The goal of this clinical trial is to evaluate whether voice or capnometry, alone or in combination with other (non invasive) biomarkers can be used to detect emphysema on chest CT-scan in people with chronic obstructive pulmonary disease (COPD). The main question it aims to answer is: • Can a machine-learning based algorithm be developed that can classify the extent of emphysema on chest CT scan from patients with COPD, based on voice and/or capnometry. Participants will: * perform different voice-related tasks * perform capnometry twice (before/after exercise) * perform a light exercise task between tasks ( 5-sit-to-stand test) * undergo one venipuncture
This is a cross sectional, single center study. At the clinic, patients with COPD will be invited to perform several voice related tasks (paced reading, sustained vowels, cough, quiet breathing) and will be instructed to perform capnometry measurements. These measurements will be performed before and after a light exercise task (5-STS: 5-sit-to-stand test). Clinical characterisation of patients including pulmonary function tests (spirometry, body plethysmography, diffusion capacity) and CT scans have been performed in all patients as a part of routine workup in the COPD care pathway. Emphysema will be quantified as low attenuation areas with a density below -950 Hounsfield units (HU) using Syngovia (Siemens, Erlangen, Germany). The primary outcome will fit a simple machine learning classification model (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify logistic regression model for the outcome of emphysema (\>25% vs ≤ 25%) from speech features and capnometry. with explanatory variables of speech features. Similar classification methods with incremental models using capnography features will be explored. Prior to carrying out the above analyses, data has to be pre-processed, including merging data, quality control, handling of missing data and feature extraction.
Study Type
OBSERVATIONAL
Enrollment
200
Patients with COPD will perform several voice-related tasks and capnometry at rest. Thereafter a 5-STS will follow and the voice-related task/capnometry will be repeated
Patients with COPD will perform several voice-related tasks and capnometry at rest. Thereafter a 5-STS will follow and the voice-related task/capnometry will be repeated
Dept of Respiratory Medicine, Maastricht University Medical Centre
Maastricht, Limburg, Netherlands
RECRUITINGLaurentius Ziekenhuis
Roermond, Limburg, Netherlands
NOT_YET_RECRUITINGpercentage of participants having moderate to severe emphysema on a chest CT (defined as > 25%)
A baseline chest CT scan from each participant will be analysed using a lung parenchyma analysis software with automated 3-D quantification of emphysema. Emphysema will be defined as low attenuation areas with a density below -950 Hounsfield units. Patients will be either classified as having low emphysema (less or equal to 25% of emphysema on chest CT scan) or moderate to high emphysema (more than 25% of emphysema on chest CT scan)
Time frame: baseline
number of (non-linguistic) inhalations per syllable from sustained vowel
Participants will perform a sustained vowel (saying 'a') at rest and after light exercise from which several measurements can be obtained: Syllables per breath group, speaking rate, articulation rate, mean frequency, mean intensity, pitch variability, mean center of gravity, inhalations, non-linguistic inhalations, ratio voice/silence intervals. Based on previous research (Merkus J, 2020) HNR, shimer, vowel duration en number of (non-linguistic) inhalations per syllable were putative vocal biomarkers in COPD. First key determinant therefore is the number of (non-linguistic) inhalations per syllable during sustained vowel of each participant. This will be used as input for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
harmonics-to-noise-ratio from sustained vowel
Participants will perform a sustained vowel (saying 'a') at rest and after light exercise from which several measurements can be obtained: Syllables per breath group, speaking rate, articulation rate, mean frequency, mean intensity, pitch variability, mean center of gravity, inhalations, non-linguistic inhalations, ratio voice/silence intervals. Based on previous research (Merkus J, 2020) HNR, shimer, vowel duration en number of (non-linguistic) inhalations per syllable were putative vocal biomarkers in COPD. Second key determinant therefore is the harmonics-to-noise ratio during sustained vowel. This will be used as input for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
vowel duration from sustained vowel
