The Cost-Effectiveness of an Artificial Intelligence Software for Stroke Imaging and Treatment Decisions in NHS Hospitals
Author(s)
Nichola R. Naylor, BSc, MSc, PhD1, Jeff Wyrtzen, BA2, Zoe Woodhead, BSc, MSc, PhD2, George Harston, FRCP, MBBChir, MA, DPhil2.
1VP Health Economics, OI Pharma Partners Ltd, London, United Kingdom, 2Brainomix Limited, Oxford, United Kingdom.
1VP Health Economics, OI Pharma Partners Ltd, London, United Kingdom, 2Brainomix Limited, Oxford, United Kingdom.
OBJECTIVES: Brainomix 360 Stroke is an intervention consisting of AI software decision-support modules that supports patient treatment decisions from CT images, for patients with suspected stroke. Our main objective was to evaluate the cost-effectiveness of this intervention when deployed across multiple stroke hospital networks in England. A secondary objective, linked to this, was to provide an open-access model in R simulating the relevant patient pathways.
METHODS: A healthcare system perspective was used. A Markov model was constructed to estimate the long-term (lifetime) impacts of thrombolysis (TPA) and mechanical thrombectomy (MT). The outcomes of this were fed through a decision tree model based in NHS England hospitals. This modelled the treatment pathways of ischaemic stroke patients as an annual cohort. An R model was constructed, using OpenAI's GPT-4-turbo and GitHub co-pilot to assist with coding. We compare the intervention to a ‘no AI software decision support’ scenario for stroke care. Costs (in GBP) and Quality-Adjusted Life Year (QALY) impacts of the different scenarios tested are provided in 2023 values. Scenario analyses and probabilistic sensitivity analyses were also performed.
RESULTS: The deterministic model estimates the intervention to dominate, with lower costs and higher QALY gains. How long term impacts are modelled (across costs, utility and mortality) had a relatively large impact on net monetary benefit. Additionally, the cohort average start age, the discount rate and percentage of patients eligible for TPA and MT impacted results markedly, though the intervention was estimated to be cost-effective across all scenarios modelled. The probabilistic model suggests at a £20,000 per QALY gained threshold there is over a 95% chance of cost-effectiveness.
CONCLUSIONS: We highlight the cost-effectiveness of an intervention that increases the number of TPAs and MTs performed in those eligible, whilst providing the first downloadable cost-utility model that covers the patient pathway in this setting.
METHODS: A healthcare system perspective was used. A Markov model was constructed to estimate the long-term (lifetime) impacts of thrombolysis (TPA) and mechanical thrombectomy (MT). The outcomes of this were fed through a decision tree model based in NHS England hospitals. This modelled the treatment pathways of ischaemic stroke patients as an annual cohort. An R model was constructed, using OpenAI's GPT-4-turbo and GitHub co-pilot to assist with coding. We compare the intervention to a ‘no AI software decision support’ scenario for stroke care. Costs (in GBP) and Quality-Adjusted Life Year (QALY) impacts of the different scenarios tested are provided in 2023 values. Scenario analyses and probabilistic sensitivity analyses were also performed.
RESULTS: The deterministic model estimates the intervention to dominate, with lower costs and higher QALY gains. How long term impacts are modelled (across costs, utility and mortality) had a relatively large impact on net monetary benefit. Additionally, the cohort average start age, the discount rate and percentage of patients eligible for TPA and MT impacted results markedly, though the intervention was estimated to be cost-effective across all scenarios modelled. The probabilistic model suggests at a £20,000 per QALY gained threshold there is over a 95% chance of cost-effectiveness.
CONCLUSIONS: We highlight the cost-effectiveness of an intervention that increases the number of TPAs and MTs performed in those eligible, whilst providing the first downloadable cost-utility model that covers the patient pathway in this setting.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
EE692
Topic
Economic Evaluation
Disease
Cardiovascular Disorders (including MI, Stroke, Circulatory)