Abstract 366: Evaluating an AI‐Driven Stroke Triage Platform: Stratified Insights from 4,548 Transfers to Four Thrombectomy Hubs from Sixty Spokes

Abstract

Background We evaluated the impact of implementing an artificial intelligence (AI)‐based acute ischemic stroke (AIS) triage system across a large academic health network. Methods A prospectively maintained database was analyzed for equivalent time periods before and after Viz.ai implementation (January 2021‐December 2022). Outcomes included transfer rates from 60 spokes to four hubs, futile transfer cost estimates, and workflow metrics [CTA‐to‐door‐out, door‐in‐door‐out (DIDO), and door‐to‐puncture (DTP)]. Clinical and procedural outcomes, as well as intravenous thrombolysis (IVT) and endovascular therapy (EVT) rates, were also assessed. Data were stratified by admission type (direct vs transferred), Viz‐enabled versus non‐Viz spokes, and academic versus community thrombectomy centers. Results Among 4,548 AIS admissions (2,341 pre‐Viz vs 2,207 post‐Viz), EVT was performed in 394 and 450 cases, respectively. Following Viz.ai adoption, EVT rates among transferred patients increased from 39.5% to 47.3% (19.8% relative increase; P = 0.01), with the largest effect at Viz‐enabled spokes (36.3% to 51.9%; 42.9% relative increase; P = 0.001), corresponding to estimated cost savings of $3.16M per 1,000 patients ( P interaction < 0.001). DIDO times improved by 14 minutes [118 (99‐147) vs 104 (98‐116); 11.9% reduction; P = 0.004] in Viz spokes but not in non‐Viz spokes ( P interaction < 0.001). DTP decreased overall by 6.5 minutes [62 (24‐91) vs 55.5 (22‐74); 10.5% reduction; P = 0.0002], with the most pronounced reduction at the community EVT center [85.5 (47‐138.5) vs 51.5 (22‐115.5); P < 0.001] compared to academic centers [60 (23‐87) vs 56 (22‐74); P = 0.02; P interaction < 0.001]. Conclusions Implementation of an automated AI triage system reduced futile transfers, shortened workflow times, and yielded substantial cost savings, particularly in community hospital settings. These findings support AI integration as an effective tool for optimizing stroke care in hub‐and‐spoke networks. image image

Ähnliche Arbeiten