Will Dynamic Evaluation of Cardiogenic Shock Using Machine Learning Models Lead to Improved Survival?

Cardiogenic shock (CS) is characterised by tissue hypoxia as a result of circulatory failure arising from inadequate cardiac output and is commonly a complication of acute myocardial infarction (AMI). Despite improvement in reperfusion strategies for AMI, the survival among patients with CS remains poor. While mechanical circulatory support (MCS) technologies in AMI-CS offer promise, they have not translated to consistent improvements in patient survival, which may reflect an inability to recognise evolving CS at a reversible stage. Hence, reducing the mortality from CS requires solutions focused on timely diagnosis. CS is heterogenous, being dependent on interpreting acute haemodynamics and biomarkers, which often delays diagnosis and intervention. The continued digitisation of health information, particularly within the emergency and acute care environments has made the development of artificial intelligence (AI)-driven diagnostic decision support for the acutely deteriorating patient feasible. Such approaches have been effectively deployed in hospitals to alert frontline staff or "shock teams" to patient deterioration, with evidence of reductions in mortality. Further, these integrated systems that can "dynamically phenotype" patients and their clinical deterioration within the flow of data not only support clinical decision-making but also allow the establishment of virtual clinical registries assimilated within real-world practice, continuously evaluating clinical practice and outcomes. This review aims to delineate CS pathophysiology, limitations within our current diagnostic approach, understand the difficulties in conducting randomised clinical trials and explores the role of an integrated AI-based approach for early diagnosis and intervention in patients with CS.