ICF: Using Explainable Artificial Intelligence to predict future stroke using routine historical investigations
Lead Research Organisation:
Plymouth University
Department Name: Peninsula Medical School
Abstract
There are more than 100,000 strokes in the UK each year causing 38,000 deaths, making it a leading cause of death and disability (NICE 2019). Ninety five percent of those who had a stroke had at least one untreated risk factor for it and 13.7% of these strokes were preventable. Findings that are predictive of future stroke are often identifiable on brain scans, electrocardiograms (ECG), heart scans (Echocardiogram or 'echo') and laboratory tests undertaken taken to investigate other medical problems. Often these signs are not picked up, which means evidence based treatments to reduce the risk of future stroke are not given. This is because of a lack of a dedicated system to do so. The first five years of care post stroke cost £3.60 billion (mean per patient cost: £46,039). Hence as well as improving many lives, a cost effective and accurate system to identify those at high risk of future stroke them will deliver major cost savings.
Using 10,000 stroke cases seen at University Hospitals Plymouth NHS Trust (UHPNT) and with the close support of the Sentinel Stroke National Audit Programme (SSNAP - a national audit where all UK stroke cases are recorded), we hope to build a database of laboratory results, Magnetic resonance imaging (MRI) and computer tomography (CT) brain scans, ambulatory ECGs (ECGs which are worn for 24hr+) and echocardiograms, collected in those who later developed and did not develop a stroke. We shall use this data to train an artificial intelligence computer programme (model) which can predict who will later develop strokes based on patterns within the data collected. We believe that this approach will not only identify known risk factors for stroke, but may identify new patterns/features in one or across a number of investigations to predict future stroke. We hope this model will be the first step to building an automated system (which interfaces directly to GPs) for determining stroke risk and implementing treatments and lifestyle modifications to reduce this risk.
Using 10,000 stroke cases seen at University Hospitals Plymouth NHS Trust (UHPNT) and with the close support of the Sentinel Stroke National Audit Programme (SSNAP - a national audit where all UK stroke cases are recorded), we hope to build a database of laboratory results, Magnetic resonance imaging (MRI) and computer tomography (CT) brain scans, ambulatory ECGs (ECGs which are worn for 24hr+) and echocardiograms, collected in those who later developed and did not develop a stroke. We shall use this data to train an artificial intelligence computer programme (model) which can predict who will later develop strokes based on patterns within the data collected. We believe that this approach will not only identify known risk factors for stroke, but may identify new patterns/features in one or across a number of investigations to predict future stroke. We hope this model will be the first step to building an automated system (which interfaces directly to GPs) for determining stroke risk and implementing treatments and lifestyle modifications to reduce this risk.
