Abstract
In this cohort study involving 9,399 current and former smokers from the COPDGene study, we assessed the relationship between AI-quantified mucus plugs on chest CTs and all-cause mortality. Our results revealed a significant positive association, particularly for those with COPD GOLD stages 1–4, with hazard ratios of 1.18 for 1–2 mucus-obstructed bronchial segments and 1.27 for ≥3 obstructed segments. This corroborates previous visual mucus plug counting research and demonstrates the relevance of mucus plugs in COPD pathology and as a marker for risk assessment. Automated mucus plug quantification methods may provide an efficient tool for both clinical evaluations and research.
Introduction
Chronic obstructive pulmonary disease (COPD) affects millions worldwide, ranking as a leading cause of mortality.[1] Central to COPD pathology is mucociliary dysfunction, leading to mucus plugs that can occlude the airways.[2] Mucus plugs, detectable in CT scans of many COPD patients, are linked to several adverse outcomes, including impaired airflow, lower oxygen levels, and reduced exercise tolerance.[3] Furthermore, mucus plugs can persist for years without symptoms like cough or sputum production.[4] Finally, the presence of mucus plugs in medium- to large-sized airways has been associated with all-cause mortality in COPD (GOLD 1–4), through meticulous visual counting of the number of mucus-obstructed bronchial segments.[5] The investigation was carried out on a subset of the data from the Genetic Epidemiology of COPD (COPDGene) study.[6]
Our study employed an artificial intelligence (AI) based platform (LungQ) for automated mucus quantification on chest CT scans to explore its association with mortality in the full cohort of all Phase 1 COPDGene participants, across all COPD stages including GOLD 0 and PRISm. We hypothesized that automated quantification would confirm visual scoring findings and would provide enhanced detail and efficiency in assessing the prognostic significance of mucus plugs in COPD.
Methods
COPDGene is a multicenter, prospective study on COPD genetics and epidemiology. It enrolled non-Hispanic Black and White participants aged 45–80 with a significant smoking history (≥10 pack-years). Exclusion criteria and ethical considerations have been outlined in the original COPDGene protocol.[5] The study included 10198 (ex-)smokers, enrolled from November 2007 to April 2011, followed up at 5 and 10 years. Data collection involved questionnaires, spirometry, and standardized chest CT scans using <1mm slice protocols. Mortality, spirometry and demographic data were sourced from the COPDGene database.
For our study all 10198 Phase 1 ever-smoker participants were included in the analysis, meaning that COPD GOLD stage 1–4, GOLD 0, and PRISm were all included.[7] Additionally, CT scans of 107 never-smoker COPDGene controls were analysed for comparison.
Automatic mucus plug quantification was performed using the LungQ platform (Thirona, Nijmegen, The Netherlands). LungQ uses AI-based algorithms to segment the bronchial tree and identify each bronchopulmonary segment. Mucus plugs are detected throughout the lung and linked to their respective segments. The detection algorithm, trained on expert annotations, identifies full mucus obstructions with clear proximal and distal airways, providing both location and volumetric assessments. The segmentation combines seed-based and voxel-based methods, providing accurate detection and quantification of mucus plugs along the entire bronchi, including the peripheries. (Figure 1a)
Figure 1.
Mucus Plug Detection and GOLD Class Distribution
a. LungQ reconstruction of the bronchial tree with detected mucus plugs indicated in red and the corresponding transversal CT images. b. Distribution of participants’ GOLD class for segment scores 0, 1–2 and ≥3 mucus-obstructed segment categories. c. Scatterplot of total mucus plug counts and GOLD classes. NB. total plug count can be >2 also at 1–2 bronchial segments affected, as multiple plugs can be found in the same segment.
