Cardiac Output Estimation in the Intensive Care Unit.

Area of Science:

• Cardiovascular physiology
• Critical care medicine
• Machine learning applications

Background:

• Cardiac output (CO) estimation is crucial for managing critically ill patients.
• Existing methods like thermodilution (TD) and estimated Fick (eFick) have limitations, necessitating novel approaches.

Purpose of the Study:

• To develop and validate novel machine learning-based cardiac output estimators for critical care settings.
• To address the limitations of current CO estimation techniques in intensive care units (ICUs).

Main Methods:

• Machine learning models were trained and validated using a large dataset (13,172 measurements from 4,825 patients) of TD-CO.
• Performance was assessed using regression metrics, trajectory analysis, and CO tracking accuracy.
• Models utilized routine physiological measurements from ICU or cardiac catheterization lab settings.

Main Results:

• Established eFick models performed poorly in the ICU due to static oxygen consumption estimates (e.g., R² of -1.5).
• The novel CORE (Catheter Optimized caRdiac output Estimation) model achieved 14% MAPE and R² of 0.58, significantly outperforming eFick (P < 0.001).
• CORE demonstrated robustness in the presence of tricuspid regurgitation (16% MAPE, R² of 0.65) and adaptability to different catheter types.

Conclusions:

• Machine learning models incorporating dynamic physiology improve CO estimation accuracy in ICU patients compared to eFick and TD.
• The CORE model offers versatility, ease of use, and broad applicability across diverse ICU environments.
• These advanced CO estimators have the potential to enhance patient care in critical illness settings.