Truace
Subscribe

Enhanced Early Risk Stratification in Cardiogenic Shock Using Machine Learning: A Hierarchical Clustering Analysis

Background Cardiogenic shock (CS) exhibits marked clinical heterogeneity, limiting the performance of conventional risk stratification tools like the Society for Cardiovascular Angiography and Interventions (SCAI) staging system. Machine learning (ML) has identified reproducible CS phenotypes that improve risk prediction, but their temporal consistency across the progression of shock remains incompletely described. Objectives The purpose of this study was to evaluate the temporal evolution of ML-derived CS pheno…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-2037· cite & verify
Enhanced Early Risk Stratification in Cardiogenic Shock Using Machine Learning: A Hierarchical Clustering Analysis
Soins intensifs by Celeda. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0

In brief

Researchers analyzed a multicenter retrospective registry of 2,906 adults with cardiogenic shock admitted to intensive care units between 2016 and 2022. They applied hierarchical agglomerative clustering separately to clinical variables on ICU days 1, 2 and 3, and three patient phenotypes emerged each day.

The finding matters because conventional staging struggles with the clinical diversity of cardiogenic shock, and the phenotypes added mortality information within every SCAI stage. Because the data were retrospective, the approach has not been shown to change care, and the authors state that prospective validation is needed.

Main points

  1. The Cardiac phenotype was younger, mostly acute myocardial infarction, had preserved renal function and 15% mortality.
  2. The Cardio-Renal phenotype was older with isolated renal impairment, mostly SCAI A and B, and had 27% mortality.
  3. The Cardio-Metabolic phenotype had severe renal and tissue malperfusion, mostly SCAI D and E, and had 55% mortality.
  4. Phenotypes stayed consistent across ICU days 1 to 3 and reproduced previously identified phenotypes.

The gain

Clustering of clinical variables found three stable patient phenotypes in cardiogenic shock that separated mortality risk within each SCAI stage during the first three ICU days.

The rundown

The study addresses the limits of the SCAI staging system, which performs poorly partly because cardiogenic shock is clinically heterogeneous. Earlier machine learning work had found reproducible phenotypes, but their consistency over the progression of shock was not well described.

Using a registry covering 2016 to 2022, the team clustered clinical variables independently on each of the first three ICU days. This tested whether phenotype membership held steady over time and whether trajectories added to SCAI staging.

Mortality rose from 15% in the Cardiac group to 27% in Cardio-Renal and 55% in Cardio-Metabolic on day 1. The authors conclude that the framework improves early risk stratification, pending prospective validation.

Sources

  1. Peer-reviewedJACC: Advances2026-10-13

The debate