Structured Light Plethysmography in COPD/1.0
Research type
Research Study
Full title
An Observational, Comparative, Multi Centre Study, Validating the Structured Light Plethysmography against standard of care (Spirometry) in the diagnosis of Chronic Obstructive Pulmonary Disease for Patients who plan to undergo Spirometry Testing
IRAS ID
279303
Contact name
Gin Lee
Contact email
Sponsor organisation
(PneumaCare Ltd)
Clinicaltrials.gov Identifier
Duration of Study in the UK
1 years, 6 months, 30 days
Research summary
Research Summary:
This is an observational, comparative, multicentre study to validate the Thora3Di™ against standard practice in patients who are undergoing investigation for COPD. The core methodology involves capturing of data during a short period of measurement of breathing using SLP against spirometric outcomes.
This study is a comparative study with two study visits (Part A and Part B ) to generate data to characterise the tidal breathing patterns and parameters with Thora3Di™againt spirometry FEV1/FVC and %predicted. Subjects will have a Part A visit 1(Development Phase), and a Part B visit 2(Validation Phase) after developing algorithms for COPD diagnosis from Part A. At Part A visit 1, subjects will have two 5-minute SLP measurements (Pre and Post bronchodilator). At Part B visit 2, subjects will be seen in the clinics and have three 5-minute SLP measurements (pre-bronchodilator, post-bronchodilator and post spirometry testing) along with spirometry testing (pre-bronchodilator and post-bronchodilator). The SLP measurement should be performed prior to standard lung function tests with minimal impact on clinical time and no change to hospital attendance. Also, each visit subjects will be asked to report concomitant medications and adverse events, and fill in COPD assessment test (CAT™).Summary of Results:
Study Title: A Comparative, Multi Centre Study, to develop a model for Structured Light Plethysmography against standard of care (Spirometry) in the diagnosis of Chronic Obstructive Pulmonary Disease for Patients who plan to undergo Spirometry Testing.
Study Acronym: PMC-SLPCOPD-2020
IRAS ID: 279303
Public Database Registration Number: clinicaltrials.gov NCT04584801 Study Sponsor: PneumaCare Ltd.We extend our thanks to everyone who participated in this clinical study sponsored by PneumaCare Ltd. This work has been vital in advancing research to improve the diagnosis of Chronic Obstructive Pulmonary Disease (COPD).
COPD is a progressive lung disease marked by persistent airflow limitation and remains a major global cause of illness and death. Its true prevalence is difficult to quantify, but undiagnosed COPD is thought to drive substantial preventable morbidity, hospitalisations, and healthcare costs. Current diagnosis relies heavily on post bronchodilator spirometry and fixed thresholds for airflow obstruction, which have important limitations: they require forced manoeuvres, depend on patient technique, struggle to detect early disease and cannot capture regional variation in lung function. These shortcomings highlight the need to reassess diagnostic approaches and develop new tools. As Artificial Intelligence (AI) becomes more integrated into healthcare, machine learning–based methods offer promising opportunities for earlier COPD detection and more accessible, comprehensive assessment of lung function.
Purpose of the Study
This study aimed to develop and test a classification model using Structured Light Plethysmography (SLP) to support COPD diagnosis.Study Participants
The study recruited adults with COPD and smokers aged 35 years or older with a history of at least 10 pack‑years, from 7 sites across England.
A total of 201 subjects were enrolled (3 screen failures) leaving 198 participants (92 men, 106 women). A further 15 subjects were excluded due to insufficient SLP data yielding 183 datasets for analysis.Study Design
Demographic, medical history and clinical data were collected, along with the validated COPD Assessment Test (CAT). Pre and post bronchodilator SLP scans were recorded during tidal (normal) breathing in a single visit, alongside standard spirometry.Results
Study data were cleaned, normalised and pre processed before analysis. Multiple predictive models were built using different combinations of clinical variables and SLP derived parameters, then evaluated, optimised, and validated for performance and stability. The final model was selected based on diagnostic accuracy, consistent data availability and clinical relevance.
To support use across different clinical settings, the model includes four adjustable sensitivity thresholds (80%, 82%, 85% and 90%), allowing clinicians to balance false positives and false negatives. Overall performance was strong, with an area under the curve (AUC) of 0.865. Sensitivity (Sn) and specificity (Sp) remained robust across thresholds—for example, at 80% sensitivity the model achieved a specificity of 0.808 (Sn 0.775, Sp 0.808), and at 90% sensitivity, specificity was 0.643 (Sn 0.892, Sp 0.643)—demonstrating its adaptability to varying diagnostic priorities.Two post-hoc analyses were performed:
i) Using Lower Limit of Normal (LLN) criteria, which adjust for age-related physiological changes, and
ii) Comparing COPD-severity groups to assess differences between LLN-based and fixed threshold classifications.The model distinguished severe COPD from non COPD smokers very effectively (AUC 0.91 and 0.90 for fixed ratio and LLN criteria), indicating that SLP captures real, physiologically meaningful differences in breathing patterns that are most evident in advanced disease. Using LLN, discrimination between mild–moderate COPD and non COPD smokers was weaker (AUC 0.67). Age is a major driver of separation between non COPD smokers and those with severe COPD, as older smokers are more likely to have progressed to advanced disease. Within the COPD group, age varies little, so the AUC of 0.72 for mild–moderate versus severe COPD reflects the true discriminative power of breathing pattern data alone. Mild–moderate COPD remains harder to classify because these patients resemble severe COPD demographically but have breathing patterns closer to non COPD smokers.
Conclusion
A high performing machine learning model was successfully developed to identify COPD using SLP. These results suggest that SLP based models could support clinical decision making and serve as an accessible screening tool for earlier diagnosis. The next step is to validate the model’s accuracy and reliability in an independent study cohort.Further Information
Further details about the study can be found at https://gbr01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.clinicaltrials.gov%2F&data=05%7C02%7Csheffield.rec%40hra.nhs.uk%7C3aa2d7126ba541c985b108de8c1e7374%7C8e1f0acad87d4f20939e36243d574267%7C0%7C0%7C639102259570254408%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=a4BYz6KLwN3iNFBYZL3oUauKYcNgZr2sI%2FYr0QXSL%2Bk%3D&reserved=0 (ID number: NCT04584801).REC name
Yorkshire & The Humber - Sheffield Research Ethics Committee
REC reference
21/YH/0004
Date of REC Opinion
11 Feb 2021
REC opinion
Further Information Favourable Opinion