A Deep Learning Model to Guide Personalized Mechanical Circulatory Support Use in Cardiogenic Shock Patients...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Cardiogenic shock (CS) following percutaneous coronary intervention (PCI) presents complex, dynamic clinical, hemodynamic, and metabolic challenges.
• Current predictive models struggle to integrate these multifaceted factors effectively for optimal patient management.

Purpose of the Study:

• To develop and validate a deep learning (DL) model, named OPtiMCS.
• To guide the personalized selection and utilization of mechanical circulatory support (MCS) devices in CS patients undergoing PCI.

Main Methods:

• Analysis of data from 1,408 CS patients treated with intra-aortic balloon pump (IABP) or microaxial flow-pump (mAFP) between 2004-2019.
• Development of the OPtiMCS DL model using longitudinal clinical, hemodynamic, and metabolic data, implemented in Python 3.7 with TabNet in PyTorch.
• Validation of the model for predicting 30-day mortality, bleeding, acute kidney injury (AKI), 1-year mortality, and 1-year stroke, including simulation of alternate MCS device outcomes.

Main Results:

• The OPtiMCS model demonstrated high predictive performance with AUCs ranging from 83% (AKI) to 98% (mortality).
• Key predictive features were identified, such as cardiac arrest predicting mortality.
• The model successfully predicted the comparative benefits of switching between IABP and mAFP, estimating potential reductions in mortality and AKI.

Conclusions:

• A validated DL model (OPtiMCS) can predict outcomes and personalize MCS device selection for CS patients undergoing PCI.
• External validation and clinical implementation could enable better risk-benefit assessment of MCS devices.
• The model holds potential to reduce adverse outcomes, improve survival rates, and advance the care of CS patients.