Participants were categorized by the number of mucus-obstructed standard bronchopulmonary segments: 0, 1–2, or ≥3. Emphysema percentage was based on the lung parenchyma with attenuation below −950 Hounsfield Units (−950HU%) and airway wall thickness by taking the square root of the wall area for a hypothetical airway with a 10-mm inner perimeter (Pi10).[8]
Cox proportional hazard regression assessed the relationship between mucus plug scores categories and mortality in three models as in the study by Diaz et al.[5] The first model adjusted for demographics, smoking history, FEV1, emphysema and Pi10. The second added coronary disease, chronic bronchitis, asthma, and annual exacerbations. The third model consisted of the first model plus the BODE index, the most validated COPD mortality prediction score.[9] Analyses, performed using R (4.3.2), considered p-values <.05 significant without adjustment for multiple testing.
Results
The final cohort for analysis consisted of 9399 ever-smoker participants, after exclusion of 799 participants due to missing CT scans (n=297), poor-quality CT scans (n=82), technical issues (n=360), or missing spirometry data (n=60). In total 4165 participants had COPD GOLD 1–4 and 5234 participants GOLD 0 and PRISm. Over a median follow-up of 3957 days there were 2633 (28.0%) deaths. Of all participants, 7200 (76.6%) participants had a score of 0, indicating no mucus-obstructed segments, 1535 (16.3%) had 1–2, and 664 (7.1%) had 3 or more mucus-obstructed segments. See Table 1 for participant characteristics and Figure 1b and 1c for mucus plug distribution across GOLD stages.
Table 1.
Participant characteristics.
| COPDGene Phase 1 cohort (n=9399) | Mucus plug score category (No. of segments w/mucus plugs) | ||
|---|---|---|---|
| Characteristics | 0 (n=7200) | 1–2 (n=1535) | ≥3 (n=664) |
| Age, median (IQR), y | 57.7 (51.3-65.1) | 62.3 (54.5-69.0) | 64.4 (57.1-70.7) |
| Sex | |||
| Female | 47.6% | 44.7% | 39.5% |
| Male | 52.4% | 55.3% | 60.5% |
| Race and ethnicity | |||
| Non-Hispanic White | 4733 (64.3%) | 1173 (75.8%) | 540 (81.2%) |
| Non-Hispanic African American | 2618 (35.7%) | 376 (24.2%) | 125 (18.8%) |
| BMI, median (IQR) | 28.3 (24.7-32.6) | 27.1 (23.5-31.5) | 25.5 (22.3-29.4) |
| Current smoker, No. (%) [No.] | 3901 (54.2%) | 747 (48.7%) | 291 (43.8%) |
| Pack-years of smoking, median (IQR) | 37.6 (25.9-51.7) | 44.0 (32.5-63.3) | 47.4 (34.7-68.3) |
| Medical history | |||
| Chronic bronchitis | 1143 (15.9%) | 414 (27.0%) | 230 (34.6%) |
| Coronary artery disease | 754 (10.5%) | 197 (12.8%) | 85 (12.8%) |
| Asthma | 760 (11.4%) | 247 (16.1%) | 165 (24.8%) |
| Exacerbations / year (mean, SD) | 0.28 (0.79) | 0.58 (1.09) | 1.02 (1.50) |
| COPD GOLD stage of severity | |||
| PRISm | 971 (13.5%) | 145 (9.4%) | 23 (3.5%) |
| 0 (≥10 packyears with FEV1/FVC>0.7) | 3732 (51.8%) | 320 (20.8%) | 43 (6.5%) |
| 1 (Mild) | 613 (8.5%) | 115 (7.5%) | 21 (3.2%) |
| 2 (Moderate) | 1265 (17.6%) | 396 (25.8%) | 143 (21.5%) |
| 3 (Severe) | 478 (6.6%) | 359 (23.4%) | 237 (35.7%) |
| 4 (Very severe) | 141 (2.0%) | 200 (13.0%) | 197 (29.7%) |
| BODE index, median (IQR) | 0 (0.0–2.0) | 2 (0.0–4.0) | 4 (2.0–5.0) |
