A Simple Index for Predicting Mucus Plugs in Patients with COPD

To the Editor:

Airway-occluding mucus plugs (MPs) identified on computed tomography (CT) are a frequent and modifiable pathologic manifestation in people with COPD.^1,2 The phenotype has been used as an imaging endpoint in asthma trials, supporting its use in COPD.^3,4 However, screening patients with a low likelihood of having MPs can lead to unnecessary radiation exposure and increased study costs. Developing an index to identify patients with a higher likelihood of airway mucus plugs on CT could provide a more efficient screening strategy. The goal of this study was to develop such an index, using data from the COPDGene Study and Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints (ECLIPSE) cohorts, as development and validation datasets.^5,6

We analyzed current and former tobacco-exposed participants (history of ≥10 pack years smoked) with Global Initiative for Obstructive Lung Disease (GOLD) grades 1–4 COPD, defined as FEV1/FVC <0.7.⁷ Of 4480 participants aged 45–80 years with GOLD 1–4 COPD, those with missing CT-based MP scores and a history of asthma were eliminated, leaving 3397 for this analysis. Out of 2163 ECLIPSE COPD participants aged 40–75 years, 1215 had GOLD 2–4 COPD without asthma history and complete data on MP scores. Trained readers identified and scored mucus plugs, as previously reported.¹ The score consists of the number of lung segments with mucus plugs, ranging from 0 to 18. The score was dichotomized for this analysis as 0–2 vs. ≥3. The cut-off was upon a prior study showing the increased mortality with a score of ≥3.¹ To develop an index to predict MPs on CT, the initial selection started from a pool of 12 variables based on prior literature, including 1) age (<65, 65+ years), 2) sex (female, male), 3) body mass index (BMI, kg/m², <20, 20–24.9, 25–34.9, >35 [control]), 4) smoking status (former vs. current), 5) pack-years smoked, 6) cough, 7) phlegm, 8) dyspnea (defined as Medical Research Council Dyspnea Scale ≥1), 9) history of exacerbations in the year before enrollment (0 vs. ≥1), 10) chronic bronchitis (yes/no), 11) 1–4 GOLD grade, and 12) FEV1/FVC (≥.6, .5 - <.6, .4- <.5, <.4).^1,2,7–11 We used mixed-effect models with above categorical factors as fixed effects and clinical site as a random effect. We applied several variable selection methods and compared various models, and then selected the best model based on the Akaike Information Criterion and Bayesian Information Criterion.¹² The model was developed in COPDGene and then tested in ECLIPSE. Out of 12 variables, we identified eleven factors that predicted MPs on CT scans. We multiply the coefficients by 20 according to their variable category, round them to integers, and sum them to create an index, ranging from 0 to 106. The same procedure was repeated with continuous variables (i.e., age, BMI, FEV1 % predicted, and FEV1/FVC ratio), yielding similar results. Results are presented only with categorical variables for an easy-to-use and interpret index. The analysis was conducted by a statistician (WW) using the SAS software 9.4 (SAS Institute, Cary, NC).

The characteristics of the participants by cohort are shown in Table 1. Of 3397 COPDGene and 1215 ECLIPSE participants with COPD, 547 (16.1%) and 234 (19.3%), respectively, presented with MP scores ≥3. The final model was built with eleven factors directly related to MPs. The development and testing index’s areas under the receiver operating characteristic curve were 0.75 and 0.73, respectively. The more influential index factors associated with MPs on CT scan were GOLD grade, BMI, and age; intermediate were sex, phlegm, FEV1/FVC, and current smoking; and the lowest were dyspnea, cough, prior exacerbation, and chronic bronchitis (Figure 1, upper panel). These eleven attributes enabled the development of the Simple Index for predicting Mucus Plugs (SIMP), with 5% higher odds of having MPs on CT per each additional SIMP point (odds ratio, 1.05 [95% confidence interval, 1.05–1.06]) (Figure 1, lower panels).

Table 1.

Participants’ characteristics from the COPDGene and ECLIPSE cohorts.^a

Characteristic	COPDGene (N = 3397) N (%)	ECLIPSE (N = 1215) N (%)
Age, years
<65	1848 (54.4)	634 (52.2)
≥65	1549 (45.6)	581 (47.8)
Sex
Female	1398 (41.2)	405 (33.3)
Male	1999 (58.8)	810 (66.7)
Raceb
Non-Hispanic Black	667 (19.6)	18 (1.5)
Non-Hispanic White	2730 (80.4)	1197 (98.5)
BMI, kg/m2
<20	204 (6.0)	132 (10.9)
20–24.9	1018 (30.0)	378 (31.1)
25–34.9	1803 (53.1)	633 (52.1)
≥35	372 (11.0)	72 (5.9)
Smoking status
Former	1896 (55.8)	767 (63.1)
Current	1501 (44.2)	448 (36.9)
Pack-years smoked, median (Q25 -Q75)	46.6 (35.3–66.0)	45.0 (32.0–60.0)
Symptoms
Dyspneac	2354 (69.5)	1027 (86.2)
Cough	1405 (41.4)	585 (50.3)
Phlegm	1510 (44.5)	711 (60.9)
Chronic bronchitisd	826 (24.3)	397 (32.7)
Prior exacerbatione	998 (29.4)	515 (42.4)
GOLD gradef
1 (mild)	668 (19.7)	-
2 (moderate)	1459 (42.9)	540 (44.4)
3 (severe)	845 (24.9)	521 (42.9)
4 (very severe)	425 (12.5)	154 (12.7)
FEV1/FVC
≥0.6	1402 (41.3)	150 (12.4)
0.5 — <0.6	688 (20.3)	248 (20.4)
0.4 — <0.5	596 (17.5)	371 (30.6)
<0.4	711 (20.9)	444 (36.6)
Mucus plug scoreg ≥3	547 (16.1)	234 (19.3)

Data are presented as frequency (proportion, %) and as median (Q25 - Q75)

^aCOPDGene and ECLIPSE were used as development and testing datasets.

