Patients with COPD from the Korean Obstructive Lung Disease study cohort were enrolled from the pulmonary clinics of 16 hospitals in South Korea, from June 2005 to April 2018. Patients with a smoking history of more than 10 pack-years were spirometrically diagnosed with COPD, as were those with a postbronchodilator FEV₁/FVC ratio < 0.7. Patients with baseline and 3-year follow-up RV and TLC were included in this study. Pulmonary function tests (PFTs) and CT scans were performed on the same day. Written informed consent was obtained from each patient prior to the evaluation. The names of the ethics committees are listed in the online supplement.

The Institutional Review Boards of the participating hospitals approved the use of the clinical and imaging data. The Ethics Committees were as follows: Asan Medical Center, Bundang CHA Hospital, Ewha Womans University Mokdong Hospital, Korea University Anam Hospital, Hanyang University Guri Hospital, Inje University Ilsan Paik Hospital, Kangbuk Samsung Hospital, Hallym University Kangnam Sacred Heart Hospital, Kangwon National University Hospital, Seoul National University Hospital, Seoul National University Bundang Hospital, Ajou University Hospital, Konkuk University Hospital, Yeouido St. Mary’s Hospital, Seoul St. Mary’s Hospital (KC11OIME0668), and the National Medical Center.

All patients underwent volumetric CT scans at full inspiration and expiration. The scan parameters included 100 effective mAs, 0.75-mm collimation, and 140 kVp with a pitch of 1.0. The CT scans were done craniocaudally with the patient in the supine position.

The CT attenuation ranged from −1,024 to 3,072 HU, and all CT scans were non-enhanced. The images were reconstructed from the thoracic inlet to the lung base using the soft kernel (B30f, Siemens Medical Systems, Malverm, PA, USA). Using in-house software, whole-lung images were extracted automatically and the pixel attenuation coefficient was measured.

The MLD and volume fraction of the lung were calculated automatically using software. The ATI was defined as the E/I MLD. Airway dimensions were measured at the site near the origin of four segmental bronchi (RB1, LB1 + 2, LB10, and RB10).

The airway was measured more accurately using a modified sharpening filter with a 3 × 3 × 3 kernel size. The percentage of bronchial wall area (WA) was defined as WA/(WA + lumen area) × 100, and was obtained in each segmental bronchus. The mean values of airway dimensions were used in the statistical analysis. Validation was done using polyacryl tubes. The software automatically discriminated the inner and outer boundaries of the airway wall and the airway lumen by the full-width-maximum method [16,17].

Data including the COPD Assessment Test (CAT) score, St. George’s Respiratory Questionnaire (SGRQ) score, body mass index (BMI), age, and sex of the patients were assessed at baseline and follow-up. The total number of COPD exacerbations during the observation period was evaluated. FEV₁, FVC, FEV₁/FVC ratio, RV, and TLC were evaluated at the baseline and follow-up PFTs. DLCO and exercise capacity on the 6MWT were also evaluated.

The study patients were grouped into four phenotypes according to the pattern of change in RV/TLC over 3 years. The RV/TLC was considered abnormal when it was ≥ 40% and normal when < 40% [9]. When the RV/TLC was < 40% at baseline and remained < 40% after 3 years, it was classified as “normal to normal” (N→N), but if it increased to ≥ 40% after 3 years it was classified as “normal to abnormal” (N→A). When the baseline RV/TLC was ≥ 40% and remained at ≥ 40% after 3 years, it was classified as “abnormal to abnormal” (A→A), but when it changed to < 40%, it was classified as “abnormal to normal” (A→N).

All statistical analyses were performed using the IBM SPSS version 20.0 (IBM Corp., Armonk, NY, USA). Continuous data are presented as mean with range. The chi-square test was used to analyze categorical parameters. Continuous variables were analyzed using one-way analysis of variance or the Kruskal-Wallis test, depending on the normality of the data distribution. A negative binomial regression analysis was used to evaluate the association between moderate to severe exacerbation frequency of COPD and other clinical parameters. Factors significant in the univariate analysis were entered into a multivariable analysis. Overall survival (OS) was the time between enrollment and observation, which took place on April 26, 2018. The log-rank test was used to compare OS between the COPD phenotypes. A p < 0.05 was considered significant; however, for multiple comparisons, alpha level adjustment was performed.

