Phenotypic clustering of patients hospitalized in intensive cardiac care units: Insights from the ADDICT-ICCU study

Abstract

Background: Intensive cardiac care units (ICCUs) were created to manage ventricular arrhythmias after acute coronary syndromes, but have diversified to include a more heterogeneous population, the characteristics of which are not well depicted by conventional methods.

Aims: To identify ICCU patient subgroups by phenotypic unsupervised clustering integrating clinical, biological, and echocardiographic data to reveal pathophysiological differences.

Methods: During 7-22 April 2021, we recruited all consecutive patients admitted to ICCUs in 39 centers. The primary outcome was in-hospital major adverse events (MAEs; death, resuscitated cardiac arrest or cardiogenic shock). A cluster analysis was performed using a Kamila algorithm.

Results: Of 1499 patients admitted to the ICCU (69.6% male, mean age 63.3±14.9 years), 67 (4.5%) experienced MAEs. Four phenogroups were identified: PG1 (n=535), typically patients with non-ST-segment elevation myocardial infarction; PG2 (n=444), younger smokers with ST-segment elevation myocardial infarction; PG3 (n=273), elderly patients with heart failure with preserved ejection fraction and conduction disturbances; PG4 (n=247), patients with acute heart failure with reduced ejection fraction. Compared to PG1, multivariable analysis revealed a higher risk of MAEs in PG2 (odds ratio [OR] 3.13, 95% confidence interval [CI] 1.16-10.0) and PG3 (OR 3.16, 95% CI 1.02-10.8), with the highest risk in PG4 (OR 20.5, 95% CI 8.7-60.8) (all P<0.05).

Conclusions: Cluster analysis of clinical, biological, and echocardiographic variables identified four phenogroups of patients admitted to the ICCU that were associated with distinct prognostic profiles.

Trial registration: ClinicalTrials.gov identifier: NCT05063097.

Keywords: Acute cardiac event; Cardiac intensive care unit; Clustering analysis; Heart failure; Unsupervised machine learning.