Available: 02/05/24, Expires: 02/01/25
Pulmonary embolism (PE) remains a significant cause of morbidity and mortality and is a critical “cannot miss” diagnosis in the Emergency Department (ED). Despite this, it is difficult to diagnose as the spectrum of disease ranges from patients being asymptomatic to having hemodynamic instability and cardiopulmonary collapse. Due to this challenge, there has been a lot of work to create clinical decision rules, such as the Pulmonary Embolism Rule Out Criteria (PERC) Rule. The PERC Rule can be quickly applied by clinicians to determine which patients for which PE is part of the differential diagnosis are at extremely low risk for PE and do not require further diagnostic testing or for whom further diagnostic testing should be pursued. Unfortunately, due to challenges related to ED boarding and crowding, patients may wait to be seen by a clinician for many hours. This creates an environment where it is possible that a patient with a “cannot miss” diagnosis waits for hours before the appropriate work-up can be started. Fortunately, all patients that arrive in the ED are seen by skilled nursing staff who are trained in ED triage. Their role is to identify patients for whom extended waiting periods in the ED would be unsafe. They do this by collecting initial vital signs, demographic data, basic medical history, and a one to two sentence “triage note.” As one might imagine, ED triage is a high stakes, stress, and volume role.
Within the Massachusetts General Hospital, there have been several quality improvement (QI) initiatives to improve ED triage including EPIC incorporated “sepsis flags” which flag a patient as “at risk” for sepsis based on their triage information. The goal of this project is to use machine learning (ML), artificial intelligence, and ChatGPT to help design an MGH ED PE flag. The goal is to recognize patients who present to the ED and are at high-risk for PE such that they can be flagged and considered for an expedited work-up for PE. The primary aim of this project will be to design, train, test, and validate ML models compared to a basic ChatGPT model, as well as nurse triage alone to determine if a high-quality model for estimating an ED patient’s probability for PE can be created from triage data alone. For this project, you will work with both Chris Kabrhel, MD, MPH, an MGH ED attending physician, director of the Center for Vascular Emergencies (CVE), and a world-renowned PE expert and Drew Birrenkott, MD, DPhil, a third-year emergency medicine resident who completed his doctorate in biomedical engineering with a focus on ML. The data for this project is derived from a data set containing all of the triage information of all patients who were deemed to be at high-enough risk for a PE that they were evaluated in the ED with a computed tomography pulmonary angiogram (CTPA), the gold standard diagnostic study for PE, whether or not they were ultimately diagnosed with a PE.