An abstract presented at the 2024 American Venous Forum (AVF) annual meeting (March 3–6) revealed that, in the treatment of patients with pulmonary embolism (PE), the implementation of artificial intelligence (AI) software has been linked to a shorter time to assessment.
At the Tampa, Florida, meeting, Jacob Shapiro, MD, shared this and other key findings from a review of patients diagnosed with PE at a single center over a six-year period.
Shapiro, a resident at TriHealth in Cincinnati, Ohio, and colleagues outline in their abstract that the serious and potentially life-threatening nature of PE necessitates a “streamlined PE workflow with timely assessment and initiation of treatment” to potentially improve a patient’s chance of survival. They add that AI has been increasingly used in healthcare to improve clinical efficiency.
The authors detail that, in October 2022, TriHealth implemented an AI-powered parallel workflow tool designed to automatically detect and triage patients with suspected PE. The aim of the study, they share, was to evaluate the clinical impact of AI software on time to assessment, time to anticoagulation, and patient outcomes at the institution.
Shapiro et al reviewed 150 patients diagnosed with PE between January 2017 and July 2023, retrospectively collecting data on these patients prior to AI implementation and comparing it against those of PE patients following AI implementation. The researchers looked at scan-to-assessment time, scan-to-alert time—which they note was used as a surrogate for scan-to-assessment time following AI implementation assuming best practice—time to anticoagulation administration, Pulmonary Embolism Response Team (PERT) activations, and in-hospital mortalities.
Shapiro reported at AVF that scan-to-alert time in the post-AI group of 45 patients was “significantly faster” than scan-to-assessment time in the pre-AI group of 113 patients. He added that anticoagulants were administered faster for post-AI cases with PERT activation compared to cases without PERT activation, and that in-hospital mortalities decreased from 8.4% (pre-AI) to 2.2% (post-AI), with all mortalities occurring in cases without PERT activation.
“Adoption of AI into our workflow was associated with faster time to assessment of PE patients,” Shapiro and colleagues conclude in their abstract. They elaborate that, with an average AI alert time of under six minutes, AI “optimizes standard of care by promoting quicker triage.”
Furthermore, they note that the combined benefit of AI and PERT activation was highlighted by faster anticoagulation administration and decreased mortality in their sample. “These findings suggest a link between earlier anticoagulation and reduced risk of mortality,” they write.
Speaking ahead of AVF, Shapiro stated that “further research is needed to determine if long-term patient outcomes are impacted by this technology.”
On the learning curve with this AI software, Shapiro remarked that it is “very minimal.” He explains: “The interface for reviewing CT [computed tomography] scans is very intuitive and the chat feature is easy to navigate. The patient’s lab values are also automatically imported and obviously displayed.”
Shapiro commented that overreliance on AI could be a drawback of the technology. However, he emphasized his overall opinion that, as an adjunct to an expert reviewer, there are “no real downsides.” Specifically, he gives the example of the ability to look at scans in views other than axial (i.e. coronal or sagittal) as a benefit for reviewers.
On the wider endorsement of this software, Shapiro notes that it has already been widely adopted by neurosurgeons for strokes and that his institution has been using it for PE for over 18 months. “We are already seeing it integrated into clinical practice,” he says.
