Clin Res Cardiol (2021) DOI DOI https://doi.org/10.1007/s00392-021-01843-w

Identification of Three Clinical Phenotypes of Cardiogenic Shock

E. Zweck1, K. L. Thayer2, O. K. L. Helgestad3, M. Kanwar4, M. Ayouty5, A. R. Garan6, J. Hernandez-Montfort7, C. Mahr8, D. Wencker9, S. Sinha10, E. Vorovich11, W. O'Neill12, J. Abraham13, G. W. Hickey14, J. Josiassen15, C. Hassager15, L. O. Jensen16, L. Holmvang15, H. Schmidt16, H. B. Ravn17, J. Moeller3, D. Burkhoff18, N. Kapur2

1The Cardiovascular Center, Tufts Medical Center, Boston; 2The Cardiovascular Center, Tufts Medical Center, Boston, US; 3Department of Cardiology, Odense University Hospital, Odense, DK; 4Allegheny Health Network, Pittsburgh, US; 5Tufts University School of Medicine, Boston, US; 6Beth Israel Deaconess Medical Center, Boston, US; 7Cleveland Clinic Florida, Weston, US; 8University of Washington Medical Center, Seattle, US; 9Baylor Scott & White Advanced Heart Failure Clinic, Dallas, US; 10Inova Heart and Vascular Institute, Falls Church, US; 11Northwestern Medicine, Chicago, US; 12Henry Ford Hospital, Detroit, US; 13Providence Heart Institute, Portland, US; 14UPMC Heart and Vascular Institute, Pittsburgh, US; 15Department of Cardiology, Rigshospitalet, Kopenhagen, DK; 16Department of Cardiothoracic Anesthesia, Odense University Hospital, Odense, Dänemark; 17Department of Cardiac Anesthesiology, Rigshospitalet, Kopenhagen, DK; 18Cardiovascular Research Foundation, New York, US;

Aims: Cardiogenic shock (CS) is a heterogeneous syndrome that represents an acute and fulminant form of heart failure (HF). Clinical presentation and outcomes in CS are highly varied, based on etiology, demographics and treatment approach. We employed a data-driven, machine learning (ML) approach to test the hypothesis that CS patients can be subdivided at presentation into distinct phenotypes and that these phenotypes are associated with unique clinical profiles and in-hospital mortality. Methods: We included admission data from 1959 CS patients from two international cohorts: The Cardiogenic Shock Working Group Registry [(further defined by etiology as myocardial infarction (CSWG-MI, n=410) and acute on chronic heart failure (CSWG-HF, n=480))] and the Danish Retroshock Registry (DRR, n=1069). Consensus k means clustering of CSWG-MI was used for derivation, and the derived phenotypes were subsequently validated in the CSWG-HF and DRR cohorts. Patients in each phenotype were further categorized by the highest SCAI stage reached during hospitalization to assess their use in risk stratification. Results: The ML algorithms revealed three distinct clusters in CS: ‘non-congested (I)’, ‘cardio-renal shock (II)’ and ‘cardio-metabolic shock (III)’. Amongst the three cohorts (CSWG-MI vs DDR vs CSWG-HF), in-hospital mortality was 21% vs 28% vs 10%, 45% vs 40% vs 32%, and 55% vs 56% vs 52% for Clusters I, II, and III, respectively. Cluster III ‘cardio-metabolic shock’ had the highest mortality amongst the CS phenotypes, regardless of CS etiology. Despite baseline differences, each cluster showed a reproducible demographic, metabolic, and hemodynamic profile across the 3 cohorts. The risk of developing stage D or E shock during the hospital stay was lowest in Cluster I and highest in Cluster III for both, CSWG-MI and CSWG-HF patients. Within SCAI stages C-E, the phenotypes further stratified mortality and vice versa. Conclusion: Using ML we identified and externally validated three distinct CS phenotypes, with specific and reproducible associations with short-term mortality. These phenotypes may provide an additional approach to risk stratification, allow for targeted patient enrollment in clinical trials, and foster development of tailored treatment strategies in unique physiological subsets of CS patients.

https://dgk.org/kongress_programme/jt2021/aP197.html