A Rapid Diagnostic of Risk in Hospitalized Patients Using Machine Learning

A Rapid Diagnostic of Risk in Hospitalized Patients With COVID-19, Sepsis, and Other High-Risk Conditions to Improve Outcomes and Critical Resource Allocation Using Machine Learning

In this study, the investigators will deploy a software-based clinical decision support tool (eCARTv5) into the electronic health record (EHR) workflow of multiple hospital wards. eCART's algorithm is designed to analyze real-time EHR data, such as vitals and laboratory results, to identify which patients are at increased risk for clinical deterioration. The algorithm specifically predicts imminent death or the need for intensive care unit (ICU) transfer. Within the eCART interface, clinical teams are then directed toward standardized guidance to determine next steps in care for elevated-risk patients.

The investigators hypothesize that implementing such a tool will be associated with a decrease in ventilator utilization, length of stay, and mortality for high-risk hospitalized adults.

Study Overview

• Status: Active, not recruiting
• Conditions: COVID-19; Sepsis; Respiratory Failure; Cardiac Arrest; Hemodynamic Instability; Septicemia; Clinical Deterioration
• Intervention / Treatment: Device: eCARTv5 clinical deterioration monitoring; Other: Standard of care control

Detailed Description

The objective of this proposal is to rapidly deploy a clinical decision support tool (eCARTv5) within the electronic health record of multiple medical-surgical units. eCART combines a real-time machine learning algorithm for identifying patients at increased risk for intensive care (ICU) transfer and death with clinical pathways to standardize the care of these patients based on a real-time, quantitative assessment of patient risk.

The investigators hypothesize that implementing such a tool will be associated with a decrease in ventilator utilization, length of stay, and mortality for high-risk hospitalized adults.

Background:

Clinical deterioration occurs in approximately 5% of hospitalized adults. Delays in recognition of deterioration heighten the risk of adverse outcomes. Machine learning algorithms enhance clinical decision-making and can improve the quality of patient care. However, their impact on clinical outcomes depends not only on the sensitivity and specificity of the algorithm but also on how well that algorithm is integrated into provider workflows and facilitates timely and appropriate intervention.

Preliminary Data:

eCART has been built upon more than a decade of ongoing scientific research and chronicled in numerous peer-reviewed publications. eCART was developed at the University of Chicago by Drs. Dana Edelson and Matthew Churpek. The first version (eCARTv1) was derived and validated using linear logistic regression in a dataset of nearly 60,000 adult ward patients from a single medical center. That model had 16 variables in it and was subsequently validated in silent mode, demonstrating that eCART could alert clinicians more than 24 hours in advance of ICU transfer or cardiac arrest. eCARTv2, derived and validated in a dataset of nearly 270,000 patients from 5 hospitals, improved upon the earlier version by utilizing a cubic spline logistic regression model with 27 variables and demonstrated improved accuracy over the Modified Early Warning Score (MEWS), a commonly used score that can be hand- calculated by nurses at the bedside (AUC 0.77 vs. 0.70 for cardiac arrest, ICU transfer or death). In a multicenter clinical implementation study, eCARTv2 was associated with a 29% relative risk reduction for mortality. In further development of eCART, the University of Chicago research team demonstrated that upgrading from a cubic spline model to a machine learning model, such as a random forest or gradient boosted machine (GBM), could increase the AUC. In the most recent development - eCART v5 - the research team has advanced the analytic using a gradient boosted machine learning model trained on a multi-center dataset of more than 800,000 patient records. Now with 97 variables, this more sophisticated model increases the accuracy by which clinicians can predict clinical deterioration.

• Study Type: Interventional
• Enrollment (Estimated): 30000
• Phase: Not Applicable

Contacts and Locations

Study Locations

• United States
  • Connecticut
    • New Haven, Connecticut, United States, 06510
      • Yale New Haven Health System
  • Florida
    • Clearwater, Florida, United States, 33759
      • BayCare Health System
  • Wisconsin
    • Madison, Wisconsin, United States, 53792
      • University of Wisconsin Health

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

• 18 years old
• Admitted to an eCART-monitored medical-surgical unit (scoring location)

Exclusion Criteria:

• Younger than 18 years old
• Not admitted to an eCART-monitored medical surgical unit (scoring location)

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Prevention
• Allocation: Non-Randomized
• Interventional Model: Parallel Assignment
• Masking: Triple

