In a recent study published in JAMA Neurology, researchers evaluate the implementation of automated software to detect large vessel occlusion (LVO) from computed tomography (CT) angiograms to improve endovascular stroke therapy workflows.
Study: Automated Large Vessel Occlusion Detection Software and Thrombectomy Treatment Times: A Cluster Randomized Clinical Trial. Image Credit: SquareMotion / Shutterstock.com
Background
The timely implementation of endovascular thrombectomy is critical for improving patient outcomes after an acute ischemic stroke (AIS) with LVO. The time between the patient’s arrival at the hospital and initiation of endovascular thrombectomy has become an important metric for a hospital to receive a stroke center certification, with many concerted efforts made to reduce this time.
Some challenges to reducing this workflow time have been the detection of a possible AIS with LVO by the clinicians or radiologists, as well as communicating the need for an endovascular thrombectomy to the care team for its execution.
The use of artificial intelligence (AI) in the diagnosis of various medical conditions using CT images is being extensively explored. Thus, using automated AI-based methods for LVO screening of CT angiograms of patients presenting with possible AIS could reduce the time between assessment and endovascular thrombectomy.
About the study
In the present study, researchers utilize a randomized stepped-wedge clinical trial to determine the efficiency of an AI-based automated system in detecting LVO in possible AIS patients and improving the assessment and workflow time between hospital arrival and the initiation of endovascular thrombectomy. The randomized stepped-wedge method was implemented to circumvent issues associated with randomizing the analysis at the individual patient level while retaining the robustness of randomized evaluation.
The trial was conducted across four comprehensive stroke centers in the greater Houston region between January 2021 and the end of February 2022. After being provided clearance from the United States Food and Drug Administration (FDA) for the use of this AI platform for clinical care, in addition to significant financial support received for the implementation of the software, a stepped rollout in hospital-level clusters was performed.
Trial participants included patients who presented at the emergency departments of these four comprehensive stroke centers with symptoms of AIS with LVO and underwent CT angiography imaging. All patients who underwent endovascular thrombectomy for AIS with LVO of the middle cerebral, internal carotid, anterior cerebral, posterior cerebral, basilar, or intracranial vertebral arteries were included in the study.
Patients who presented as in-hospital stroke codes or had been transferred from other centers that did not perform endovascular thrombectomy were excluded from the analysis, as the workflow time for these patients was significantly different. For patients transferred from other centers, the decision for an endovascular thrombectomy has already been made, and they are taken directly for the procedure without further imaging, which would change the workflow time.
The intervention included activation of the automated AI-based LVO detection from the CT angiogram, which was coupled with a secure messaging system. This system was activated in the four comprehensive stroke centers in a random-stepped manner. The activated system alerted radiologists and clinicians on their mobile phones of a possible LVO minutes after the completion of CT imaging.
Primary study outcomes included the impact of the AI-based automated LVO detection system on the door-to-groin time, which was determined using a linear regression model. The secondary outcome was the time elapsed between arrival at the hospital and administration of the intravenous tissue plasminogen activator, the time between initiating the CT scan and beginning of the endovascular thrombectomy, and the duration of hospitalization.
Study findings
Implementing the AI-based automated LVO detection system, coupled with a secure application for communication using mobile phones, significantly improved the workflow time for in-hospital AIS. The implementation of this software across the four comprehensive stroke centers was associated with clinically relevant reductions in the treatment time for performing endovascular thrombectomy.
During the trial, about 250 patients presented at the emergency department of the four centers with LVO AIS. Implementing the AI-based automated system reduced the door-to-groin time by 11 minutes. Furthermore, mortality rates decreased by 60%, with the time between the initial CT scan and the start of the endovascular thrombectomy also associated with similar reductions.
Conclusions
The implementation of the automated AI-based system for detecting LVO among possible AIS patients, coupled with a secure application for communication, significantly reduced the in-hospital workflow and led to clinically significant reductions in endovascular thrombectomy treatment times.
Journal reference:
- Martinez-Gutierrez, J. C., Kim, Y., Salazar-Marioni, S., et al. (2023). Automated Large Vessel Occlusion Detection Software and Thrombectomy Treatment Times: A Cluster Randomized Clinical Trial. JAMA Neurology. doi:10.1001/jamaneurol.2023.3206
Machine learning helps predict stroke risk after CF-LVAD implantation