Assistant Professor Yale School of Medicine, United States
Purpose: Acute pulmonary embolism (PE) remains a predominant driver of mortality and morbidity despite advances in imaging and therapeutic modalities. In recent years, artificial intelligence (AI) algorithms have been utilized to triage the interpretation of computed tomography pulmonary angiography (CTPA) images by radiologists and have been shown to improve operational statistics such as report turnaround time (RTAT) and hospital length of stay (LOS). However, the effect these algorithms have on clinical outcomes warrants further exploration. The aim of this study was to examine how an AI triage algorithm affected medium-term pulmonary health in intermediate-high- and high-risk PE patients.
Materials and Methods: 154 intermediate-high-risk and high-risk PE patients were identified from an academic health center’s medical records. Intermediate-high-risk was defined as elevated brain natriuretic peptide or troponin with right heart strain, while high-risk was defined as systolic blood pressure < 90. Within this cohort, 52 patients were admitted prior to AI implementation (pre-AI), and 102 patients were admitted after AI implementation (post-AI). Pulmonary health outcomes recorded in the patient's chart within 1-year of PE were quantified using an internal, non-validated, scoring scale of 1-5, where 5 indicates the most severe symptoms.
Results: Results are summarized in the table. No improvement was found in mean pulmonary health score, time-to-treatment from initiation of CTPA, or LOS. Compared to the pre-AI cohort, the post-AI cohort had a statistically significantly longer mean RTAT. No differences were found in a subset analysis comparing RTAT for images acquired during the day versus night.
Conclusion: These findings suggest that triage AI algorithms may not significantly improve workflow or pulmonary health outcomes in intermediate-high-risk and high-risk PE patients. Additional studies are warranted with larger sample sizes.