| FEV1, L, median (IQR) | 2.41 (1.85-3.01) | 1.69 (1.09-2.41) | 1.16 (0.77-1.71) |
| FEV1, % predicted, median (IQR) | 85.3 (70.6-97.6) | 62.0 (39.9-82.8) | 40.5 (27.0-59.2) |
| Emphysema on CT, median (IQR), % | 1.54 (0.45-4.85) | 4.52 (1.05-16.7) | 11.33 (2.90-24.15) |
| Airway wall thickness, median (IQR), mm | 2.14 (1.84-2.53) | 2.61 (2.19-3.06) | 3.02 (2.58-3.46) |
|
| |||
| Total number of plugs | |||
| Median (IQR), no. | 0.0 | 1.0 (1.0–2.0) | 7.0 (5.0–12.0) |
| Total plug volume (mm3) | |||
| Median (IQR) | 0 (0–0) | 39.8 (14.6–93.7) | 318.1 (164.0-742.4) |
|
| |||
| Never-smokers (control group, n=107) | 101 (94.4%) | 5 (4.7%) | 1 (0.9%) |
In the adjusted model, automated mucus plug score categories were significantly associated with all-cause mortality. Hazard ratios (HR) were 1.14 (CI: 1.03–1.26) for 1–2 mucus-obstructed segments and 1.24 (CI: 1.09–1.42) for 3 or more. In the second model adjusted for additional confounders HR 1.13 (CI: 1.02–1.25) and HR 1.21 (CI: 1.06–1.38). In the third model adjusting for the BODE index, hazard ratios were 1.10 (CI: 0.995–1.22) and 1.15 (CI: 1.0–1.31), respectively. See Table 2.
Table 2.
Association Between Mucus Plug Score and All-Cause Mortality.
| COPDGene Phase 1 cohort (n=9399) | Mucus plug score (No. of segments w/ mucus plugs) | |||||
|---|---|---|---|---|---|---|
| No. | 0 (n =7200) | 1–2 (n =1535) | ≥3 (n= 664) | |||
| Deceased, n (%) | 1625 (22.6%) | 638 (41.6%) | 371 (55.9%) | |||
| HR (95% CI) | HR (95% CI) | p-value | HR (95% CI) | p-value | ||
| Adjusted model* | 9397 | Reference | 1.14 (1.03-1.26) | 0.010 | 1.24 (1.09-1.42) | 0.001 |
| Adjusted model plus coronary artery disease, chronic bronchitis, current asthma and exacerbations per year |
9397 | Reference | 1.13 (1.02-1.25) | 0.016 | 1.21 (1.06-1.38) | 0.006 |
| Adjusted model plus BODE index† | 9272 | Reference | 1.10 (0.995-1.221) | 0.062 | 1.15 (1.001-1.312) | 0.048 |
| Participants with GOLD stage 1–4 | Mucus plug score (No. of segments w/ mucus plugs) | |||||
| No. | 0 (n= 2497) | 1–2 (n= 1070) | ≥3 (n= 598) | |||
| Deceased, n (%) | 868 (34.8%) | 552 (51.6%) | 355 (59.4%) | |||
| HR (95% CI) | HR (95% CI) | p-value | HR (95% CI) | p-value | ||
| Adjusted model* | 4163 | Reference | 1.18 (1.05-1.32) | 0.005 | 1.27 (1.10-1.46) | 0.001 |
| Adjusted model plus coronary artery disease, chronic bronchitis, current asthma and exacerbations per year |
4163 | Reference | 1.17 (1.04-1.31) | 0.008 | 1.24 (1.07-1.43) | 0.004 |
| Adjusted model plus BODE index† | 4068 | Reference | 1.14 (1.02-1.28) | 0.027 | 1.22 (1.05-1.41) | 0.008 |
| Particpants with GOLD 0 and PRISm | Mucus plug score (No. of segments w/ mucus plugs) | |||||
| No. | 0 (n=4703) | 1–2 (n=465) | ≥3 (n=66) | |||
| Deceased, n (%) | 757 (16.1%) | 86 (18.5%) | 16 (24.2%) | |||
| Adjusted model* | 5234 | Reference | 1.05 (0.83-1.31) | 0.692 | 1.31 (0.79-2.17) | 0.303 |
| Adjusted model plus coronary artery disease, chronic bronchitis, current asthma and exacerbations per year | 5234 | Reference | 1.04 (0.83-1.30) | 0.754 | 1.26 (0.76-2.11) | 0.372 |
| Adjusted model plus BODE index† | 5204 | Reference | 1.04 (0.83-1.30) | 0.738 | 1.24 (0.75-2.06) | 0.412 |
Cox proportional hazard regression models.