Participants will perform a sustained vowel (saying 'a') at rest and after light exercise from which several measurements can be obtained: Syllables per breath group, speaking rate, articulation rate, mean frequency, mean intensity, pitch variability, mean center of gravity, inhalations, non-linguistic inhalations, ratio voice/silence intervals. Based on previous research (Merkus J, 2020) HNR, shimer, vowel duration en number of (non-linguistic) inhalations per syllable were putative vocal biomarkers in COPD. Third key determinant therefore is the vowel duration (in seconds) during sustained vowel. This will be used as input for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
shimmer from sustained vowel
Participants will perform a sustained vowel (saying 'a') at rest and after light exercise from which several measurements can be obtained: Syllables per breath group, speaking rate, articulation rate, mean frequency, mean intensity, pitch variability, mean center of gravity, inhalations, non-linguistic inhalations, ratio voice/silence intervals. Based on previous research (Merkus J, 2020) HNR, shimer, vowel duration en number of (non-linguistic) inhalations per syllable were putative vocal biomarkers in COPD. Fourth key determinant therefore is shimmer (in Hz) during sustained vowel. This will be used as input for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
end-tidal CO2 from capnography (ETCO2)
Participation perform a quiet breathing (tidal volume) at rest and after light exercise to measure CO2 during exhalation (capnogram) from which several parameters can be measured, of which end-tidal CO2 (etCO2), phase 2 slope and phase 3 slope are most distinctive for COPD phenotyping (Pereira 2016). First key determinant from capnography is therefore end-tidal CO2 (in mm Hg). This will be used as input for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
phase-2 slope from capnography (slp2)
Participation perform a quiet breathing (tidal volume) at rest and after light exercise to measure CO2 during exhalation (capnogram) from which several (more than 80) parameters can be measured, of which end-tidal CO2 (etCO2), phase 2 slope and phase 3 slope are most distinctive for COPD phenotyping (Pereira 2016). Second key determinant from capnography is therefore phase-2 slope (in mm Hg/L). This will be used as input for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
phase-2 slope from capnography (slp3)
Participation perform a quiet breathing (tidal volume) at rest and after light exercise to measure CO2 during exhalation (capnogram) from which several parameters can be measured, of which end-tidal CO2, phase 2 slope and phase 3 slope are most distinctive for COPD phenotyping (Pereira 2016). Third key determinant from capnography is therefore phase 3 slope (in mm Hg/L). This will be used as input for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
serum sRAGE
Serum soluble receptor for advanced glycation end-products (sRAGE) from peripheral blood will be determined in each participant. Serum sRAGE is considered a blood biomarker for emphysema (Klont 2022). Serum sRAGE levels (in ng/mL) from each participant will be used as input variable for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
ratio of residual volume to total lung capacity (RV/TLC) on body plethysmography
Emphysema can be measured using body plethysmography. Several variables can be measured with body plethysmography: total lung capacity (TLC), inspiratory capacity (IC), functional residual capacity (FRC), residual volume (RV), ratio of IC/TLC, ratio FRC/TLC and ratio RV/TLC. The ratio of RV/TLC might be the most sensitive measure for airtrapping as the first sign of emphysema and is therefore chosen as the key outcome measure of body plethysmograpy. RV/TLC ratio (expressed as Z-score) from each participant will be used as input variable for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
diffusion capacity of the lungs for carbon monoxide
Diffusion capacity of the lungs for carbon monoxide (DLCO) is a measure of the lungs ability to transfer gas from air to the blood stream and a decrease in DLCO is associated with the extent of emphysema in chest CT scans. DLCO (expressed a Z-score) in each participant will be measured and used as input variable for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
Time frame: baseline
forced expiratory volume in one second
Forced expiratory volume in one second (FEV1) is a measure of severity of the underlying COPD. postbronchodilator FEV1 (expressed a Z-score) in each participant will be measured via spirometry. according to ERS/ATS guidelines. FEV1 (Z-score) will be used as input variable for machine learning classification models (e.g. using logistic regression, support vector machines, random forests and/or decision tree) to classify emphysema on CT scan (\>25% vs ≤ 25%)
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.
Time frame: baseline