Unsupervised Agglomerative Cluster Phenotyping of Young Patients With Acute Coronary Syndrome Undergoing Percutaneous...

Young patients with acute coronary syndrome (ACS) exhibit diverse demographic, clinical and angiographic characteristics. We hypothesized that unsupervised machine learning (ML) can identify phenotypic clusters and predict the prognosis among young patients with ACS who undergo percutaneous coronary intervention (PCI).

We used the Houston Methodist Young-ACS PCI registry (2010-2022) and performed agglomerative clustering to analyze demographic, clinical, and angiographic variables among young patients (≤ 50 years) treated with PCI for ACS.

Among 452 patients, cluster analysis identified three distinct phenotypes. Low risk (n = 273) exhibited the lowest proportions of heart failure (HF: 10.3%) and peripheral arterial disease (PAD): 5.9%), normal left ventricular ejection fraction (LVEF) (median = 0.60) and shorter median culprit vessel stent length (16 mm). Moderate risk (n = 101) was characterized by moderate rates of diabetes (36.6%), hypertension (77.2%), prior myocardial infarction (16.8%), and prior PCI (18.8%). High risk (n = 78) had the highest proportion of diabetes (67.9%), hypertension (93.6%), HF (65.4%), PAD (17.9%), prior MI (41.0%), prior PCI (46.2%), the lowest median LVEF (0.3) and the most frequent use of mechanical circulatory device (5.1%). All-cause mortality differed significantly, with high-risk cluster showing a fourfold increase in all-cause mortality (HR: 4.50 [1.68-12.1]) compared to low-risk cluster; however moderate and low-risk clusters did not differ significantly (HR: 1.52 [0.44-5.19]). Sensitivity analysis based on two-clusters showed similar outcomes, indicating the robustness of the analysis.

Three distinct phenotypic clusters among young patients with ACS treated with PCI exhibited varying degrees of comorbidity burden and mortality risk, underscoring the potential for phenotype-guided therapeutic approaches.