M23 Identification of Asthma-COPD overlap using a novel hand-held capnometer and interpretable machine learning

<h3>Rationale</h3> It is estimated that 27% of asthma patients and 30% of COPD patients have Asthma-COPD Overlap Syndrome (ACOS). Patients with ACOS experience more frequent exacerbations, quicker deterioration and higher mortality than patients with only asthma or COPD. According to the Dutch Hypothesis, asthma and airway hyper-responsiveness predispose patients to COPD. This has been supported by research showing asthmatics have a 10x higher risk of developing chronic bronchitis and a 17x higher risk of developing emphysema than non-asthmatics. Spirometry is the current gold-standard for diagnosing ACOS but is technique-dependent, non-specific, and requires administration by a highly-skilled healthcare professional. This, along with a lack of precise clinical criteria for ACOS, results in significant under and misdiagnosis. There is therefore a need for a simple, reliable, and precise diagnostic test to identify when asthmatics have progressed to ACOS. The objective of this study was to accurately differentiate ACOS from Asthma by applying machine learning techniques to a tidal breathing CO₂ capnogram captured using TidalSense’s N-Tidal™ handheld capnometer. <h3>Methods</h3> Capnograms were collected from five longitudinal observational clinics that recruited Asthma and ACOS patients from primary and secondary care. A logistic regression model was trained on 80 features derived from these capnograms and performance was measured on an unseen test set of 52 Asthma, and 15 ACOS participants. Varying levels of obstruction were present, with% predicted FEV1 in both cohorts ranging from 19 to 101. <h3>Results</h3> The classification model yielded AUROC of 0.96, sensitivity 92%, specificity 87%, positive predictive value (PPV) 87% and negative predictive value (NPV) of 92%. A likely clinical use case involves ruling-in or ruling-out a diagnosis in patients classified with high confidence. Evaluating performance on the 71% of capnograms classified with &gt;80% confidence yielded PPV 0.96 and NPV 0.99. Waveform features driving classification were related to the alpha angle region, which represents the transition of alveolar gas to larger airways. These features correlated with FEV1/FVC, supporting their hypothesized role as markers of obstruction. <h3>Conclusion</h3> The N-Tidal^TM device could be used alongside interpretable machine learning as an accurate, rapid, point-of-care test to identify asthmatics’ progression to COPD.