Adjusted for: age, gender, race, BMI, smoking status, packyears, FEV1%, emphysema (−950HU%), airway wall thickness (Pi10), scanner model;
Including all variables from the adjusted model except FEV1 and BMI plus adjustment for the BODE index (continuous). Proportional hazard assumptions were evaluated using Schoenfeld residuals.
Among 4165 participants with COPD (GOLD stages 1–4), hazard ratios were 1.18 (CI: 1.05–1.32) for 1 or 2 obstructed segments and 1.27 (CI: 1.10–1.46) for 3 or more obstructed segments. In the second model 1.17 (CI: 1.04–1.31) and 1.24 (CI: 1.07–1.43). In the third model 1.14 (CI: 1.02–1.28) and 1.22 (CI: 1.05–1.41), respectively. In COPD stages (GOLD 0 and PRISm), mucus plug scores were lower and showed no significant association with mortality.
In the control group of 107 never-smokers, LungQ detected mucus plugs in 6 participants (5.6%), with 4 participants with one plug, 1 participant with three plugs in two segments, and 1 subject with four plugs in four segments.
Discussion
This study confirms automated mucus plug analysis in the COPDGene cohort associations with higher mortality, consistent with visual scoring methods, even when adjusting for confounders and the BODE-index. Hazard ratios for COPD stages 1–4 subgroup of 1.18 and 1.27, for 1–2 and ≥3 obstructed segments, respectively, were highly similar to those found by visual methods. No mortality association was found in non-COPD subgroups (GOLD 0 and PRISm).
LungQ detected similar mucus plug counts to Diaz et al., with 25.7% (1070/4165) of participants showing 1–2 obstructed segments and 14.4% (598/4165) with ≥3, reflecting both method’s identification of mucus plugs in 40% of GOLD 1–4 participants. The variation in the standard bronchopulmonary segment anatomy and the algorithm’s conservative design, emphasizing specificity, may account for differences. The prevalence of mucus plugs in GOLD 1–4 is substantially higher than what was observed in the PRISM (14%) and GOLD 0 participants (10%). Furthermore, also substantially higher than what can be observed in never-smokers.
The study’s strengths include the inclusion of all COPDGene Phase 1 participants including 5234 participants classified as GOLD 0 and PRISm as control groups, along with 107 never-smokers. Additionally, the use of automated analysis provides consistency compared to visual methods. However, one limitation is the absence of direct validation of the automated measurement against visual scoring. Due to the exceptionally laborious nature of visual mucus plug counting, large-cohort comparison is challenging. Therefore, we have focussed on evaluating the association of automated mucus plug score categories with mortality. The observational design is another limitation as it limits causal conclusions.
The three mucus score categories (0, 1–2, and ≥3 obstructed segments) were chosen for their demonstrated stratification of risk.[5] Continuous variables of total mucus plug numbers and total mucus volume are highly left-skewed, limiting their utility for the current models. The distribution of mucus across segments and quantifying the total number and volume of mucus plugs may add further relevant information, particularly for longitudinal analysis at the individual level. These aspects warrant further exploration in future studies.
Overall, the results across various COPD severities confirm the association of mucus plugs with mortality, likely mediated through inflammation, infection, and ventilation/perfusion mismatch. This underscores their potential as markers of disease severity and may guide treatment interventions. Automated analysis of mucus plugs could identify high-risk subgroups by integrating patient data and mucus metrics, opening new opportunities for personalized risk assessment, research, and therapeutic strategies in COPD.
