The Impact of a Machine Learning Early Warning Score on Hospital Mortality: A Multicenter Clinical Intervention Trial

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Elevated-risk adult inpatients face significant mortality risks.
• Early detection and intervention are crucial for improving outcomes in hospitalized patients.
• Existing early warning systems may lack the precision to effectively stratify risk.

Purpose of the Study:

• To evaluate the impact of the electronic Cardiac Arrest Risk Triage (eCART) machine learning early warning risk score on mortality.
• To assess if implementing eCART improves outcomes for elevated-risk adult inpatients.

Main Methods:

• A pragmatic pre- and post-intervention study was conducted across four hospitals.
• Adult patients on medical-surgical wards were included, with 6,681 meeting criteria.
• The intervention involved presenting eCART scores to clinicians, triggering specific assessment workflows for high- and intermediate-risk patients.

Main Results:

• Hospital mortality significantly decreased from 13.9% to 8.8% during the intervention period (adjusted OR, 0.60).
• Significant mortality reductions were observed in high- and intermediate-risk subgroups.
• The intervention also led to increased ICU transfers, reduced time to transfer, and more frequent vital sign reassessments.

Conclusions:

• Implementing a machine learning-driven early warning score protocol reduced inhospital mortality.
• Earlier and more frequent intensive care unit (ICU) transfers appear to be the primary driver of this mortality reduction.