Triage Tool for Accurate Disposition of Patients in Disaster Response
Scott Levin’s project seeks to develop a decision support tool to improve surge capacity in mass casualty incidents.
To improve hospital surge capacity by developing a rapid
evidence-based tool to predict appropriate level of in-hospital
care using emergency department information.
Completed Project Overview
Research activities for the PACER funded project "Triage Tool
for Accurate Disposition of Emergency Patients in Disaster
Response" has beendirected toward developing 'HopScore' a novel,
electronic, outcomes-based triage tool. The tool, 'HopScore'
(developed at Johns Hopkins) aims to support objective triage
decisions and improve patient differentiation based on
time-sensitive and critical patient outcomes.
HopScore uses easily obtained patient demographic and clinical
information commonly collected at ED triage to predict patients
risk for time-sensitive outcomes(i.e., in-hospital mortality,
intensive care unit admission, emergent surgery or cardiac
catheterization) and general inpatient admission. Inputinformation
includes basic demographics, chief complaint, mode of arrival, and
vital signs (temperature, heart rate, systolic blood
pressure,respiratory rate, and oxygen saturation). HopScore was
originally derived on a sample of 25,198 (97 million weighted) ED
visits for adult (>18years) patients from the National Hospital
Ambulatory Medical Care Survey (NHAMCS), a nationally
representative probability sample collectedby the Center for
Disease Control and Prevention (CDC).
The tool was later refined to include over a 100,000 patients
that were seen at the JohnsHopkins emergency department (ED).
HopScore addresses deficiencies in the current standard, Emergency
Severity Index (ESI) triage tool used by 72% of EDs across the
United States (US). Deficiencies of ESI include: (1) no link to
critical patient outcomes, (2) a strong reliance on
triageevaluator's subjective judgment which results in variable
application, (3) poor distribution (i.e., discrimination) of
patients across the 5-levelindex; almost half of all patients
nationally are undifferentiated, classified as acuity level 3.
HopScore uses machine learning (i.e., decision tree)algorithm to
render predictions (i.e., probability estimates) and then
stratifies patients to a 5-level index based on risk of critical
outcomes. Much of our analyses compares HopScore to the current