Acknowledgments:
COPDGene study participants and teams.
Disclose sources of financial support for the study:
This work was supported by NHLBI grants U01 HL089897 and U01 HL089856 and by NIH contract 75N92023D00011. The COPDGene study (NCT00608764) has also been supported by the COPD Foundation through contributions made to an Industry Advisory Committee that has included AstraZeneca, Bayer Pharmaceuticals, Boehringer-Ingelheim, Genentech, GlaxoSmithKline, Novartis, Pfizer, and Sunovion.
G.J. Braunstahl (GB): honoraria for lectures and consultancy from GSK, AstraZeneca, Novartis and Sanofi Genzyme, as well as research grants from Sanofi Genzyme, GSK and AstraZeneca, not related to the work under consideration.
H.A.W.M. Tiddens (HT): received in the last three years multiple grants from the following public and institutional grant institutions for lung structure and function research: NHMRC, NIH, CFF, ECFS, IMI, and ErasmusMC Sophia Foundation; received unconditional grants for investigator-initiated research from Novartis and Insmed. Acted as consultant for Insmed, Thirona, Neupharma and Boehringer Ingelheim. Chief medical officer for Thirona since April 2022. Vice-chair and faculty for the ADVANCE course sponsored by Vertex.
Footnotes
Conflicts of interest:
T. van der Veer (TV): no conflicts of interest involving the work under consideration for publication, no relevant financial activities outside the submitted work, no other relationships or activities that readers could perceive to have influenced, or that give the appearance of potentially influencing the current manuscript.
E.R. Andrinopoulou (EA): no conflicts of interest, no relevant financial activities or other relevant relationships or activities regarding the work under consideration.
J. Charbonnier (JC): Employee and shareholder at Thirona
V. Kim (VK): no conflicts of interest, no relevant financial activities or other relevant relationships or activities regarding the work under consideration.
R. Latisenko (RL): Senior Deep Learning Engineer at Thirona
D. A. Lynch (DL): no conflicts of interest, no relevant financial activities or other relevant relationships or activities regarding the work under consideration.
References
- 1.Collaborators, G.B.D.C.R.D., Global burden of chronic respiratory diseases and risk factors, 1990–2019: an update from the Global Burden of Disease Study 2019. EClinicalMedicine, 2023. 59: p. 101936. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Boucher RC, Muco-Obstructive Lung Diseases. N Engl J Med, 2019. 380(20): p. 1941–1953. [DOI] [PubMed] [Google Scholar]
- 3.Dunican EM, et al. , Mucus Plugs and Emphysema in the Pathophysiology of Airflow Obstruction and Hypoxemia in Smokers. Am J Respir Crit Care Med, 2021. 203(8): p. 957–968. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Okajima Y, et al. , Luminal Plugging on Chest CT Scan: Association With Lung Function, Quality of Life, and COPD Clinical Phenotypes. Chest, 2020. 158(1): p. 121–130. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Diaz AA, et al. , Airway-Occluding Mucus Plugs and Mortality in Patients With Chronic Obstructive Pulmonary Disease. JAMA, 2023. 329(21): p. 1832–1839. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Regan EA, et al. , Genetic epidemiology of COPD (COPDGene) study design. COPD, 2010. 7(1): p. 32–43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Agusti A, et al. , Global Initiative for Chronic Obstructive Lung Disease 2023 Report: GOLD Executive Summary. Am J Respir Crit Care Med, 2023. 207(7): p. 819–837. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Grydeland TB, et al. , Quantitative computed tomography: emphysema and airway wall thickness by sex, age and smoking. Eur Respir J, 2009. 34(4): p. 858–65. [DOI] [PubMed] [Google Scholar]
- 9.Celli BR, et al. , The body-mass index, airflow obstruction, dyspnea, and exercise capacity index in chronic obstructive pulmonary disease. N Engl J Med, 2004. 350(10): p. 1005–12. [DOI] [PubMed] [Google Scholar]