{
  "abstract": "Introduction and objectives Chronic obstructive pulmonary disease (COPD) consists of two main phenotypes: emphysema-predominant and airway-predominant COPD. The reference standard for diagnosing emphysema is the high-resolution computed tomography (HRCT) scan. However, its widespread use is challenged by cost, diagnostic availability, and exposure to radiation. The development of easy-to-use and non-invasive diagnostic methods is therefore necessary to detect emphysema in patients with COPD. A proposed method is capnography, using the TidalSense N-Tidal® handheld capnometer, which has already been proven to be an accurate test for distinguishing the severity of COPD through quantification of changes in physiologic dead space and alveolar compliance.The objective of this research was to accurately differentiate emphysema-predominant COPD from healthy controls by applying interpretable machine learning (ML) techniques to a 75-second tidal breathing CO2 capnogram captured using the N-Tidal® capnometer.Methods 31 patients with emphysema-predominant COPD (collected on the Exploring Novel Biomarkers for Emphysema Detection (ENBED) study, NCT05825261) and 41 healthy controls (HC) were recruited. HRCT was performed in all patients to confirm emphysema, with% emphysema ranging from 3% to 45%. Low complexity discriminative machine learning models were trained on 93 features extracted from the breath waveform. Leave-One-Out Cross-Validation (LOOCV) was used to evaluate model performance.Results Logistic regression classification yielded the following metrics: AUROC 0.991, sensitivity 0.936, specificity 0.976, positive predictive value (PPV) 0.967 and negative predictive value (NPV) 0.952. The primary waveform features driving classification were derived from the alpha angle region, which represents the transition of alveolar gas to larger airways. This is consistent with hypothesis based on the pathophysiologic changes that are known to occur in emphysema. Features from this region also correlated with FEV1/FVC (highest r = 0.61), supporting their hypothesised role as markers of obstruction.Abstract P150 Figure 1Receiver operating characteristic (ROC, left figure) and precision-recall (PR, middle figure) curves for the logistic regression model. The right figure shows the normalised average waveforms (averaged across the cohort) for emphysema-predominant COPD patients and healthy controlsConclusions This initial dataset has shown that the N-Tidal® device can be used alongside interpretable ML algorithms to accurately distinguish emphysema-predominant COPD patients from healthy subjects.",
  "authors": [
    {
      "affiliations": [
        "TidalSense Limited, Cambridge, UK"
      ],
      "name": "RH Lim"
    },
    {
      "affiliations": [
        "Maastricht University Medical Centre, Maastricht, Netherlands"
      ],
      "name": "LG Reinders"
    },
    {
      "affiliations": [
        "Roche Pharma Research and Early Development, Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland"
      ],
      "name": "J Axmann"
    },
    {
      "affiliations": [
        "Genentech Inc, South San Francisco, USA"
      ],
      "name": "WT Gaik"
    },
    {
      "affiliations": [
        "TidalSense Limited, Cambridge, UK"
      ],
      "name": "L Talker"
    },
    {
      "affiliations": [
        "TidalSense Limited, Cambridge, UK"
      ],
      "name": "G Lambert"
    },
    {
      "affiliations": [
        "Roche Pharma Research and Early Development, Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland"
      ],
      "name": "A Mackay"
    },
    {
      "affiliations": [
        "Roche Pharma Research and Early Development, Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland",
        "Genentech Inc, South San Francisco, USA"
      ],
      "name": "D Mohan"
    },
    {
      "affiliations": [
        "Roche Pharma Research and Early Development, Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland"
      ],
      "name": "V Erpenbeck"
    },
    {
      "affiliations": [
        "Roche Pharma Research and Early Development, Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland"
      ],
      "name": "J Dorn"
    },
    {
      "affiliations": [
        "TidalSense Limited, Cambridge, UK"
      ],
      "name": "AX Patel"
    },
    {
      "affiliations": [
        "Maastricht University Medical Centre, Maastricht, Netherlands"
      ],
      "name": "SO Simons"
    }
  ],
  "title": "P150 Diagnostic classification of the emphysema subtype of COPD using N-Tidal® capnography",
  "uid": "062c47ce-5cd5-5f5e-b556-e4d8e873b202"
}