Future Innovations in Novel Detection of Atrial Fibrillation after Stroke: Regulatory approvals and API integration to enable use in the NHS

Ischaemic stroke is the most common cause of acquired brain injury and severe disability in the UK,(1) including limb weakness, visual problems, and language and communication problems. There are 85,000 ischaemic strokes (strokes caused by thrombosis/infarction) per year in the NHS,(2) with an aggregate societal cost of £22 billion per year.(3) Atrial fibrillation (AF) is the cause of more than 20% strokes.(4) AF detection after stroke is important to prevent recurrent stroke but for many patients with stroke AF is not detected during hospital stay.(5) Accordingly guidelines recommend prolonged cardiac monitoring after ischaemic stroke to detect AF and prevent recurrent stroke,(6, 7) but doing this at scale in the NHS is infeasible due to staff and hardware costs and burden for patients. Utilising artificial intelligence in routine health data could address this emerging healthcare need to target prolonged cardiac monitoring after ischaemic stroke, and thereby reduce costs, efficiently diagnose AF, and prevent recurrent stroke to reduce hospitalization and long-term care needs. FIND-AFDAS (Future Innovations in Novel Detection of Atrial Fibrillation After Stroke) is a high impact, scalable patient-focused Tier C digital health technology as artificial intelligence as a medical device (AIaMD). It is a fixed meta-machine learning algorithm that uses routinely collected health data (age, sex, five comorbidities) to calculate the probability (probability score between 0 and 1) for an individual to have AF detected after a stroke event. Thus FIND-AFDAS acts to drive clinical management by informing healthcare professionals of patients most likely to benefit from prolonged cardiac monitoring for AF detection after stroke. FIND-AFDAS was developed and validated in data from 68,000 patients with an acute stroke event across eight countries (UK, Switzerland, Greece, Israel, Japan, Taiwan, and China).(8) Performance was accurate and robust across these different countries incorporating different ethnic groups, and equally good in men and women.(8) We then validated FIND-AFDAS in the relevant environment by calculating the FIND-AFDAS score in patients in the PER DIEM randomised clinical trial (RCT), where all 300 participants had experienced a stroke and then undergone prolonged cardiac monitoring, with 30 new AF cases detected. FIND-AFDAS demonstrated exceptional prediction accuracy (area under the receiver operating characteristic curve 0.98), and a FIND-AFDAS score of 0.11 led to sensitivity 100%, specificity 88%, negative predictive value 100% and positive predictive value 48% for AF detection after stroke using prolonged cardiac monitoring, thereby reducing the number needed to screen from 10 to 2.(8) Based on this evidence FIND-AFDAS is at technology readiness level 4 (component validated in laboratory environment). The FIND-AFDAS patient and public involvement group have steered and governed this research from inception. They identified detection of AF after stroke as an important approach to prevent brain injury by preventing future stroke. They co-designed the algorithm by highlighting the importance of utilising only routinely-collected variables as information that might be missing in routine data (genetics, laboratory tests) would disadvantage patients from under-served communities who less frequently access healthcare services and thus for whom there may be health data poverty for specialist investigations.