Emergency department (ED) nurses triage over 136 million patients each year in the United States. The goal of triage is to assess and identify clinical conditions in order to prioritize those with the most significant risk of morbidity and mortality. Current practice uses the Emergency Severity Index (ESI) score to group patients by resource utilization. ESI has significant limitations including: racial bias, poor relation to patient-centered outcomes, subjectivity, and failure to differentiate acute patients (poor specificity). As such, the ESI tool fails to identify patient-specific factors, that are present at the time of triage, to accurately predict critical conditions requiring life-saving treatments. Due to its time sensitivity, complex symptomology, variable outcomes, and a national cost burden of $21 billion, acute coronary syndrome (ACS) will be used as an exemplar time-sensitive condition to develop a new predictive machine learning algorithm to be used for ED triage. Of 800,000 new annual ACS cases in the United States, nurses fail to identify approximately 50% during triage. This suggests an urgent need to develop triage tools, specifically ones that correctly identify ACS early, which could potentially reduce mortality by 10%-20%. This project proposes to use big data analytics to address the critical gaps of the ESI tool and nurse failure to identify ACS at triage. A large cohort of patients presenting to 17 different EDs with symptoms suspicious of ACS will be used to create a multidimensional database, extracting routinely collected patient factors from the electronic health record data acquired at initial nurse triage. This project will use state-of-the-art machine learning approaches that incorporate the complex interactions between patient factors to identify patients with true critical coronary occlusion that require time dependent treatment. The innovation of this study stems from having access to a world renowned academic medical center that is able to conduct full-scale studies using electronic health records of over 4.2 million patient encounters. This project aligns with the NINR?s strategic vision for nurses to use emerging technologies (big data) to predict patient trajectories, inform interventions and support real time clinical decision making. By using advanced machine learning concepts, we will translate our final machine model into a robust clinical tool to assist nurses in making real time clinical decisions to accurately identity ACS events and initiate timely treatment, thereby improving patient outcomes. Study findings have potential to change the paradigm of ED nurse triage to be more objective and data-driven, thereby recognizing critical conditions at initial triage and eliminating unnecessary morbidity and mortality.
Emergency department nurses triage over 136 million patients a year, using the Emergency Severity Index score that has significant limitations and often fails to identify patients with critical time-sensitive conditions such as acute coronary syndrome (ACS) approximately 50% of the time. To address these gaps, this project aims to develop a predictive algorithm using state-of-the-art machine learning concepts that incorporates the complex interactions between medical data available at nurse triage in order to identify patients with true critical coronary occlusion that requires time dependent treatment. Such a model can be translated into a robust clinical decision support tool to assist nurses in making real time clinical decisions to accurately identify ACS events and initiate timely treatment, thereby improving patient outcomes.
