Implementation of an artificial intelligence-enabled stethoscope in primary care: utilisation and incentive characterisation in the TRICORDER programme

Abstract

Abstract Background The TRICORDER (Triple Cardiovascular Disease Detection using an Artificial Intelligence Stethoscope) cluster randomised controlled trial and implementation study is the largest cardiovascular artificial intelligence (AI) deployment in the National Health Service (NHS). For such a newly introduced AI tool shown to be effective to be sustainably adopted in care, the role of incentivisation, updates and reinterventions need to be understood in order to avoid attrition of use and failure of clinical impact. The AI-enabled stethoscope supports primary care physicians to detect reduced left ventricular ejection fraction (LVEF), atrial fibrillation (AF), and valvular heart disease (VHD) within fifteen seconds during routine examinations. Purpose To examine AI stethoscope utilisation in TRICORDER stratified by locality and identify clinician-ranked incentives for consistent adoption. Methods 200 UK Primary Care Practices were cluster randomised to receive AI Stethoscopes for use in clinical practice (intervention) or to continue with standard care (control). Utilisation data were extracted from the manufacturer’s cloud database between 5th November 2023 and 16th February 2025. Practices were stratified by locality (borough). After 16 months of follow-up, a review of utilisation data informed an expert consensus on questionnaire items to formulate strategies for maximising utilisation. Practices were categorised into low, medium, and high utilisation groups, and for each user within these practices, three incentives for consistent use of the AI Stethoscopes - upgraded devices, electronic health record (EHR) integration and financial reimbursement per new patient diagnosis - were ranked. Results Among the 94 practices in the intervention arm, 15,988 AI stethoscope examinations were performed between 05 November 2023 and 16 February 2025. The mean (±SD) number of recordings per practice per month was 10.63 ± 12.25. Significant variation in utilisation rates was observed across boroughs (ANOVA, P < 0.001). Deprivation rates by borough relative to the rest of the country were compared. (Figure 1). The reintervention questionnaire received 146 responses within 30 days. Among high-utilisation practices, 72% ranked EHR integration as the most influential incentive for sustained AI stethoscope use, surpassing financial incentives and device upgrades (Figure 2). Conclusions Utilisation of the AI Stethoscope varies by borough, highlighting geographic differences in adoption and the need for tailored reintervention strategies. High-utilisation practices prioritised EHR integration as an incentive, emphasising the importance of seamless digital workflows for sustained clinical use.

Ähnliche Arbeiten