The main goal of this observational, study is to develop a clinical decision support tool utilizing Impella 5.5 pump parameters to predict native heart recovery and prevent adverse events, by leveraging data science and real-world clinical data of cardiogenic shock patients.

Therefore, secondary objectives are essential to consolidating a retrospective longitudinal analysis of Impella 5.5 pump data alongside ICU digital health record datasets to:

1. Validate the Impella 5.5 placement signal by comparing it with ICU arterial line waveforms.
2. Integrate pump data with ICU clinical data to identify patterns associated with therapy outcomes, including native heart recovery, heart replacement therapy, and mortality while on device support.
3. Define clinical scenarios linked to hemolysis, HRAEs, and arrhythmias and develop predictive models to mitigate their occurrence.

The clinical management of patients experiencing severe cardiogenic shock requires precise, real-time monitoring to optimize hemodynamic support and guide therapeutic transitions. The Impella 5.5 micro-axial flow pump provides left ventricular unloading, generating automated internal continuous parameters that reflect moving cardiac states. This study establishes a retrospective, longitudinal framework that integrates these high-frequency device metrics with corresponding clinical data housed within intensive care unit (ICU) digital health records (DHR). By synthesizing these disparate data streams, this research aims to build an advanced analytical framework to support clinical decisions in the cardiogenic shock landscape.

Signal Validation and Data Preprocessing:

The initial phase of the study validates the physiological fidelity of the continuous data stream. High-frequency digital logs generated by the pump console-specifically the optical placement signal-will undergo time-series alignment with standard physiological waveforms recorded in the ICU, using indwelling arterial line pressure data as the reference standard. This signal validation ensures that the longitudinal parameter data accurately capture mechanical positioning and true left ventricular dynamics prior to entering the downstream modeling pipeline.

Analytical Framework and Modeling Strategy:

Following data integration and signal validation, the consolidated dataset will be leveraged to develop predictive models aimed at distinguishing patient trajectories and forecasting complications. The computational pipeline is divided into two primary analytical pathways:

Endpoint Classification:

An artificial neural network will be developed to evaluate patient trajectories toward distinct clinical endpoints: native heart recovery, escalation to heart replacement therapy, or death. The modeling pipeline incorporates a rigorous framework to ensure generalizability and guard against overfitting. The complete dataset will be partitioned into an 80% development subset and a 20% independent testing subset. The development subset will undergo 5-fold cross-validation to drive comprehensive model architecture optimization, systematically testing structural variations to identify the highest-performing network configuration.

Adverse Event Forecasting:

Separate statistical and machine learning architectures will be constructed to evaluate risk patterns and clinical scenarios associated with severe on-device complications, specifically clinical hemolysis, new-onset arrhythmias, and hemocompatibility-related adverse events (HRAEs). These models focus on identifying early-warning clusters within the high-frequency pump log data to identify sub-clinical changes before manifest physiological degradation occurs.

Through these combined pathways, this observational study seeks to lay the foundational algorithmic groundwork for a real-time clinical decision support tool utilizing objective, automated device analytics to improve safety and personalization in mechanical circulatory support.

Inclusion Criteria:

• Adult patients who were treated for cardiogenic shock and supported with an Impella 5.5 micro-axial flow pump
• Only patients with available high-resolution pump data (downloaded from the clinical console) and ICU digital health record datasets

Exclusion Criteria:

• Patients supported with an Impella 5.5 for indications other than cardiogenic shock (e.g., protected PCI or CABG)
• Patients younger than 18 years
• Patients with incomplete data, procedural records, or demographic information

The study population comprises adult patients admitted to the intensive care unit (ICU) presenting with severe cardiogenic shock who required temporary mechanical circulatory support between January 2019 and March 2026, treated at the Medical University of Vienna, Austria. Eligible individuals are those who undergo clinical management utilizing the Impella 5.5 micro-axial flow pump as part of their standard of care. Only patients with available high-resolution pump data (downloaded from the clinical console) and ICU digital health record datasets extracted from ICCA (IntelliSpace Critical Care and Anesthesia, Philips Medical Systems Development and Manufacturing Centre, Hamburg, Germany) will be included for analysis.

IPD will not be shared because the dataset contains highly sensitive, high-frequency physiological data linked with intensive care health records. Further, IPD is currently not planned for public sharing as the dataset is being actively used to develop a proprietary clinical decision support tool and model architecture optimization.