What is the clinical evidence from this study?

Study design: Cohort. Population: Type 2 Diabetes (n=1333). Intervention: Phenotypic clustering models (discrete subgroup membership or continuous probabilities) vs. Original clinical variables. Primary outcome: Discrimination for mortality, CVD, and CKD (C-indices: mortality 0.696-0.700; CVD 0.642-0.647; CKD 0.669-0.677).

What does this research mean for the field?

Established phenotypic clustering of Type 2 diabetes into subgroups shows limited separation, substantial instability over time, and provides similar prognostic performance for mortality, CVD, and CKD compared to standard clinical variables. Novelty: ClaimNovelty.INCREMENTAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

This study aims to evaluate the phenotypic clustering of type 2 diabetes into specific subgroups and assess their prognostic implications.

June 7, 2026

2351-P: Assessment of Phenotypic Clustering of Incident and Prevalent Type 2 Diabetes: A MESA Pilot Study in the Definition, Etiology, Function: Integration to Enhance Type 2 Diabetes Treatment (DEFINE-Type 2 Diabetes) Consortium

Key Result

Phenotypic clustering of Type 2 Diabetes showed substantial subgroup transitions and similar prognostic performance for mortality, CVD, and CKD compared with standard clinical variables.

Key Points

This study aims to evaluate the phenotypic clustering of type 2 diabetes into specific subgroups and assess their prognostic implications.
Utilized established k-means clustering models in 871 prevalent and 462 incident T2D cases.
Analyzed data across four exams (2000-2007) including age at onset, BMI, HbA1c, HOMA2-IR, and HOMA2-β.
Evaluated mortality and disease outcomes using three models in 869 prevalent cases with long-term follow-up.
MARD was the most common subgroup (50% prevalent; 64% incident), whereas SIDD was rare in incident cases (0.6%).
Cumulative incidence of mortality, CVD, and CKD were 38%, 32%, and 50%, with modest discrimination performance across models.
Transition rates were notable, with SIDD instability showing 21-42% retention across exams and higher mortality rates (4-10%).

Study Design

Type

Cohort (n=1,333)

Multicenter

Yes

Structured PICO

Does phenotypic clustering of T2D improve prognostic discrimination for mortality, CVD, and CKD compared to standard clinical variables?

Population

1,333 prevalent and incident Type 2 Diabetes cases from the Multiethnic Study of Atherosclerosis followed for up to 18 years.

Exposure

Established k-means clustering models for T2D subgroups (MARD, MOD, SIDD, SIRD)

Comparator

Original clinical variables (age at onset, BMI, HbA1c, HOMA2-IR, and HOMA2-β)

Outcome

Discrimination for mortality, CVD, and CKDhard clinical

Established T2D phenotypic clustering does not improve prognostic discrimination for mortality, CVD, or CKD over standard clinical variables in a multiethnic cohort.

Main Result

Effect estimate: C-indices: mortality 0.696-0.700; CVD 0.642-0.647; CKD 0.669-0.677

Limitations

Did not account for diabetes duration and progression

Abstract

Introduction and Objective: Phenotypic clustering of T2D into aging-related (MARD), obesity-related (MOD), insulin-deficient (SIDD), and insulin-resistant (SIRD) subgroups may capture heterogeneity, but performance and stability in diverse populations are uncertain. Methods: We performed established k-means clustering models in 871 prevalent and 462 incident T2D cases with data for age at onset, BMI, HbA1c, HOMA2-IR, and HOMA2-β in the Multiethnic Study of Atherosclerosis across four exams (2000-2007). We assessed subgroup distribution, transitions, and mortality between exams. In 869 prevalent cases with long-term follow-up, we evaluated discrimination for mortality, CVD, and CKD using three models: discrete subgroup membership, continuous subgroup probabilities, or the original variables. Results: MARD was the most common subgroup (50% of prevalent; 64% of incident), while SIDD was rare among incident T2D (0.6%). Incident cases had higher BMI and HOMA2-β and lower HbA1c than prevalent T2D. Silhouette indices were modest (0.22-0.23). Across exams, MARD and SIRD increased in prevalence, and SIDD showed the greatest instability (only 21-42% remaining SIDD) and higher between exam mortality (4-10%). Over 18 years, cumulative incidence of mortality, CVD, and CKD were 38%, 32%, and 50%. Discrimination of these outcomes was similar across the three models (C-indices: mortality 0.696-0.700; CVD 0.642-0.647; CKD 0.669-0.677). Conclusion: In a multiethnic cohort, established T2D clustering shows limited separation, substantial subgroup transitions—especially for SIDD—and similar prognostic performance compared with clinical variables. Accounting for diabetes duration and progression may improve future clustering approaches. Disclosure L. Olson: None. F. Hsu: None. K. Smith: None. R. Dagostino Jr: Other - I am on a DSMB for this company; Current; Daiichi Sankyo. Consultant; Ended; Merck Current; AstraZeneca. M. Bancks: None. M. Udler: Advisory Panel; Ended; Novo Nordisk. Research Support; Current; Novo Nordisk. Funding National Institutes of Health (1U01DK140778 and 1U01DK140757)

Bookmark