Clinical evaluation of a machine learning-based early warning system for patient deterioration.

Area of Science:

• Healthcare Technology
• Clinical Informatics
• Predictive Analytics

Background:

• Implementation and clinical impact of ML-based EWS for patient deterioration are not well-described.
• Study focused on a multifaceted, real-time ML-based EWS in an academic medical center's general internal medicine (GIM) unit.

Purpose of the Study:

• To describe the implementation and evaluation of a novel ML-based EWS.
• To assess the association between the ML-EWS and clinical outcomes, specifically non-palliative in-hospital death.

Main Methods:

• Nonrandomized, controlled study using propensity score-based overlap weighting.
• Compared GIM unit patients during intervention (Nov 2020-June 2022) with pre-intervention (Nov 2016-June 2020).
• Difference-in-differences analysis compared GIM with subspecialty units (cardiology, respirology, nephrology) not receiving the intervention.

Main Results:

• 13,649 GIM admissions and 8,470 subspecialty admissions included.
• Non-palliative deaths decreased in GIM during intervention (1.6% vs 2.1%, aRR 0.74).
• High-risk GIM patients with alerts had lower deaths (7.1% vs 10.3%, aRR 0.69).

Conclusions:

• Implementing an ML-based EWS in the GIM unit was linked to a reduced risk of non-palliative death.
• ML-based EWS represent promising technology for enhancing clinical outcomes in hospital settings.