The American Heart Association (AHA) has released a scientific statement on virtual stroke networks and the role of artificial intelligence (AI), mobile imaging applications, and telehealth, as reported in Stroke.

The statement focuses on initial triage, transport, diagnosis, and treatment of stroke, and does not address in-hospital care, recovery, or rehabilitation.

Telestroke networks have increased patient access to care(NOT RECOVERY!) and the use of thrombolysis in underresourced areas, and mobile and broadband technology have expanded telestroke networks outside the hospital. Mobile access to neuroimaging has enabled optimization of transfer and triage decision-making for patients who need neuroendovascular care(NOT RECOVERY!) for acute ischemic stroke, and AI gives nonradiology professionals access to complex neuroimaging results remotely in real time. AI-related challenges include building robust stroke datasets, validating AI applications for acute stroke care(NOT RECOVERY!) safely, and assessing newly designed AI technologies before implementation, the AHA writing group noted.

AI software platforms are intended for triage and do not replace formal interpretations by a radiologist or clinician. “Most AI tools for stroke care(NOT RECOVERY!) in development are in the untested pilot phase and are not ready for widespread clinical use,” the authors cautioned.

 

Clinicians should lead in creating best practices to guide education and implementation of AI-enabled virtual stroke care(NOT RECOVERY!) in compliance with current practices.

Teleproctoring and telerobotics may be practical alternatives in a virtual stroke network, although barriers include lack of sufficient reliable broadband, manual procedural components, and complex postprocedure care(NOT RECOVERY!) needs.

Technologic advances in AI-based applications and mobile health could potentially support acute stroke identification, transfer, triage, and treatment outside the hospital, and in-hospital technology-supported care(NOT RECOVERY!) could improve diagnosis, treatment, prognosis, and monitoring of stroke-related performance, according to the writing group.

Use of AI-based tools in virtual acute stroke networks involves ethical concerns. “The risk of automation bias from passively accepting the outputs of AI algorithms risks dampening acute stroke diagnostic skills for professionals, enhancing disparities in care(NOT RECOVERY!) stemming from biased data sets, and clouding the interpretation and explainability of acute stroke treatment decisions,” the group wrote.

AI health care(NOT RECOVERY!) algorithms can be limited by the quality of the data used for training, as women, underrepresented racial and ethnic groups, and individuals from lower socioeconomic backgrounds frequently are underrepresented in digital health data sets. “AI models should be evaluated for their generalizability across diverse populations, and ongoing bias surveillance throughout the AI life cycle is essential to promote equitable health care(NOT RECOVERY!) outcomes and reduce disparities,” the authors wrote.

Access, outcomes, and cost should be considered during research into the use and implementation of these new technologies. Cost-efficiency analyses should be conducted on AI-applied health care(NOT RECOVERY!) delivery compared with standard health care(NOT RECOVERY!) delivery models.

“Clinicians should lead in creating best practices to guide education and implementation of AI-enabled virtual stroke care(NOT RECOVERY!) in compliance with current practices,” the AHA writing group wrote. “Implementation within a stroke system should be informed by evidence, regional resources, and stakeholder feedback.”

Disclosure: Some of the study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.