After a stroke, clot-busting drugs (thrombolysis) or mechanical removal of clots (thrombectomy) can dramatically improve outcome. The effects depend on how quickly the right patients are selected and treated.
Our programme brought together researchers, people with lived experience, clinicians and policymakers to substantially improve services and patient outcomes.
We used AI to understand variation in thrombolysis use between units and its effects on outcome. Even after adjusting for case-mix, patients do better at hospitals with higher rates of thrombolysis. We identified approaches to enable units to increase appropriate treatment rates which would result in 400 more people not having lasting disability, and 280 fewer dying or experiencing severe disability after a stroke each year.
Based on our findings, NHS England funded work to successfully improve stroke treatment in six underperforming hospitals, now extended to 12 more hospitals. Our work was included in the National Clinical Guidelines for Stroke and the National Stroke Audit incorporates our computer code, allowing individual hospitals to actively identify problems and improve service delivery.
Thrombectomy requires specialist centres. Deciding where these centres should be located is complicated by the varying needs of patients, locations of existing services, and geographical differences. With national policymakers we built an AI model to identify optimal location of new centres. Conservative calculations suggest that c.300 people/year would be spared death or disability if this could be achieved. The announcement of new centres in Kent and Norfolk, as we recommended, is the first step towards this.
David Hargroves, National Clinical Director for Stroke, described PenARC’s work as “pivotal in the planning for the delivery of thrombectomy”, and stated,
“We are now able to deliver 4,500 procedures per year, which means 1,000 patients are less disabled. Their work has been critical to our ability to deliver this to the NHS.”
Learn more about Stroke Audit Machine Learning (SAMueL).
Updated Feb 2025.