BACKGROUND Early detection of acute exacerbations of chronic obstructive pulmonary disease (AECOPD) remains a critical challenge in COPD management. This study introduces a novel machine learning score that leverages vital signs collected at home through remote patient monitoring to detect AECOPD events in advance. OBJECTIVE The study aims to evaluate the predictive performance of the BVS3 composite risk score in identifying AECOPD events ahead of clinician-defined episodes. METHODS The eMEUSE-SANTÉ clinical trial (NCT04963192), conducted in France from 2021 to 2024, involved 220 COPD patients who were remotely monitored for six months using a CE-certified (Class IIa) connected wristband measuring oxygen saturation (SpO2), breathing rate (BR) and heart rate (HR). The BVS3 risk score was computed based on the Z-scores of these vital signs and its performance was assessed against objectively physician validated AECOPD events. RESULTS Continuous 24-hour monitoring of vital signs using a connected wristband was well accepted over the long term, with a median adherence of 86% indicating strong patient compliance. A total of 42 physician validated exacerbations of COPD (7 severe and 35 moderate) with no missing remote monitoring data were documented in 39 patients at a general hospital. The BVS3 risk score demonstrated excellent predictive performance, achieving an AUC of 0.94 for severe events and 0.88 for moderate and severe AECOPD combined. For severe exacerbations, BVS3 achieved a sensitivity of 86% at 94% specificity. When considering both moderate and severe AECOPD events, the BVS3 score anticipated exacerbations episodes an average of 4.4 ± 3.1 days before clinical confirmation, with an overall accuracy of 84.8% and sensitivity of 74% with 85% specificity. Individual Z-scores for each vital sign also showed specific predictive capabilities for moderate and severe events, with the heart rate (z-HR), breathing rate (z-BR) and heart oxygen saturation (z-SpO2) yielding AUCs of 0.83, 0.82 and 0.71 respectively, but with inferior performances compared with the combination of the 3 vital signs Z-scores. CONCLUSIONS Our findings highlight the acceptability and predictive performance of remote patient monitoring associated with machine learning algorithms for the early detection of AECOPD events. Passive collection of vital signs yields strong patient compliance, and the interpretable BVS3 risk score demonstrates high accuracy and anticipation for AECOPD prediction. This end-to-end digital solution, which requires limited patient involvement, may offer a scalable approach for early identification of AECOPDs in COPD care. By enabling earlier intervention, this solution could significantly improve patient outcomes through proactive disease management. CLINICALTRIAL ClinicalTrials NCT04963192; https://clinicaltrials.gov/study/NCT04963192
Tilquin et al. (Tue,) studied this question.