Fully Connected Digital Ecosystems within Hospitals – AI/ML Solutions for Improved Patient Care

Abstract

Cardiogenic shock (CS) in the context of acute myocardial infarction (AMI) remains a significant challenge in critical care, with high mortality rates despite the availability of advanced mechanical circulatory support (MCS) devices like the Impella pump. However, adoption of these devices in clinical practice remains limited. This thesis explores two complementary strategies to address these challenges: developing machine learning (ML) models to predict shock severity and assessing the feasibility of integrating hospital Electronic Medical Record (EMR) data into Abiomed’s digital ecosystem to support standardized shock care. In the first phase, ML models were trained on multiple clinical datasets to predict Society for Cardiovascular Angiography and Interventions (SCAI) shock stages based on patient data. While these models demonstrated strong predictive performance, feature analysis revealed that SCAI stages often reflect physician treatment decisions rather than purely patient physiology. This raises concerns about their utility as real-time clinical decision tools and suggests that ML applications may be better suited to prompting early data collection and intervention before severe shock develops. The second phase evaluated the feasibility of EMR integration to support the broader adoption of standardized shock protocols. After considering regulatory, operational, and technical factors, third- party data aggregation emerged as the most practical path forward. Integrating EMR data could improve outcome tracking, support protocol adoption, and strengthen partnerships between Abiomed and hospitals, creating a foundation for more consistent and proactive shock management. Together, these findings highlight the need for predictive tools that guide early clinical action and infrastructure that supports seamless data integration. By advancing both, Abiomed can expand its role in cardiogenic shock care, improve patient outcomes, and lead the evolution of data-driven, standardized treatment strategies.