^bRace was self-reported by the participant. Other races and Hispanics were not included in COPDGene.

^cDyspnea defined as a modified medical Research Council Dyspnea Scale ≥1

^dChronic bronchitis is defined as cough and phlegm for at least 3 months per year for at least 2 consecutive years.

^eExacerbation is a new onset or increase in cough, phlegm, or dyspnea requiring antibiotic or systemic steroids prescription. Participants were asked about the number of exacerbation episodes the year before study entry. The number of episodes was dichotomized as 0 vs ≥1.

^fGOLD grades of COPD were defined with postbronchodilator FEV1 percentage of predicted (pp) values as follows: 1 (mild), FEV1 pp ≥80; 2 (moderate), FEV1 pp 50 to <80; 3 (severe) FEV1 pp 30 to <50; and 4 (very severe), FEV1 pp <30.

^gMucus plug score is defined as the number of lung segments with mucus plugs, ranging from 0 to 18, and dichotomized as 0–2 vs. ≥3

COPDGene participants with missing data: dyspnea, 8.

ECLIPSE participants with missing data: dyspnea, 24; cough, 51; phlegm, 48; FEV1/FVC, 2.

Abbreviations: ECLIPSE, Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints; GOLD, Global initiative for obstructive lung disease; COPD, chronic obstructive pulmonary disease; FEV1, forced expiratory volume in one second; FVC, forced vital capacity; Q, quartile

Figure 1.

Weighted factors (upper panel) and associated probabilities of the simplex index for predicting mucus plugs (SIMP) (lower panel). (Upper panel) Factors of SIMP are shown as categories. Numbers inside of each bar represent category variables and corresponding weighted points based on the coefficients from a multivariable model. The total score ranges from 0 to 106 points. A score of 106 points represents the sum of the categories for each variable with the highest weighted points. (Lower panel) Estimated probabilities (blue line) of mucus plugs as a function of SIMP points based on a univariate logistic regression model. Odds ratios of mucus plugs for each additional SIMP point (upper left corner). Green and red lines represent the probability of mucus plug on CT for a patient with SIMP of 22 and 95 points, respectively (see text for details). 215×279mm (300 × 300 DPI)

We developed a simple tool — the SIMP — for predicting MPs on CT scans using a well-characterized COPD cohort. The devised SIMP is easy to implement because it is based on readily available demographic, clinical, and spirometric attributes. Most of these attributes were found to be associated with mucus plugs in prior studies and are typical characteristics assessed when screening study participants.^1,2,8–10 A factor strongly associated with MPs was disease severity as measured with GOLD grades, suggesting a greater susceptibility to form plugs in severe than mild disease. In addition to GOLD grades, which are based on FEV1% predicted values, we used FEV1/FVC with cutoff values recently proposed to assess COPD severity.¹¹ The strength of the association between FEV1/FVC ratio and MPs followed those of GOLD grade, BMI, and age. COPD patients with MPs consistently present with lower BMI, supporting our finding; however, further investigation is needed to elucidate the underlying mechanisms of such an association. Participants aged 65 years and older, as well as females, had an elevated risk of MPs on CT. Reduced ability to clear secretions from the airways and hormonal changes may partially explain these associations. Additionally, respiratory symptoms, chronic bronchitis and exacerbations were associated with mucus plugs, all which might be considered clinical manifestations of mucus pathology. ^8,13

The SIMP combines eleven variables, with each assigning a specific number of points, into a simple scale. For example, a 71-year-old (10 points) male (0), BMI of 38 (0 point), current. smoker (6), with GOLD 3 COPD (26), FEV1/FVC of 0.60 (0), dyspnea (3), cough (5), phlegm (9), history of chronic bronchitis (3), and exacerbations in the past year (0), totals 62 points, corresponding to a 22% probability of having mucus plugs on CT. In contrast, the probability increases to 61% for a 70-year-old (10) female (9), BMI of 22.8 (16), current smoker (6), with GOLD 3 COPD (26), FEV1/FVC of 0.44 (8), dyspnea (3), cough (5), and phlegm (9), a history of chronic bronchitis (3), and no exacerbations in the past year (0), totaling 95 points. The equation to compute the probabilities of mucus plugs using the total points is provided in the upper panel of Figure 1.

Study limitations include heavy smokers only, two racial groups, and a time-consuming lung segment-based scoring of mucus plugs. The score may underestimate the mucus burden as it does not include partially occluded airways. Additionally, categorizing continuous variables for prediction models is not typically recommended by guidelines. Finally, lack of eosinophil data prevented from evaluating its potential role in predicting mucus plugs, an area for further research.

Based on demographic, clinical, and spirometric attributes, SIMP can predict mucus plugging on CT scans in people with COPD. This tool could inform the recruitment of participants to clinical studies using mucus plugs as an endpoint.