Abstract 327: Electronic Health Record Integration of Predictive Analytics to Select High-risk Stable Patients With Non-st-segment Elevation Myocardial Infarction for Intensive Care Unit Admission

Abstract

Background: Nationwide, intensive care unit (ICU) utilization for initially stable patients with non-ST-segment elevation myocardial infarction (NSTEMI) is not associated with patient risk. Use of the ACTION ICU risk score, which predicts clinical deterioration requiring ICU care in initially stable NSTEMI patients, could guide admission of high-risk patients to the ICU and low-risk patients to a lower acuity unit. Methods: We created a modified best practice advisory (BPA) within the electronic health record (EHR) at a single institution. The BPA semi-automatically calculates the ACTION ICU score (5 elements automatically populate from the EHR and 4 are entered manually) and recommends a location for admission based on a 10% risk threshold for clinical deterioration over the course of admission. The BPA was triggered for all ED patients with serum 4 th generation troponin T above the local upper limit of normal. Physicians could temporarily hide the BPA, permanently cancel it if they felt the patient’s presentation was not primarily due to NSTEMI, or generate the ACTION ICU score. We measured how ED physicians used the BPA, and clinical and utilization outcomes for patients admitted through the ED with a discharge diagnosis of NSTEMI in the 12 months before and after BPA roll-out. Results: Between August 14, 2017 and August 13, 2018, the BPA triggered 972 times. It was hidden until the patient left the ED 230 times (23.7%) and canceled 561 times (57.7%). Providers opted to calculate a risk score 181 times (18.6%), and a score was successfully calculated 146 times. Among 135 patients for whom the BPA triggered that had a final hospital diagnosis of NSTEMI, the BPA was inappropriately canceled in 62 (45.9%) and hidden in 16 (11.9%). Overall, there were 190 NSTEMI admissions through the ED in the year after BPA integration into the EHR and 253 in the year prior. In the year after BPA integration 32.6% of the NSTEMI patients were admitted directly to the ICU compared with 37.5% admitted to ICU prior to BPA (p=0.32). No change was found in the distribution of ACTION ICU scores of NSTEMI patients admitted to the ICU prior to vs after BPA integration, as well as no differences in ICU length of stay (p=0.96), hospital length of stay (p=0.27), the proportion of patients transferred from the ward to the ICU (p=0.78), or in-hospital mortality (p=0.18). Conclusions: Embedding the ACTION ICU risk score into the EHR did not affect clinical or utilization outcomes for patients presenting to the ED with NSTEMI, but was limited by inappropriate cancellation of the risk score calculator. Better EHR mapping to enable risk calculation without the need for user input may be needed for successful deployment of predictive analytics in this patient population.