USF-LVHN SELECT

Leveraging Artificial Intelligence to Reduce Neuroscience ICU Length of Stay.

Publication/Presentation Date

3-1-2025

Abstract

GOAL: Efficient patient flow is critical at Tampa General Hospital (TGH), a large academic tertiary care center and safety net hospital with more than 50,000 discharges and 30,000 surgical procedures per year. TGH collaborated with GE HealthCare Command Center to build a command center (called CareComm) with real-time artificial intelligence (AI) applications, known as tiles, to dynamically streamline patient care operations and throughput. To facilitate patient flow for our neuroscience service line, we partnered with the GE HealthCare Command Center team to configure a Downgrade Readiness Tile (DRT) to expedite patient transfers out of the neuroscience intensive care unit (NSICU) and reduce their length of stay (LOS).

METHODS: As part of an integrated NSICU performance improvement project, our LOS reduction workgroup identified the admission/discharge and transfer process as key metrics. Based on a 90%-plus average capacity, early identification of patients eligible for a downgrade to lower acuity units is critical to maintain flow from the operating rooms and emergency department. Our group identified clinical factors consistent with downgrade readiness as well as barriers preventing transition to the next phase of care. Configuration of an AI-powered model was identified as a mechanism to drive earlier downgrade and reduce LOS in the NSICU. A multidisciplinary ICU LOS reduction steering committee met to determine the criteria, design, and implementation of the AI-powered DRT. As opposed to identifying traditional clinical factors associated with stability for transfer, our working group asked, "What are clinical barriers preventing downgrade?" We identified more than 76 clinical elements from the electronic medical records that are programmed and displayed in real-time with a desired accuracy of over 95%. If no criteria are present, and no bed is requested or assigned, the DRT will report potential readiness for transfer. If three or more criteria are present, the DRT will suggest that the patient is not eligible for transfer.

PRINCIPAL FINDINGS: The DRT was implemented in January 2022 and is used during multidisciplinary rounds (MDRs) and displayed on monitors positioned throughout the NSICU. During MDRs, the bedside nurses present each patient's key information in a standardized manner, after which the DRT is used to recommend or oppose patient transfer. Six months postimplementation period of the DRT and MDRs, the NSICU has seen a 7% or roughly eight-hour reduction in the ICU length of stay (4.15-3.88 days) with a more than three-hour earlier placement of a transfer order. Unplanned returns to the ICU (or bouncebacks) have remained low with no change in the preimplementation rate of 3% within 24 hours. As a result of this success, DRTs are being implemented in the medical ICUs.

PRACTICAL APPLICATIONS: This work is uniquely innovative as it shows AI can be integrated into traditional interdisciplinary rounds and enable accelerated decision-making, continuous monitoring, and real-time alerts. ICU throughput has traditionally relied on direct review of a patient's clinical course executed during clinical rounds. Our methodology adds a dynamic and technologically augmented touchpoint that is available in real time and can prompt a transfer request at any time throughout the day.

Volume

70

Issue

2

First Page

126

Last Page

136

ISSN

1096-9012

Disciplines

Medical Education | Medicine and Health Sciences

PubMedID

40059204

Department(s)

USF-LVHN SELECT Program, USF-LVHN SELECT Program Students

Document Type

Article

Link to Full Text

Share

COinS