Among the 675 patients with COPD enrolled in this study, 52 without baseline RV/TLC data and an additional 344 without follow-up RV/TLC data for year 3 were excluded (Supplementary Fig. 1). A total of 279 patients were evaluated. The baseline RV/TLC showed a significant negative linear correlation with the baseline FEV₁ (%; r = −0.649, p < 0.001), FVC (%; r = −0.451, p < 0.001), FEV₁/FVC ratio (r = −0.351, p < 0.001), DLCO (absolute value; r = −0.242, p < 0.001), and BMI (r = −0.133, p = 0.028) (Fig. 1). On the other hand, the baseline RV/TLC showed a positive linear correlation with the baseline total SGRQ score (r = 0.457, p < 0.001), total CAT score (r = 0.324, p = 0.003), E/I MLD (r = 0.354, p < 0.001), and mean bronchial WA (%; r = 0.236, p < 0.001) (Fig. 1). According to the pattern of change in the RV/TLC, patients were categorized into four groups: 76 in “N→N,” 34 in “A→N,” 33 in “N→A,” and 136 in “A→A.” Table 1 shows the clinical characteristics of the four groups. The annual rate of RV/TLC (%) decline varied among the study patients (−0.18 ± 3.36%/yr, mean ± standard deviation) (Supplementary Fig. 2). The mean age was highest in the A→A group (67.5 years, p < 0.001). No significant group differences were seen in sex or the proportion of ever-smokers. Prevalence of comorbidities were also compared between the groups: statistically significant difference was observed only in pulmonary tuberculosis (p = 0.041). In terms of inhaler medications during the observation period, statistically significant difference was observed in uses of long-acting beta-agonist (LABA), long-acting muscarinic antagonist (LAMA), inhaled corticosteroid (ICS)/LABA, and LABA/LAMA (p = 0.009, p = 0.044, p = 0.002, and p = 0.002, respectively). Comparisons in moderate to severe exacerbation frequency in follow-up years after the observation time showed no significant difference between the groups. Three-year changes in the FEV₁, FVC, and FEV₁/FVC differed significantly among the four groups. For FEV₁ and FVC, respectively, group A→N showed a large increase of 266 mL (p < 0.001) and 381 mL (p < 0.001), group N→A showed a marked decrease of 216 mL (p < 0.001) and 332 mL (p = 0.029), and group A→A showed a decrease of 16 mL (p = 0.426) and 6 mL (p = 0.011) compared to group N→N. No significant difference between the subgroups was shown in the change in DLCO.

Group A→N showed a significant within-group increase in FEV₁, while group N→A showed a significant decrease in both % predicted and the absolute values. The absolute value of FVC changed significantly in all groups except A→A. While groups N→N and N→A showed significant decreases in the absolute value of FVC, group A→N showed a significant increase. Groups N→A and A→N showed significant changes in FEV₁/FVC (%): group A→N showed an increase of 2.8%, while group N→A showed a decrease of −2.3% (Table 1).

All subgroups showed significant within-group decreases in the absolute value of DLCO, while only groups N→N and A→A showed significant decreases in % predicted DLCO. A significant change was detected in the BMI only in group A→A. The mean BMI decreased significantly, by −0.33, in group A→A.

Table 2 shows the baseline, follow-up, and delta (difference between year 3 and baseline) values of the four groups stratified by 3-year changes in the RV/TLC. Group A→A showed the highest E/I MLD value in both the baseline and follow-up measurements. Groups A→N and A→A showed the highest baseline EI values. The mean baseline bronchial WA was highest in groups A→N and A→A (67.2% and 67.6%, respectively; p = 0.002). Group A→A had the highest baseline inspiratory and expiratory lung volumes among the four subgroups (p = 0.003 and p = 0.001, respectively).

All subgroups showed significant within-group change in E/I MLD. Group A→N showed a significant decrease of −0.013 (p < 0.001), while group A→N showed an increase of 0.005 (p < 0.001). Three-year change in EI differed significantly among the four groups (p = 0.002). Groups N→A and A→A showed significant within-group changes in EI: group N→A and group A→A showed a significant mean increase of 4.81 (p < 0.001) and 0.99 (p < 0.001), respectively.

Table 3 shows the results of analyses of factors associated with moderate to severe exacerbation frequency. We entered age, sex, FEV₁ (%), SGRQ score, BMI, smoking status, major comorbidities, bronchodilator use, and COPD phenotype according to the change in RV/TLC in a univariate analyses (Table 3). FEV₁ (%), SGRQ score, and COPD phenotype according to the change in RV/TLC were significant on the univariate analysis, and they were entered in a multivariate analysis. The multivariable analysis revealed that COPD phenotype according to the change in RV/TLC was not significant. FEV₁ (%) and SGRQ score were significant in the multivariable analysis.

The OS of the four groups was compared using a log-rank test. No pairwise comparison showed a significant difference in OS. The A→A group did not show statistically significant difference in mortality when compared to other groups (Table 4). Cox regression analyses were performed for OS, but no significant associations were observed between shorter survival and pattern of change in the RV/TLC.