RMS: a ML-based system for ICU respiratory monitoring and resource planning

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Respiratory Physiology

Background:

• Acute hypoxemic respiratory failure (RF) is a common, serious complication in critically ill patients.
• Effective management of RF is crucial for reducing morbidity, mortality, and healthcare resource utilization.

Purpose of the Study:

• To develop and validate a machine-learning (ML) based monitoring system for comprehensive ICU patient management.
• To enable early detection, continuous monitoring, and prediction of extubation readiness and failure (EF).

Main Methods:

• Development of a comprehensive ML monitoring system for ICU physicians.
• The system focuses on early RF detection, continuous monitoring, extubation readiness assessment, and EF prediction.
• Model performance was evaluated against standard clinical monitoring and validated in an external ICU cohort.

Main Results:

• The ML model predicted 80% of RF events with 45% precision, detecting 65% of events over 10 hours earlier than standard methods.
• The system successfully predicted extubation failure risk, aiding in prevention and optimizing ventilation duration.
• The model accurately predicted ICU ventilator demand 8-16 hours in advance with a mean absolute error of 0.4 ventilators per 10 patients.

Conclusions:

• The ML-based monitoring system significantly enhances early detection and management of respiratory failure in ICUs.
• This AI tool improves patient outcomes by predicting extubation failure and optimizing mechanical ventilation resource allocation.
• The validated system offers a valuable tool for ICU physicians, improving patient care and operational efficiency.