Model could help select initial glucose-lowering therapy for type 2 diabetes

A new model could help predict the effectiveness of glucose-lowering therapy in patients with type 2 diabetes.

Researchers in the United Kingdom developed and validated a model to predict the relative glycemic effectiveness of initiating dipeptidyl peptidase-4 inhibitors, glucagon-like peptide-1 receptor agonists, sodium-glucose cotransporter-2 inhibitors, sulfonylureas, or thiazolidinediones in patients with type 2 diabetes. Nine routinely available variables were included in the model: age, duration of diabetes, sex, baseline body mass index, and baseline levels of HbA1c, estimated glomerular filtration rate, high-density lipoprotein cholesterol, total cholesterol, and alanine aminotransferase. The model was developed and validated with observational data from England in people with type 2 diabetes who were 18 to 79 years of age and initiating therapy with one of the five drug classes between Jan 1, 2004, and Oct 14, 2020. Holdback validation was done by geographic region and by time, with further validation using individual-level data from three published randomized trials.

For validation using observational data, differences in observed glycemic effectiveness were assessed between matched groups who were receiving therapy that was concordant or discordant with the optimal therapy predicted by the model. “Optimal” was defined as the drug class with the highest predicted glycemic effectiveness, or the lowest predicted HbA1c level at 12 months. Additional validation involved pairwise comparisons of drug classes in all datasets. The researchers also assessed five-year risks for glycemic failure (confirmed HbA1c level ≥69 mmol/mol [≥8.5%]), all-cause mortality, major adverse cardiovascular events or heart failure, renal progression, and microvascular complications. The results were published March 1 by The Lancet.

A total of 100,107 drug initiations were used to develop the model. In the initial development and initial validation cohorts, 32,305 (15.2%) of 212,166 initiations of therapy used the drug class predicted as optimal. In model-concordant groups, mean observed benefit of HbA1c at 12 months was 5.3 mmol/mol (2.6%) in the geographic validation cohort (12,373 matched pairs) and 5.0 mmol/mol (2.6%) in the temporal validation cohort (4,841 matched pairs) versus matched model-discordant groups. Predicted differences in HbA1c were well calibrated with observed differences in pairwise drug-class comparisons using the three clinical trials and the observational cohorts.

The five-year risk for glycemic failure was lower in model-concordant versus model-discordant groups in the observational cohorts (adjusted hazard ratio [aHR], 0.62; 95% CI, 0.59 to 0.64). Model-concordant versus model-discordant groups had similar five-year risks for all-cause mortality (aHR, 0.95; 95% CI, 0.83 to 1.09) and lower risks for major adverse cardiovascular events or heart failure (aHR, 0.85; 95% CI, 0.76 to 0.95), renal progression (aHR, 0.71; 95% CI, 0.64 to 0.79), and microvascular complications (aHR, 0.86; 95% CI, 0.78 to 0.96).

The researchers noted that active comparator trial data were not available for glucagon-like peptide-1 receptor agonists and that they could not include semaglutide or tirzepatide in their models, among other limitations. “In conclusion, we have developed and validated an individualised treatment selection model for people with type 2 diabetes, based on glycaemic effectiveness and considering five major glucose-lowering drug classes. The model is based solely on routinely collected clinical parameters, supporting low-cost application worldwide,” they wrote. “Implementation could lead to improvements in glycaemic control, a marked increase in time on stable glucose-lowering therapy before additional intensification, and a potential reduction in diabetes complications.” A calculator demonstrating the five-drug class model is available online.