP DecScore,” a model designed to guide selection between sodium-glucose cotransporter-2 inhibitors (SGLT-2i) and glucagon-like peptide-1 receptor agonists (GLP-1RA). Although both drug classes are widely used and offer heart and kidney benefits, clinicians have had limited tools to determine which is best suited for an individual patient.

"The improved glycemic control observed in patients adhering to model recommendations underscores the clinical utility of the TiP DecScore in facilitating individualized therapeutic decisions, thus enhancing treatment outcomes for patients with T2D," the authors said.

Data from more than 24,000 patients

To build the model, the researchers analyzed data from more than 24,000 patients in the China Metabolic Analytics Project and then tested it in a separate group of 1,459 patients. The tool uses 15 routine clinical factors such as body mass index, baseline HbA1c, and other lab values to predict which medication is more likely to help a patient reach target blood sugar levels.

The model showed good predictive performance, with “the receiver operating characteristic curve 0.71-0.78,” the authors reported. Overall, it more often recommended GLP-1RA therapy than SGLT-2i, particularly for patients with higher body weight, higher HbA1c levels, or shorter duration of diabetes.

Patients whose prescribed treatment matched the model’s recommendation were more likely to achieve good blood sugar control. The authors noted that higher rates of HbA1c control are observed in concordant versus discordant groups. The difference was especially notable at 12 months among patients younger than 55 years and among men.

The authors concluded that the TiP DecScore “effectively guides” personalized selection between SGLT-2i and GLP-1RA therapies for T2D patients.

“We found that patients whose medications matched TiP DecScore recommendations achieved better glycemic control compared with those with mismatched treatments,” they said.

The study included investigators affiliated with multiple institutions across China, including the Shanghai National Clinical Research Center for Metabolic Diseases and the School of Artificial Intelligence at Nanjing University.

The authors reported no competing interests and noted that the data supporting the findings are available within the article and its supplementary materials.