Stroke & AI

Unmet need

8 out of 10 stroke patients who leave the hospital do so with a disability. As a result, stroke costs UK society £26 billion per year. The most severe and disabling strokes are due to a large vessel occlusion (LVO).

The standard of care for LVO stroke in the UK has been the administration of thrombolysis (a clot-busting drug called alteplase). This is only modestly effective for LVO stroke patients, and the majority are dead or disabled at 1 year after their stroke. In 2015 a series of research trials demonstrated the efficacy of mechanical thrombectomy to open the occluded vessel for LVO stroke patients. This has transformed the potential care of patients with LVO stroke: patients who would have ended up in nursing homes are walking out of the hospital the next day. Each thrombectomy is thought to save the economy £80,000. 

In the UK fewer than 1% of stroke patients undergo thrombectomy, but it is estimated that over 10% would benefit. The key barrier to delivering this treatment is patient identification due to a lack of real-time neuroradiology imaging expertise at all sites to which stroke patients present, compounded by communication barriers across stroke networks. 

Within the Thames Valley, Oxford University Hospitals NHS Foundation Trust (OUHFT), based at the John Radcliffe Hospital, provides the specialist stroke care for patients requiring mechanical thrombectomy. In Oxford, approximately 30-50 patients undergo thrombectomy each year. The estimated number of patients across the Thames Valley who would benefit from thrombectomy is 300-400 per year.

Project aim

This project aims to evaluate the ability of an artificial intelligence (AI)-driven imaging support software to improve the delivery of acute stroke care in the Thames Valley stroke network. Identification of patients for interventions requires specialist radiological image interpretation in real-time, which is not readily available in most hospitals. We plan to use AI technology to aid image interpretation by front line physicians, facilitating patient diagnosis and speeding up decision making in this time-critical condition.
The primary objective will be to assess the impact of AI-based decision support on the identification of patients for transfer to the John Radcliffe Hospital in Oxford for specialist intervention. This intervention, mechanical thrombectomy, is for the most severe strokes and involves removing the blood clot causing the stroke using wires guided by X-ray studies. Additionally, within individual hospitals, the impact of AI-decision support in delivering acute stroke care such as thrombolysis will be assessed. The data generated will provide a case for wider NHS adoption, providing a cost-effectiveness analysis to form part of a NICE technology appraisal.
The AI-decision support is the CE-marked e-stroke supplied by Brainomix Ltd, an Oxford-based company. The e-stroke includes tools for non-contrast CT and CT angiography brain scans, in addition to tools that connect physicians and facilitate information transfer. The software supports physicians to identify patients who would and would not benefit from acute stroke treatments, by identifying treatment targets (large vessel occlusions) and selecting patients who are most likely to benefit from intervention.

Project objectives

The aim of this study is to evaluate the feasibility of artificial intelligence-enabled imaging support solution (e-stroke), combined with digital connectivity to improve stroke care in an NHS network. If the implementation is successful across a network, it is valued and adopted by clinicians, and workflow is improved, this will lead to a larger scale health economic analysis of the e-stroke across several NHS stroke networks.
Specific outcomes include:

1. Demonstration of that such an AI-enabled support solution can be integrated into an NHS network
2. Impact on the total number of mechanical thrombectomy procedures and thrombolysis given across the region
3. Evaluation of patient flow metrics including CT-to-decision time, door-in-door-out time in referring hospitals, door-to-needle time, number of patients transferred, and proportion transferred who undergo mechanical thrombectomy
4. Clinician satisfaction and practice impact evaluation at both referring and accepting hospital sites
5. Cost-effectiveness analysis to contribute to a NICE technology appraisal