P239 Quantification of small airways disease in severe asthma using a novel, fast-response capnometer and interpretable machine learning

Résumé fourni par la source

Rationale Asthma is a major noncommunicable disease characterized by airway inflammation and hyper-responsiveness. It affected an estimated 262 million people in 2019 - causing 455,000 deaths - and amounts to $50 billion annually in direct healthcare costs in the USA alone. Small airways disease is a poorly understood contributory factor to asthma, targeting the non-cartilaginous eighth and higher generation of the tracheobronchial tree. It has been demonstrated to triple the odds of systemic corticosteroid use, and increase the odds of acute exacerbation six-fold. The prevailing measure of small airways obstruction is a low percentage predicted forced expiratory flow rate between 25% and 75% of vital capacity (% predicted FEF 25–75%) as measured during spirometry. This work aims to demonstrate the utility of signal-processing and statistical techniques on capnography data to assess the degree of small airways disease as measured by% predicted FEF 25–75%. Methods Capnograms were drawn from two longitudinal observational clinical studies (ABRS and GBRS) that recruited 85 participants with asthma from UK primary and secondary care. These capnography signals were denoised, and each capnogram translated into 25 geometric features. XGBoost was trained on the features from the capnograms of 82% of the patients to distinguish% predicted FEF 25–75% < 50% from ≥ 50%, and tested on ten capnograms from each of the remaining 18% of patients. Results The model achieved a micro-average AUROC of 93%, sensitivity of 95.0%, specificity of 76.0%, PPV of 88.8%, and NPV of 88.4%. The average machine learning model prediction probability output per participant was plotted against the average% predicted FEF 25–75% per participant, and the Pearson’s product moment correlation coefficient (r) between these two variables was calculated as -0.902. Conclusion Machine learning (ML) techniques applied to the processed capnography signal were able to accurately detect small airways disease in patients with asthma. The ML model’s probability output was purposed as an indicator of small airways obstruction, and was demonstrated to inversely correlate with% predicted FEF 25–75%. The analysis suggests that the N-TidalTM capnometer could be used as an accurate and rapid point-of-care test to identify small airways disease.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.