Unsupervised clustering identifies distinct phenotypes in acute myocardial infarction: insights from the FAST-MI 2015...

Current classification of Acute Myocardial Infarction (AMI) into ST-elevation (STEMI) and non-ST-elevation (NSTEMI) myocardial infarction does not fully reflect the clinical heterogeneity of patients.

To identify clinically meaningful phenotypes of AMI patients using an unsupervised clustering approach and assess their associations with management strategies and long-term outcomes.

Hierarchical agglomerative clustering using Ward's method was performed in 4,947 patients from the French FAST-MI 2015 nationwide registry. Clustering was based on baseline clinical, demographic, and laboratory features. Between-cluster differences in baseline characteristics, management, in-hospital complications, and 1-year mortality were analyzed.

Four phenotypic clusters were identified. Cluster 1 (n = 1,488) gathered older patients, predominantly men, with typical coronary risk factors, intermediate risk profile, and balanced STEMI/NSTEMI presentations. Cluster 2 (n = 1,305) gathered older polymorbid patients, 45% were women, 60% with NSTEMI presentation, with limited use of invasive strategies (71%), and a high 1-year mortality. Cluster 3 (n = 815) gathered young (mean age 44 years), predominantly male patients, with a high prevalence of smoking (73%) and few comorbidities, 62% STEMI, high rates of revascularization, and excellent prognosis. Cluster 4 (n = 1,339) was close to Cluster 3 with a more metabolic profile with higher rates of obesity, diabetes, and dyslipidemia, also with favorable outcomes. In-hospital complications were more frequent in Clusters 1 and 2. One-year mortality was lowest in Clusters 3 and 4 (1.5 and 2.3%), intermediate in Cluster 1 (6.0%), and highest in Cluster 2 (15.7%).

Unsupervised clustering identified four clinically relevant AMI phenotypes with distinct characteristics, management patterns, and prognoses. This data-driven classification highlights the diversity of AMI presentations and may offer complementary insights for patient profiling and risk stratification.