Study population
We employed 311 subjects with cement dust effects collected from Kangwon National University Hospital (KNUH) supported by a Korean research project called the Chronic Obstructive pulmonary disease in Dusty Areas near cement plants (CODA) cohort over approximately 10 years (3, 14, 15). This project was designed to investigate the effect of cement-dust exposure on patients’ health near cement plants located in the Kangwon and Chungbuk provinces of South Korea considering the distance of cement plants and wind direction based on meteorological data from the National Institute of Environmental Research of the Ministry. The size of cement dust varies between 0.5 and 5 µm (16). As control data, we employed 298 subjects with none or little exposure to cement dust collected from Chonbuk National University Hospital (CNUH) over 3 years (17). The control subjects had normal findings on CT imaging, such as an absence of lung lesions or air-trapping, and no known history of lung disease or surgery. Both the KNUH and CNUH studies were approved by the Institutional Review Board at individual sites (KNUH 2019-06-007 and CUH 2016-03-020-005) and used a similar imaging protocol (Table 1).
A flow chart for the subject selection procedure is provided in Fig 1. To select subjects with normal lung function, we only included subjects with FEV1/forced vital capacity (FVC)≥70% and FVC %predicted value≥80%. In addition, we excluded subjects with any prior diagnosis of asthma and/or pneumonia. This exclusion procedure allows for an objective comparison by eliminating confounding effects caused by pulmonary diseases such as asthma, pneumonia, and COPD. Then propensity score matching (PSM) method was applied for 66 DE and 274 NDE subjects, to reduce the confounding effects of age, sex, height, and smoking history. See the subsection Statistical analysis for the PSM method. In this study, pulmonary function tests (PFTs) of DE and NDE subjects were performed according to the American Thoracic Society/European Respiratory Society guideline (18).
QCT-based airway structure and lung function
In both TLC and FRC scans, we derived the luminal hydraulic diameter (Dh, TLC and Dh, FRC), airway wall thickness (WTTLC and WTFRC), and bifurcation angle (θTLC and θFRC) using Apollo software 2.0 (VIDA Diagnostics, Coralville, Iowa, USA), along with an in-house post-process. The bifurcation angle was defined as an angle between two daughter branches of a proximal airway. Dh, WT, and θ could be used to assess airway narrowing, wall thickening, and the alteration of branching structure, respectively.
To assess the deformable features of segmental airways, the deformation ratio ε between TLC to FRC was computed as follows:
where φ is any structural variable of Dh, WT, and θ, so εDh, εWT, and εθ were derived in this study. To measure the regional features of airways, structural variables were extracted from seven central airways and five subgroup regions. A detailed description of airway labeling is given in Fig 2. The seven central airways included the trachea, right main bronchus (RMB), bronchus intermedius (BronInt), trifurcation of the right lower lobe (TriRUL), main bronchus (LMB), trifurcation of the left upper lobe (TriLUL), and LLB6. The five subgroup lobes included the right upper lobe (sRUL), right middle lobe (sRML), right lower lobe (sRLL), left upper lobe (sLUL), and left lower lobe (sLLL).
Functional variables included the air volumes at TLC and FRC, inspiratory capacity (IC), and Jacobian between TLC and FRC, respectively. In addition, Emph% and fSAD% were computed using an image registration technique (19). To minimize the inter-center variability of Emph% and fSAD%, a fraction-threshold (10) method was used. Detailed formulations of these imaging variables are included in the references (10, 11, 20).
Statistical analysis
We used the PSM method to reduce the bias between two groups of DE and NDE subjects. A propensity score is the predicted probability of belonging to the treatment group, and would be calculated for each subject in the study (13). The propensity scores were estimated by multiple logistic regression analysis using the age and height variables stratified sex and smoking status. Matching was done using the Greedy 1-4 matching within a caliper, 0.2 times standard error of propensity scores. To validate PSM we used standardized difference, defined balance as an absolute value less than 10 (21). PSM method was conducted using SAS 9.4 software.
One sample t-tests were performed for the subjects matched by PSM method to compare the difference of QCT imaging-based metrics. The values are represented by means and the standard deviation (SD) in Tables 2–4 and means and the confidence interval (CI) in Fig 3–6. A significance level of p=0.05 was chosen for a total of 130 comparison tests, resulting in a false discovery rate of 8.6%. Statistical analyses were conducted using R software.