Accurate diagnosis of asthma using either single or longitudinal breath records captured on a novel fast response capnometer (ID 503)
TidalSense Limited
Funding: The studies which provided the data for this report were funded by NIHR (i4i grant) and Pfizer OpenAir. The authors had sole responsibility for the study design, data collection, data analysis, data interpretation and report writing. The ABRS study was supported by the National Institute for Health Research Invention for Innovation (NIHR i4i) Programme (Grant Reference Number: II-LA-1117-20002), the GBRS study was supported by Innovate UK (Grant Reference Number: 102977), the CBRS study was supported by SBRI Healthcare, the CBRS2 study was supported by Pfizer OpenAir.
Abstract
AimThe diagnosis of asthma can be challenging and often requires multiple diagnostic tests and forced expiratory manoeuvres, such as spirometry with reversibility testing or regular peak flow measurements in order to capture variable airflow obstruction. To assess the performance of a diagnostic model in its classification of participants with and without asthma, built using interpretable data processing and machine learning techniques applied to a dataset of CO2 breath records (75 seconds of tidal breathing), captured on TidalSense’s N-Tidal™ handheld capnometer.
Method
Participant records were drawn from 4 clinical studies (GBRS, ABRS, CBRS, CBRS2). This pooled dataset included participants recruited from primary and secondary care. Two XGBoost models were trained and validated on 82 features derived from the high-resolution CO2 data of 146 asthmatic and 133 non-asthmatic participants (which included healthy volunteers, those with COPD, bronchiectasis, pulmonary fibrosis, heart failure, anaemia, and other cardiorespiratory conditions).
The model used breath waveform features from a single breath record. The model was trained using 117 asthmatic, and 106 non-asthmatic participants and performance metrics were generated from an unseen validation set of 29 asthmatic, and 27 non-asthmatic participants. This was repeated 20 times with different validation participants for additional statistical power, and the average and variability of these metrics were recorded.
Results
The classification model achieved AUROC of 0.908 ± 0.016, sensitivity of 0.800 ± 0.043, specificity of 0.883 ± 0.012, positive predictive value (PPV) of 0.873 ± 0.010, and negative predictive value (NPV) of 0.817 ± 0.031 in detecting asthma from a single breath record.
Figure 1: Performance of the diagnostic classifier
Conclusion
TidalSense’s N-TidalTM capnometer and machine learning classifier could be used as an accurate, rapid, point-of-care diagnostic test for asthma, particularly in primary care. Future work will incorporate longitudinal capnography data into a diagnostic classifier.
Presented at BTS 2023
Conflicts of interest: LT, CD, ABS, JCC, HB, RHL, GL, AXP are currently employed, or were employed/funded at the time of the research, by TidalSense Limited. GH and HFA are funded by the National Institute for Health Research (NIHR) Community Healthcare MedTech and In Vitro Diagnostics Co-operative at Oxford Health NHS Foundation Trust. The views expressed in this publication are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.
The conference has been instigated and organised by PCRS. We are grateful to sponsors and exhibitors who have contributed funding towards this event in return for exhibition space. Neither sponsors or exhibitors have had any input into the agenda or the selection of speakers with the exception of any sponsored satellite symposia which are clearly indicated. View the full list of sponsors.