Related Experiment Video
Updated: Jul 25, 2025
Utilizing Percutaneous Ventricular Assist Devices in Acute Myocardial Infarction Complicated by Cardiogenic Shock
Published on: June 12, 2021
Clinical Course of Patients in Cardiogenic Shock Stratified by Phenotype
Elric Zweck1, Manreet Kanwar2, Song Li3
1The CardioVascular Center, Tufts Medical Center, Boston, Massachusetts, USA; Division of Cardiology, Pulmonology, and Vascular Medicine, University Hospital Dusseldorf, Dusseldorf, Germany.
Insights
Three distinct cardiogenic shock (CS) phenotypes were confirmed, showing similar mortality rates and clinical features across different patient groups. These validated CS phenotypes can guide future clinical trials and personalized treatment strategies.
Area of Science:
- Cardiology
- Machine Learning in Medicine
- Critical Care Medicine
Background:
- Cardiogenic shock (CS) has high in-hospital mortality (30-60%) despite advances, complicated by its heterogeneity.
- Three distinct CS phenotypes (noncongested, cardiorenal, cardiometabolic) were previously identified in the Cardiogenic Shock Working Group (CSWG) registry V1.
- These phenotypes present challenges for clinical trial design and patient management.
Purpose of the Study:
- To externally validate machine learning-based CS phenotypes.
- To define the clinical course and outcomes associated with each CS phenotype.
- To assess the reproducibility of CS phenotypes in an independent cohort.
Main Methods:
- Utilized 1,890 all-cause CS patients from the CSWG registry version 2.
- Applied nearest centroids method to identify CS phenotypes based on initial clusters.
- Retrospectively identified phenotypes in 796 patients for analysis.
Main Results:
- In-hospital mortality rates for phenotypes I, II, and III were 23%, 41%, and 52%, consistent with V1 findings (21%, 45%, 55%).
- Phenotype-specific demographic, hemodynamic, and metabolic characteristics were reproducible.
- Mechanical circulatory support use varied by phenotype (58.8% in I, 45.7% in II, 51.9% in III) and was linked to increased mortality in cardiorenal CS (OR 1.82).
- Phenotypes II and III, and SCAI stage E, independently predicted increased mortality.
Conclusions:
- The study supports the universal applicability of these machine learning-derived CS phenotypes.
- These validated phenotypes can inform the design of future clinical trials for CS.
- Tailored management algorithms based on specific CS phenotypes are feasible and recommended.
Background:
Cardiogenic shock (CS) patients remain at 30% to 60% in-hospital mortality despite therapeutic innovations. Heterogeneity of CS has complicated clinical trial design. Recently, 3 distinct CS phenotypes were identified in the CSWG (Cardiogenic Shock Working Group) registry version 1 (V1) and external cohorts: I, "noncongested;" II, "cardiorenal;" and III, "cardiometabolic" shock.
Objectives:
The aim was to confirm the external reproducibility of machine learning-based CS phenotypes and to define their clinical course.
Methods:
The authors included 1,890 all-cause CS patients from the CSWG registry version 2. CS phenotypes were identified using the nearest centroids of the initially reported clusters.
Results:
Phenotypes were retrospectively identified in 796 patients in version 2. In-hospital mortality rates in phenotypes I, II, III were 23%, 41%, 52%, respectively, comparable to the initially reported 21%, 45%, and 55% in V1. Phenotype-related demographic, hemodynamic, and metabolic features resembled those in V1. In addition, 58.8%, 45.7%, and 51.9% of patients in phenotypes I, II, and III received mechanical circulatory support, respectively (P = 0.013). Receiving mechanical circulatory support was associated with increased mortality in cardiorenal (OR: 1.82 [95% CI: 1.16-2.84]; P = 0.008) but not in noncongested or cardiometabolic CS (OR: 1.26 [95% CI: 0.64-2.47]; P = 0.51 and OR: 1.39 [95% CI: 0.86-2.25]; P = 0.18, respectively). Admission phenotypes II and III and admission Society for Cardiovascular Angiography and Interventions stage E were independently associated with increased mortality in multivariable logistic regression compared to noncongested "stage C" CS (P < 0.001).
Conclusions:
The findings support the universal applicability of these phenotypes using supervised machine learning. CS phenotypes may inform the design of future clinical trials and enable management algorithms tailored to a specific CS phenotype.
Related Concept Videos
01:30Cardiomyopathy II: Dilated Cardiomyopathy
01:29Cardiomyopathy V: Interprofessional Care
01:17Pathophysiology of Heart Failure
01:30Heart Failure IV: Classification and Diagnostic Evaluation
01:25Cardiomyopathy I: Introduction and Classification
01:29Cardiomyopathy III: Hypertrophic Cardiomyopathy