Abstract 057: Cost‐Effectiveness of AI in Stroke Care: A Systematic Review of Health Economic Evaluations

Introduction Artificial Intelligence has shown measurable gains in diagnostic accuracy, workflow efficiency, and clinical outcomes in stroke care. However, economic sustainability remains a critical consideration for widespread adoption. Health economic evaluations, particularly cost‐effectiveness analyses (CEAs), offer vital insights into the value of AI integration from payer and policymaker perspectives. This systematic review aims to synthesize current evidence on the cost‐effectiveness of AI technologies across the stroke care continuum. Methods Following PRISMA and CHEERS 2022 guidelines, we systematically searched MEDLINE, Embase, NHS Economic Evaluation Database, and IEEE Xplore from inception to June 2025. We included studies that performed CEAs, cost‐utility analyses, or cost‐benefit analyses on AI applications in any phase of stroke care (prehospital, acute, subacute, or rehabilitation). Key data extracted included setting, AI type, comparator strategy, perspective, time horizon, cost components, effectiveness metrics (QALYs gained), and incremental cost‐effectiveness ratios (ICERs). Data were synthesized and visualized using R and Python. Results A total of 17 eligible studies involving 11 countries were included, with sample sizes ranging from 500 to 150, 000 patients. The majority were model‐based CEAs (n=14), with 3 using real‐world implementation data. Most analyses adopted a healthcare payer or societal perspective and applied time horizons of 1 year to lifetime. Across studies, AI‐driven imaging triage and LVO detection consistently demonstrated cost‐effectiveness, with ICERs ranging from 7, 000 to 14, 000 per QALY gained. Workflow optimization tools integrating AI with telestroke networks reduced door‐to‐needle time by up to 15 minutes and achieved an ICER of 7, 000/QALY. Prognostic models and post‐stroke rehabilitation AI systems showed more variable results (16, 000‐21, 000/QALY), primarily due to limited evidence on long‐term outcome gains and higher upfront technology costs. All reviewed ICERs were well below conventional willingness‐to‐pay thresholds (50, 000/QALY), indicating high economic value. Sensitivity analyses showed robustness to cost assumptions and healthcare setting variability. However, only 4 studies accounted for implementation barriers such as clinician training, regulatory overhead, and interoperability costs. Reporting quality was high (CHEERS compliance score >85%) in 12 studies but varied in modeling transparency. Conclusion AI applications in stroke care, particularly for acute imaging triage and LVO detection, are consistently cost‐effective under established health economic thresholds. Workflow optimization AI yields the highest value, whereas rehabilitation‐focused tools need further outcome validation. Despite promising economics, few studies incorporate real‐world integration challenges. Health systems aiming to scale AI must align investment strategies with long‐term outcome modeling and equity considerations. Future research should focus on prospective CEAs embedded in clinical trials and health system implementation frameworks to better capture real‐world value. image