Future Innovations in Novel Detection of Heart Failure: A proof-of-concept study of a machine learning algorithm to increase early detection of heart failure in the community

Heart failure (HF) is most often detected late, when a patient has deteriorated to the extent that they require admission to hospital, in spite of most patients having previously presented with symptoms of HF in primary care. There are now an unparalleled range of treatments for patients with HF which reduce hospitalisation and death due to HF. Therefore early detection of HF in primary care may reduce morbidity and mortality for patients and reduce health expenditure for the NHS by avoiding hospital admissions. I have developed a supervised machine learning algorithm that can be implemented at scale in primary care electronic health records to identify individuals at risk of undiagnosed HF (FIND-HF). In this single arm interventional non-randomised proof-of-concept study I will invite individuals without a diagnosis of HF at a range of FIND-HF scores to attend for symptom assessment, point-of-care natriuretic peptide testing and AI-enabled echocardiography at a local clinical community site. I will assess the yield of undiagnosed HF for patients at different FIND-HF risk scores, as well as how treatment can be improved for newly detected HF cases. This proof-of-concept study is designed to inform a larger randomised clinical trial of FIND-HF-guided community-based early detection and treatment of HF to prevent HF hospitalisation and death. During a follow-on fellowship I aim to conduct this randomised clinical trial, investigate health economics of an early diagnostic approach, and consider barriers and facilitators to clinical implementation of FIND-HF.