Detecting COPD exacerbations using AI predictive modelling with continuous temperature sensing

<bold>Introduction:</bold> Exacerbations of chronic obstructive pulmonary disease (COPD) affect morbidity and mortality, but these events are often delayed in presentation or unreported to healthcare professionals. We aimed to assess the feasibility of continuous temperature monitoring for early detection of COPD exacerbation. <bold>Methods:</bold> Participants from the London COPD exacerbation (EXCEL) cohort underwent at-home continuous body temperature monitoring using a wearable sensor, for up to 9 months. Exacerbations occurring during this period were defined clinically according to established EXCEL criteria, using daily symptom diary cards and patient reporting. A subgroup of exacerbations and previous data, including from lower respiratory tract infections, was used to train the AI system to predict a potential exacerbation event using changes in temperature signature in the preceding 7 days. The sensitivity of this AI prediction was assessed. Control regions, where exacerbations did not occur, were used to test specificity. <bold>Results:</bold> The study recruited 37 patients with COPD who were monitored for a total of 970 weeks (mean age 71.8±8.1yrs, 51% male). Patients had 30 exacerbations diagnosed clinically, of which 24 had sufficient data quality for AI prediction (80.0%). A total of 20 out of 24 exacerbations were predicted by the AI (83.3% sensitivity). In 32 control regions, only 3 predicted exacerbations did not occur (90.6% specificity). <bold>Conclusion:</bold> Continuous temperature monitoring is feasible in patients with COPD and might aid the early detection of exacerbations. Future work in larger cohorts should be undertaken to confirm this and the possible benefits of early detection and intervention in this group.