Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Improving Emergency Department Nurse Triage via Big Data Analytics

Frisch, Stephanie Outterson (2020) Improving Emergency Department Nurse Triage via Big Data Analytics. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Download (4MB) | Preview


Background: In the United States, emergency department nurses triage approximately 145 million patients a year. Triage is the brief period of time where nurses assess and prioritize patients who have the most significant risk for morbidity and mortality. Typical or atypical patient presentation for acute coronary syndrome (ACS), or heart attack, is challenging for nurses to distinguish given the over 30 potential symptoms at triage. Machine learning algorithms using routinely collected objective data have potential to improve identification of a coronary event and to potentially eliminate known biases in the current triage system, thus improving patient outcomes.
Purpose: The purpose of this study was to develop and validate a predictive triage algorithm that can identifiy ACS that requires immediate treatment in patients presenting with suspected ACS. We aimed to: 1) assemble and annotate a large cohort of patients presenting to the emergency department for suspicion of ACS; 2) develop, validate and compare five machine learning algorithms to develop a sensitive and specific model to predict the outcome of ACS; and 3) compare the performance of our best two machine learning algortihms against routine emergency department triage practice (i.e., the Emergency Severity Index).
Methods: We conducted a retrospective observational cohort study of adult patients who were triaged at the emergency department for a suspected coronary event. We developed, validated and compared five machine learning algorithms (binary logistic regression, naïve Bayes, random forest, gradient boosting machine, and artificial neural network) using routinely collected data that could be available at triage. We used 10-fold cross validation to predict the outcome of ACS and to identify the best two performing algorithms using the area under the receiver operating characteristic curve (AUC). We used lasso regularization to select a subset of input variables for the outcome of ACS. We then compared performance of our machine learning classifiers to the dichotomized assigned scores from the Emergency Severity Index to correctly classify the diagnosis of ACS as high acuity using the AUC. We used the Delong test to compare the AUC of our best performing machine learning algorithms to correctly assigned high acuity triage scores.
Results: Our sample included 1201 patients (mean age 65±14 years, 46% female, 89% white, 1% Hispanic) with 522 (43%) patients having a diagnosis of ACS. We identified a total of 243 input variables with a subset of 43 variables chosen using lasso regularization. Artifical neural network and binary logistic regression were the best performing algorithms using the subset of 43 input variables with the AUC of 0.78 [95% confidence interval (CI), 0.76–0.80] and 0.77 (95% CI, 0.75–0.79), respectively. Both algorithms outperformed the diachotomized ESI triage scores for placing ACS as high acuity with an AUC of 0.61 (95% CI, 0.60–0.63). There was a statistically significant difference in AUC between the best performing algorithm (artificial neural network) and the correctly assigned ESI triage scores (p < 0.001).
Conclusion: Our machine learning algorithms outperformed routine triage scores in identifying highest risk patients among those with suspected ACS using baseline triage data collected in the brief period of time to assess a patient at the emergency department to identify ACS. There was a 17% accuracy rate improvement when comparing the artifical neural network accuracy rate to the correctly assigned high acuity triage scores for the outcome of ACS in our hetereogenious patient population. The application of predictive algorithms could be translated into a clinical decision support tool to enhance identification of patients with potential ACS, improving timely treatments, which could improve patient outcomes.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Frisch, Stephanie Outtersonsof9@pitt.edusof90000-0003-3692-3125
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairAl-Zaiti, Salahssa33@pitt.edussa33
Committee MemberCallaway, Cliftoncallawaycw@upmc.educallaway
Committee MemberSejdic, Ervinesejdic@pitt.eduesejdic
Committee MemberSereika, Susanssereika@pitt.edussereika
Committee MemberHravnak, Marilynmhra@pitt.edumhra
Date: 29 July 2020
Date Type: Publication
Defense Date: 18 June 2020
Approval Date: 29 July 2020
Submission Date: 28 July 2020
Access Restriction: 2 year -- Restrict access to University of Pittsburgh for a period of 2 years.
Number of Pages: 182
Institution: University of Pittsburgh
Schools and Programs: School of Nursing > Nursing
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: emergency department, triage, nursing, machine learning, acute coronary syndrome
Date Deposited: 29 Jul 2020 15:07
Last Modified: 29 Jul 2022 05:15


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item