Phenotyping Refractory Cardiogenic Shock Patients Receiving Venous-arterial Extracorporeal Membrane Oxygenation with Machine Learning Algorithms

Abstract

Abstract Background Refractory cardiogenic shock (CS) is a heterogeneous clinical condition differing widely in mortality. This research phenotyped CS patients receiving venous-arterial extracorporeal membrane oxygenation (VA-ECMO) by machine learning algorithm to explain the potential heterogeneity. Methods A prospective cohort of CS patients receiving VA-ECMO support were enrolled and analyzed. After strict machine learning (ML) methods generating and verifying cluster-determined variables, algorithm based on these covariates generated certain clusters with distinct clinical outcomes, hence the clinical and laboratory profiles were analyzed. Results Among 210 CS patients receiving ECMO, 148 (70.5%) were men, with a median age of 62 years. Overall, 142 (67.6%) survived on ECMO, and 104 (49.5%) patients survived to discharge. The patients were phenotyped into three clusters: (1) “platelet preserved (I)” Phenotype [36 (17.1%) patients], characterized by preserved platelet count; (2) “hyperinflammatory (II)” phenotype [72 (34.3%) patients], characterized by a significant inflammatory state; and (3) “hepatic-renal (III)” phenotype [102 (48.6%) patients], characterized by unfavorable conditions in hepatic and renal functions tests. The in-hospital mortality rates were 25.0%, 52.8%, and 55.9% for phenotypes I, II, and III, respectively (P = 0.005). Conclusion The research explored three phenotypes in refractory CS patients receiving VA-ECMO with distinct clinical profile and mortality. Early recognition and intervention can conduce to manage patients presenting unfavorable signs.