Stroke AI Adoption Concentrated in Well-Resourced Hospitals

Unequal distribution despite financial incentives

Radiology artificial intelligence tools designed to detect strokes are being adopted unevenly across U.S. hospitals, with patients at smaller and under-resourced facilities far less likely to benefit from the technology, according to research from the Neiman Health Policy Institute published in the American Journal of Neuroradiology.

The study examined Medicare claims data from 2020 to 2023, covering the period after the Centers for Medicare & Medicaid Services established a New Technology Add-On Payment for AI that detects large vessel occlusion in acute ischemic stroke. Despite this financial incentive—designed to help hospitals cover implementation costs during the early adoption phase—significant disparities emerged in who actually received AI-enhanced care.

Researchers analyzed over 2,100 acute ischemic stroke episodes across 1,100 healthcare facilities. By 2022, add-on-payment-backed AI was used in approximately 21% of stroke cases. However, usage was six times higher at large hospitals compared to smaller ones, and 1.5 times higher at comprehensive stroke centers. Hospitals serving more socioeconomically deprived areas were significantly less likely to deploy the technology.

Why it matters

This research reveals a troubling pattern: payment incentives alone cannot overcome the operational and infrastructure barriers that prevent smaller hospitals from adopting clinical AI. When advanced diagnostic tools concentrate in facilities that already deliver excellent care, the technology fails to reach patients who might benefit most—potentially widening rather than narrowing healthcare disparities in time-sensitive conditions like stroke.

Infrastructure gaps limit smaller hospitals

The findings suggest that economics, workflow integration challenges, and technical readiness create higher barriers for resource-constrained facilities than the add-on payment can offset. AI adoption was most common at hospitals with over 1,000 beds and those equipped with CT imaging capabilities.

"AI tools may support faster stroke evaluation, but operational readiness, infrastructure, and clinical workflows play a major role in determining whether these tools are actually used in practice," said co-author Maria X. Sanmartin, an assistant professor at the Zucker School of Medicine at Hofstra/Northwell, in a statement accompanying the research.

Documented barriers include difficulties integrating AI into existing systems, provider skepticism, and staff learning curves—challenges that disproportionately affect facilities without comprehensive stroke center designation.

Potential solutions for broader access

The researchers suggest that centralized AI-as-a-service models could help smaller hospitals access stroke detection algorithms without requiring full on-site implementation. These shared service hubs have proven effective in other contexts for bringing advanced capabilities to under-resourced settings.

"When adoption is concentrated in facilities that already excel in stroke care, it misses the opportunity to improve care in less-resourced settings where potential gains are the greatest," Sanmartin noted.

Overall, fewer than 15% of analyzed stroke cases involved AI use during the study period. Usage declined in 2023 as the temporary add-on payment code began to sunset. No disparities emerged based on patient demographics or stroke severity—the gap was driven entirely by facility characteristics.

The findings were first reported by Radiology Business, with lead author Casey Pelzl noting that "access to these technologies depends more on where a patient is treated than on their clinical needs." The study was supported by a grant from the American Heart Association.