Identifying Cardiogenic Shock Sub-Phenotypes with Machine Learning: A Multicenter Study Combining Clinical and Echocardiographic Data

doi:10.1016/j.jacadv.2025.102257

Acute cardiac care & resuscitation

Identifying Cardiogenic Shock Sub-Phenotypes with Machine Learning: A Multicenter Study Combining Clinical and Echocardiographic Data

JACC AdvPublished 28 Oct 2025Prospective multicentre observational cohort study

Read the original article ↗

Cite

Authors' conclusion

Machine learning, newly integrating echocardiographic data, identified 5 distinct CS phenotypes, each with unique clinical/echocardiographic features and mortality risks.

Study context

Population: Patients admitted to cardiac ICUs with overt cardiogenic shock, from 2021
Participants: 172 patients

Exposure: Cardiogenic shock (CS) phenotypes IV and V vs phenotype I
Outcome: In-hospital mortality

Methodological detailsunsupervised machine learning (K-Means clustering)

Unsupervised machine learning (K-Means clustering)

Main finding

As reported in the abstract

In-hospital mortality

I25%

II32%

III39%

IV41%

V60%

Each group vs IHazard ratio (95% CI)P value

IV

2.78(1.07–7.19)

P < 0.05

V

2.80(1.10–7.14)

P < 0.05

Observational data: associations, not effects.

Further findings

Mortality prediction after adjustment

Remained independent after adjustment for confounders and of SCAI stage

Original publication

Identifying Cardiogenic Shock Sub-Phenotypes with Machine Learning: A Multicenter Study Combining Clinical and Echocardiographic Data. JACC Adv. Published 28 October 2025. doi:10.1016/j.jacadv.2025.102257

Read the original article ↗PubMed

Summarised from the published abstract; every number is checked against it