Phenotyping Refractory Cardiogenic Shock Patients Receiving Venous-Arterial Extracorporeal Membrane Oxygenation Using Machine Learning Algorithms

This study used machine learning to categorize cardiogenic shock (CS) patients treated with venous-arterial extracorporeal membrane oxygenation (VA-ECMO) into distinct phenotypes. Subsequently, it aimed to clarify the wide mortality variance observed in refractory CS, attributing it to the condition's inherent heterogeneity. This study enrolled a cohort of CS patients who received VA-ECMO support. By employing rigorous machine learning (ML) techniques, we generated and validated clusters based on determinants identified through algorithmic analysis. These clusters, characterized by distinct clinical outcomes, facilitated the examination of clinical and laboratory profiles to enhance the understanding of patient responses to VA-ECMO treatment. In a study of 210 CS patients undergoing VA-ECMO treatment, 70.5% were male with a median age of 62, ranging from 53 to 67 years. Survival rates were 67.6% during VA-ECMO and 49.5% post-discharge. Patients were classified into three phenotypes based on the clinical and laboratory findings: "platelet preserved (I)", those with stable platelet counts, "hyperinflammatory (II)", those indicating significant inflammation, and "hepatic-renal (III)", those showing compromised liver and kidney functions. Mortality rates (25.0%, 52.8%, and 55.9% for phenotypes I, Ⅱ, and Ⅲ, respectively ( = 0.005)) varied significantly among these groups, highlighting the importance of phenotype identification in patient management. This study identified three distinct phenotypes among refractory CS patients treated using VA-ECMO, each with unique clinical characteristics and mortality risks. Thus, highlighting the importance of early detection and targeted intervention, these findings suggest that proactive management could improve outcomes for those showing critical signs.