Acceptability, feasibility and utility of presenting live AI-based risk prediction models to a COPD MDT

<bold>Introduction:</bold> We have developed performant, fair, explainable and actionable AI-based risk prediction models to enhance COPD care. <bold>Background:</bold> The DYNAMIC-AI observational cohort clinical investigation (<ext-link>NCT05914220</ext-link>) is evaluating the patient acceptability, technical feasibility, safety and utility of providing live 12-month mortality and 3-month readmission model risk scores to clinicians in the Lenus ‘AI insights’ app. <bold>Methods:</bold> Study recruitment was from April 2023 – January 2024. Participants with COPD were recruited through the NHS GG&C digital COPD support service, with patient information and consent flow within the established patient app. Acceptability has been measured by recruitment numbers. Feasibility has been measured by the number of real-time model scores provided for review to the COPD multi-disciplinary team. <bold>Results:</bold> 130 patients consented to participate, with only 14 people declining consent. Model inferencing was successful on 121 of 130 participants, with non-accessible electronic healthcare data preventing inferencing for 9 participants. There have been no adverse events or device deficiencies. Early MDT experience has noted utility of risk scores, with emerging respiratory failure and prescribing model features as insights which could prompt proactive clinical interventions. <bold>Conclusions:</bold> Providing AI insights to a COPD MDT is acceptable to patients and technically feasible. Early experience in live clinical use suggests considerable potential for reorientation of COPD care. Secondary objectives including prospective model performance and clinical experience will continue to be captured during the 12-month follow up phase of the DYNAMIC-AI clinical investigation.