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: Intervention Arm; Intervention Arm (experimental): eCARTv5 will monitor all adult medical-surgical (ward) patients at hospitals that implement the tool in their EHR. A pre vs. post analysis will be done to compare the impact of the tool at the intervention hospitals.	Device: eCARTv5 clinical deterioration monitoring; eCART is a predictive analytic used for the identification of acute clinical deterioration built upon more than a decade of ongoing scientific research and chronicled in numerous peer-reviewed publications. eCART draws upon readily available patient data from the EHR, rapidly quantifies disease severity, and predicts the likelihood of critical illness onset.
Active Comparator: Control Arm; Control Arm (active comparator): hospital sites that do not implement eCARTv5 will be active comparator.	Other: Standard of care control; Standard of care is the health system's clinical best practices and workflows for identifying high-risk patients for clinical deterioration, including other tools already built into the electronic health record (EHR). Hospitals that do not implement eCARTv5 will be compared as a control against hospitals that do implement eCARTv5.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Hospital mortality for elevated risk patients	Hospital mortality, a measure of how many patients died in the hospital, will come from administrative data, specifically from the discharge disposition of each eCART elevated risk patient. This data will be taken from the complete hospitalization, from admission to discharge.	The outcome of hospital mortality for elevated risk patients will be tracked across 12 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Total hospital length of stay (LOS) for elevated risk patients	Total hospital length of stay (LOS) for patients with any elevated eCART score during hospitalization, defined as the time period between hospital admission and discharge. LOS is defined as the time (hours or fraction of a day) from first vital sign to last vital sign within a patient encounter.	Total hospital length of stay (LOS) for elevated risk patients will be tracked across 12 months
ICU-free days following an eCART elevation	30-day ICU-free days, defined as the number of days patients were both alive and not being cared for in an ICU in the first 30 days following hospital admission with any elevated eCART score. Because death is biased toward fewer ICU days and is a competing outcome, patients who die prior to day 30 are assigned with 0 ICU-free days.	The outcome of 30-day ICU-free days will be tracked across 12 months
Ventilator-free days following an eCART elevation	30-day ventilator-free days, defined as the number of days patients were both alive and not mechanically ventilated in the first 30 days following hospital admission with any elevated eCART score. Because death is biased toward fewer ventilator days and is a competing outcome, patients who die prior to day 30 are assigned with 0 ventilator-free days.	The outcome of 30-day ventilator-free days will be tracked across 12 months

Other Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sepsis Mortality	Hospital mortality, a measure of how many patients died in the hospital, will come from administrative data, specifically from the discharge disposition of each eCART elevated risk patient meeting Sep-1 criteria for sepsis.	The outcome of sepsis mortality will be tracked across 12 months
Sepsis Length of Stay (LOS)	Total hospital length of stay (LOS) for patients with any elevated eCART score during hospitalization that met Sep-1 criteria for sepsis.	The outcome of sepsis length of stay (LOS) will be tracked across 12 months
COVID-19 Mortality	Hospital mortality, a measure of how many patients died in the hospital, will come from administrative data, specifically from the discharge disposition of each eCART elevated risk patient with a COVID-19 diagnosis or positive COVID-19 test result.	The outcome of COVID-19 mortality will be tracked across 12 months
COVID-19 Length of Stay (LOS)	Total hospital length of stay (LOS) for patients with any elevated eCART score during hospitalization with a COVID-19 diagnosis or positive COVID-19 test result.	The outcomes of COVID-19 length of stay (LOS) will be tracked across 12 months

Collaborators and Investigators

• Sponsor: AgileMD, Inc.
• Collaborators: Yale University; University of Chicago; University of Wisconsin, Madison; Biomedical Advanced Research and Development Authority; BayCare Health System
• Investigators: Study Chair: Dana P Edelson, MD, MS, AgileMD, Inc.

Publications and helpful links

General Publications

• Churpek MM, Yuen TC, Park SY, Meltzer DO, Hall JB, Edelson DP. Derivation of a cardiac arrest prediction model using ward vital signs*. Crit Care Med. 2012 Jul;40(7):2102-8. doi: 10.1097/CCM.0b013e318250aa5a.
• Churpek MM, Yuen TC, Winslow C, Robicsek AA, Meltzer DO, Gibbons RD, Edelson DP. Multicenter development and validation of a risk stratification tool for ward patients. Am J Respir Crit Care Med. 2014 Sep 15;190(6):649-55. doi: 10.1164/rccm.201406-1022OC.
• Kang MA, Churpek MM, Zadravecz FJ, Adhikari R, Twu NM, Edelson DP. Real-Time Risk Prediction on the Wards: A Feasibility Study. Crit Care Med. 2016 Aug;44(8):1468-73. doi: 10.1097/CCM.0000000000001716.
• Winslow CJ, Edelson DP, Churpek MM, Taneja M, Shah NS, Datta A, Wang CH, Ravichandran U, McNulty P, Kharasch M, Halasyamani LK. The Impact of a Machine Learning Early Warning Score on Hospital Mortality: A Multicenter Clinical Intervention Trial. Crit Care Med. 2022 Sep 1;50(9):1339-1347. doi: 10.1097/CCM.0000000000005492. Epub 2022 Aug 15.

Study record dates

Study Major Dates

• Study Start (Actual): December 31, 2024
• Primary Completion (Estimated): December 31, 2026
• Study Completion (Estimated): December 31, 2026

Study Registration Dates

• First Submitted: April 24, 2023
• First Submitted That Met QC Criteria: June 6, 2023
• First Posted (Actual): June 7, 2023

Study Record Updates

• Last Update Posted (Actual): July 29, 2025
• Last Update Submitted That Met QC Criteria: July 25, 2025
• Last Verified: July 1, 2025

More Information

Terms related to this study

• Keywords: machine learning; artificial intelligence; clinical decision support; early warning scores
• Additional Relevant MeSH Terms: Pathologic Processes; Disease Attributes; Respiratory Tract Infections; Infections; RNA Virus Infections; Virus Diseases; Respiratory Tract Diseases; Systemic Inflammatory Response Syndrome; Inflammation; Lung Diseases; Respiration Disorders; Pneumonia, Viral; Pneumonia; Coronavirus Infections; Coronaviridae Infections; Nidovirales Infections; Disease Progression; COVID-19; Clinical Deterioration; Respiratory Insufficiency; Sepsis
• Other Study ID Numbers: 1.0

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: Yes
• product manufactured in and exported from the U.S.: No