Effectiveness of an Artificial Intelligence-Enabled Intervention for Detecting Clinical Deterioration

Area of Science:

• Healthcare Technology
• Clinical Informatics
• Patient Safety

Background:

• Inpatient clinical deterioration poses significant risks, often missed by clinicians.
• Early warning scores aim to identify high-risk patients, but evidence of effectiveness is limited.

Purpose of the Study:

• To evaluate an artificial intelligence (AI) deterioration model-enabled intervention.
• To reduce escalations in care for hospitalized patients using a robust causal inference study design.

Main Methods:

• A cohort study employed a regression discontinuity design, utilizing the Epic Deterioration Index (EDI) prediction model.
• The intervention involved AI-driven alerts based on EDI score thresholds and a collaborative nurse-physician workflow.
• Included hospitalized adults from four general internal medicine units between January 2021 and November 2022.

Main Results:

• The AI intervention was associated with a -10.4-percentage point absolute risk reduction in escalations of care (e.g., rapid response team activation, ICU transfer, cardiopulmonary arrest).
• This reduction was statistically significant (P=.03) for patients at the EDI score threshold.
• No evidence of confounding bias was found at the EDI score threshold.

Conclusions:

• An AI deterioration model-enabled intervention significantly decreased the risk of care escalations in hospitalized patients.
• The findings support the effectiveness of AI in improving inpatient safety.
• Further expansion and testing of this AI intervention in diverse settings are warranted.