Youth-Onset T2D Clusters Do Not Outperform Simple Risk Measures

Despite heterogeneity in youth-onset type 2 diabetes (T2D), clustering does not currently outperform simple clinical measures in predicting risk for complications, according to study results published in Diabetes Care.

Clustering has emerged as a promising strategy to shift diabetes care from a one-size-fits-all model toward precision medicine. However, given the gaps in knowledge about youth-onset T2D compared with adult-onset diabetes, the comparative clinical utility between clustering and simple clinical measures remains unclear.

In this observational cohort study, researchers performed phenotypic clustering of youths with T2D onset to assess whether distinct clusters offered meaningful clinical utility.

The study included individuals diagnosed with T2D before 20 years of age from the SEARCH for Diabetes in Youth (SEARCH; n=333) and Treatment Options for Type 2 Diabetes in Adolescents and Youth (TODAY; n=525) studies. Participants were excluded if they were taking a cholesterol-lowering medication or had a positive pancreatic autoantibody test.

Nevertheless, cluster membership did not add clinical utility beyond simple clinical measures for predicting outcomes.

In a diverse subset of autoantibody-negative T2D cases, the researchers conducted data-driven clustering into 3 clusters: youth-onset insulin-deficient diabetes (YIDD-T2), youth-onset insulin-resistant diabetes, and intermediate youth-onset diabetes. Differences among clusters were tested using Kruskal–Wallis with Wilcoxon post hoc comparisons for continuous variables and χ² tests for categorical variables.

The primary outcome was loss of glycemic control and treatment failure as defined by hemoglobin A1c (HbA_1c) of at least 8% over 6 months or the need for insulin.

Three clinically distinct clusters with different rates of treatment failure emerged in both SEARCH and TODAY. Cluster 1 individuals (25.5%) showed the highest fasting glucose, HbA_1c, and triglycerides; Cluster 2 individuals (40.5%) had the highest body mass index (BMI) z-scores; and Cluster 3 individuals (33.9%) demonstrated the highest high-density lipoprotein (HDL) and lowest fasting glucose, HbA_1c, and triglycerides.

Cluster distributions did not differ by race or Tanner stage but varied significantly by sex (P =.004) and medication use (P <.0001): Cluster 3 included the most women (73%) whereas Cluster 2 included the fewest women (53%); Cluster 2 included the highest proportion of individuals on metformin alone (58%), whereas Cluster 1 included the most insulin users (60%).

In the TODAY trial, these clusters showed divergent therapeutic responses and treatment failure risks, with the YIDD-T2 group consistently exhibiting the highest failure rates and the greatest burden of diabetes complications.

Study limitations include the possibility that different input variables may yield different optimal k values; the absence of essential biomarkers, such as complete Tanner staging and liver transaminases; the likelihood that additional complications will emerge with follow-up beyond 10 years; and potential differential risk estimates due to limited follow-up duration.

The study authors concluded, “Youth-onset type 2 diabetes can be characterized into reproducible clusters that demonstrate differential response to treatments and risk of complications. Nevertheless, cluster membership did not add clinical utility beyond simple clinical measures for predicting outcomes.”