The Problem
Acute pulmonary emboli (PE) originate from thrombi in the deep veins of the legs that embolize and become lodged in the pulmonary arteries, obstructing blood flow to the lungs.(1) Several studies estimate that nearly 300,000 Americans are diagnosed with PE annually and experience a 10% 3-month and 25% 12-month mortality.(2-6) Of those that die, nearly 34% died before therapy could take effect.(7) At Duke, 1-2 PEs are diagnosed daily and are associated with a 9% mortality rate at hospital discharge.(8) Of all patients diagnosed with PE at Duke, 26% are admitted to an ICU and 9% require mechanical ventilation. Importantly, ~19% of all PEs complicate an inpatient admission for a different indication(termed in-hospital PE). Half of these in-hospital PEs will lead to ICU admission and, among these ICU transfers, 60% will require mechanical ventilation and 20% will experienc e mortality. In terms of morbidity, mean and median hospital length of stay for all PE patients are 7.5 and 5 days respectively, but reach as high as 13 days in high risk patients, and 14% of patients are eventually discharged to acute rehabilitation. As the complexity and morbidity of these patients have increased over time, annualized healthcare-associated costs have increased 2.5-fold.(9)
Depending on severity,some patients diagnosed with acute PE may be managed in the outpatient setting, some may require close monitoring for hemodynamic collapse, or some may require mechanical circulatory support for obstructive shock. Risk stratification represents a unique challenge, but several scores and guidelines are available to improve triage and aide in management decisions. In low risk patients, the Simplified Pulmonary Embolism Severity Index (sPESI) score has been shown to reliably identify patients who do not require hospital admission utilizing age, comorbidities, and initial vital signs.(10) The remaining patients are at risk for right ventricular (RV) failure and guidelines reflect numerous studies that recommend RV assess menton computed tomography and echocardiography as well as with measurement of cardiac biomarkers.(6, 11-13) Unfortunately, fewer than 50% of patients admitted to Duke are fully risk stratified,and RV dysfunction on CTis reported in only 18% of radiology reports, despite being associated with a 5-fold increased risk of mortality.
More consistent risk stratification up-front is needed for two reasons. First,it guides clinical decision-making by front-line providers(e.g. Emergency Department and Urgent Care clinicians)who must decide if and where to admit a patient. Data by Sullivan et al. shows that intermediate-high risk patients (by European Society of Cardiology criteria(6, 14)) are 50% more likely to be admitted to Duke ICUs relative to other intermediate risk patients. Second, patients must be risk stratified to identify higher risk individuals who should be evaluated by specialists for prompt interventions to reduce healthcare utilization and improve end-organ function.Emerging technologies have shown promise(15-17) and our data from Duke demonstrates a 4-day shorter hospital length of stay when these interventions are used, but application is limited by suboptimal stratification rates and specialist consultation.
Given the complex decision-making around high risk acute PE, utilization of a multidisciplinary team was recently added as a class IIA indication in the most recent PE society guidelines.(13)Duke is uniquely positioned to be a leader in the field of PE and invasive therapies, specifically catheter-directed therapies. As one of the largest health systems in the country for acute PE with a wide referral network, our active pulmonary vascular consult service can facilitate state-of-the-art PE therapies including extracorpuscular membrane oxygenation (ECMO), catheter-directed therapies, and surgical embolectomies. However, from our clinical experience and published data from similar academic institutions, we believe that the pulmonary vascular consult is woefully underutilized and even less frequently results in multidisciplinary conversation.We have already started to understand the landscape here at Duke and currently have the only 100% physician-reviewed and validated acute PE database (created as part of IRB PRO00090028),which could serve as the ground truth for machine learning (ML) technology.
Our Solution
We propose developing a PE machine learning model to facilitate early identification of patients with high risk PE who may experience clinical deterioration without prompt intervention. We will pilot this model and implement an EMR-based decision support tool which will prompt providers to order appropriate stratification tests, consults, and therapeutics after detection of acute PE via ML technology, similarly to the already established Sepsis Watch™ decision support tool.
