UK Stroke Forum 2025 | The prognosis of ischemic stroke and intracerebral hemorrhage imaging-based subtypes
Alice Hosking, MPhys, MBBS, MRCP, The University of Edinburgh, Edinburgh, UK, discusses a study examining the prognosis of ischemic stroke and intracerebral hemorrhage imaging-based subtypes. Dr Hosking highlights that the study utilized a natural language processing algorithm to identify stroke types and locations from radiology reports, and found that mortality, recurrence rates, and dementia outcomes varied significantly between different stroke subtypes. This interview took place at the UK Stroke Forum (UKSF) 2025 Conference in Aberdeen, UK.
These works are owned by Magdalen Medical Publishing (MMP) and are protected by copyright laws and treaties around the world. All rights are reserved.
Transcript
Yeah, so this was a study we did looking at the outcome of people after stroke by stroke location. So we know that outcome after stroke varies by stroke location and studies of outcome after stroke have been done in healthcare data around the world, but nobody has looked at routinely collected data and included stroke location. And this would be really useful for patients to provide them with personalized prognostic information...
Yeah, so this was a study we did looking at the outcome of people after stroke by stroke location. So we know that outcome after stroke varies by stroke location and studies of outcome after stroke have been done in healthcare data around the world, but nobody has looked at routinely collected data and included stroke location. And this would be really useful for patients to provide them with personalized prognostic information. So we had access to all the radiology reports in Scotland for CT and MRI heads from 2010 to 2018. And we applied a natural language processing algorithm developed by our team at Edinburgh in the informatics department. And it’s called NLP, but we’ve named it Eddie, and it works a bit like ChatGPT, and this algorithm was able to identify from the radiology reports stroke types and their location and also their recency. So we linked the radiology reports to hospital inpatient data, death data, and community prescription data and performed a matched case-control study to see if we could identify any differences in outcomes after stroke between people with cortical ischemic stroke and deep ischemic stroke and lobar intracerebral hemorrhage and deep intracerebral hemorrhage. And we looked in two ways. So we looked at cumulative effects over 10 years and then we also looked at hazard ratios with fully adjusted models and we found so some of our key findings include that mortality was higher for intracerebral hemorrhage and for ischemic stroke types, particularly in the first year, and mortality was higher for cortical ischemic stroke than for deep ischemic stroke. Recurrence rates were highest for lobar intracerebral hemorrhage and dementia outcomes were also highest for lobar intracerebral hemorrhage. So we’ve also looked at some other outcomes which will be available in our full paper and I think what we’ve demonstrated is that you can use this natural language processing method to get information from the radiology reports. And we’ve also found we’ve generated numbers that you can apply to your own patients from similar populations to ours.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
Read more...
