Phenotypic heterogeneity among patients with type 2 diabetes mellitus (T2DM) and atherosclerotic cardiovascular disease (ASCVD) is ill defined. We used cluster analysis machine-learning algorithms to identify phenotypes among trial participants with T2DM and ASCVD.
We used data from the Trial Evaluating Cardiovascular Outcomes with Sitagliptin (TECOS) study (n = 14,671), a cardiovascular outcome safety trial comparing sitagliptin with placebo in patients with T2DM and ASCVD (median follow-up 3.0 years). Cluster analysis using 40 baseline variables was conducted, with associations between clusters and the primary composite outcome (cardiovascular death, nonfatal myocardial infarction, nonfatal stroke, or hospitalization for unstable angina) assessed by Cox proportional hazards models. We replicated the results using the Exenatide Study of Cardiovascular Event Lowering (EXSCEL) trial.
Four distinct phenotypes were identified: cluster I included Caucasian men with a high prevalence of coronary artery disease; cluster II included Asian patients with a low BMI; cluster III included women with noncoronary ASCVD disease; and cluster IV included patients with heart failure and kidney dysfunction. The primary outcome occurred, respectively, in 11.6%, 8.6%, 10.3%, and 16.8% of patients in clusters I to IV. The crude difference in cardiovascular risk for the highest versus lowest risk cluster (cluster IV vs. II) was statistically significant (hazard ratio 2.74 [95% CI 2.29–3.29]). Similar phenotypes and outcomes were identified in EXSCEL.
In patients with T2DM and ASCVD, cluster analysis identified four clinically distinct groups. Further cardiovascular phenotyping is warranted to inform patient care and optimize clinical trial designs.
This article contains supplementary material online at https://doi.org/10.2337/figshare.16722358.