Application of Machine Learning Models to Biomedical and Information System Signals From Critically Ill Adults.

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• Impending hemodynamic instability and respiratory failure require timely physician intervention.
• Machine learning (ML) notifications offer potential for early alerts to prevent critical events.

Purpose of the Study:

• To compare the predictive performance of ML alerts, telemedicine system (TS) alerts, and biomedical monitors (BMs).
• To evaluate the superiority of these systems in predicting intubation or vasopressor administration in critically ill adults.

Main Methods:

• An ML algorithm was developed to predict intubation and vasopressor initiation events.
• The ML algorithm's performance was benchmarked against BM alarms and TS alerts.

Main Results:

• ML notifications demonstrated superior accuracy (0.87-0.94) and precision, with a 50-fold lower alarm burden compared to TS alerts for predicting vasopressor initiation and intubation.
• TS alerts showed a 10-fold lower alarm burden than BM alarms.
• ML alert performance was consistent across internal, external validation, and COVID-19 cohorts, outperforming TS and BM alerts significantly in F-score metrics.

Conclusions:

• ML-derived notifications represent a significant advancement in managing hemodynamic instability and respiratory failure.
• The substantial improvements in accuracy, precision, and reduced alarm burden enable more proactive patient care and less disruption to clinical workflows.