# Unsupervised Clustering of Routine Inflammatory Markers in Cardiogenic Shock Reveals Phenotypic Heterogeneity Without Prognostic Utility

## TL;DR

This study found that cardiogenic shock has different inflammatory patterns, but these patterns don't help predict short-term outcomes.

## Contribution

The study identifies distinct inflammatory phenotypes in cardiogenic shock using unsupervised clustering of routine markers, but finds no prognostic benefit.

## Key findings

- Two distinct inflammatory phenotypes were identified in cardiogenic shock patients using clustering of admission inflammatory markers.
- Despite biological differences between clusters, short-term clinical outcomes were similar across groups.
- Phenotypes showed differences in age, chronic kidney disease prevalence, and inflammatory profiles but not in mortality or hospital stay.

## Abstract

Background: Cardiogenic shock is a heterogeneous syndrome in which systemic inflammation may contribute to cardiovascular risk and adverse outcomes beyond hemodynamic compromise alone. Methods: We conducted a retrospective multicenter cohort study using the eICU Collaborative Research Database (2014–2015) to identify inflammatory phenotypes among adults admitted to intensive care units with cardiogenic shock. Inflammatory indices derived from admission hematologic parameters (including NLR, PLR, MLR, NPAR, SII, SIRI, and AISI) were analyzed using principal component analysis, followed by hierarchical and k-means clustering to identify biologically distinct inflammatory phenotypes. Clinical characteristics and short-term outcomes were compared across clusters. Results: Among 419 patients, two phenotypes were identified. Cluster 1 (n = 52) was characterized by older age, a higher prevalence of chronic kidney disease (CKD), more advanced renal and hepatic dysfunction, along with a hyperinflammatory, lymphopenic profile. Cluster 2 (n = 367) exhibited comparatively lower inflammatory indices and less biochemical derangement. There was a significant difference in the prevalence of CKD, the need for mechanical ventilation, and history of malignancy between clusters. Despite clear biological separation, short-term clinical outcomes, including rates of acute kidney injury requiring renal replacement therapy, vasopressor use, hospital length of stay, and in-hospital mortality, were similar across clusters. Conclusions: These findings suggest that cardiogenic shock encompasses distinct inflammatory phenotypes, but inflammatory clustering based on routine admission laboratory data alone may have limited utility for short-term risk stratification.