Lung Quantitative Computed Tomography Textures are Associated with Systemic Inflammation and Mortality in COPD

Abstract

Background:

Chronic obstructive pulmonary disease (COPD) is characterized by persistent inflammation that is responsible for remodeling the bronchovascular bundles, which may lead to poor quality of life. Quantitative computed tomography (QCT) textures of the lung can capture local disease patterns of inflammation and related respiratory morbidity.

Research Question:

Are bronchovascular bundle textures, obtained from the adaptive multiple feature method (AMFM), associated with systemic inflammation, morbidity, and mortality in COPD?

Study Design and Methods:

We analyzed data from the SPIROMICS (n = 2,981) and COPDGene (n = 10,305) studies. The predictors included two QCT biomarkers, the bronchovascular bundles (BVB) and CT density gradient (CTDG) textures, age, sex, BMI, race, smoking status, smoking pack-years, CT emphysema, and Pi10 (airway wall thickness). Outcomes included plasma biomarker concentrations from Meso Scale Discovery proteomics assays and complete blood counts, both as markers of inflammation, along with FEV₁, FEV₁/FVC ratio, SGRQ, 6MWD, and mMRC dyspnea scale. Associations of these QCT textures with FEV₁ decline and all-cause mortality were also investigated.

Results:

Increased BVB texture was significantly associated with elevated neutrophil and monocyte counts, and the neutrophil-to-lymphocyte ratio (NLR), independent of clinical covariates, CT emphysema, and Pi10. Elevated CTDG was associated with increased neutrophil count, NLR, and tumor necrosis factor (TNF)-α. Increased CTDG and BVB textures were also associated with a lower FEV₁ and six-minute walk distance. CTDG at baseline was also associated with decline in FEV₁ at five-year follow-up in COPDGene. We observed a significant association of both BVB (HR_SPIROMICS=1.084, 95% CI: 1.035, 1.135, P<0.001; HR_COPDGene=1.106, 95% CI: 1.080, 1.131, P<0.001) and CTDG (HR_SPIROMICS=1.033, 95% CI: 1.003, 1.064, P=0.03; HR_COPDGene=1.079, 95% CI: 1.061, 1.096, P<0.001) textures with all-cause mortality independent of CT emphysema and Pi10.

Interpretation:

QCT textures may provide imaging evidence of the spatial heterogeneity of lung inflammation and overall disease burden in COPD.

Clinical Trial Registration:

SPIROMICS (NCT01969344); COPDGene (NCT00608764)

Chronic obstructive pulmonary disease (COPD) is characterized by irreversible airflow obstruction associated with abnormal inflammatory response to inhaled irritants. While inflammation typically begins in the large airways, it progressively involves the small airways, lung parenchyma, and pulmonary vasculature, leading to remodeling of the bronchovascular bundles.^1–3 These structural changes can be detected in vivo through chest computed tomography (CT) and contribute to a decline in lung function and an increase in the risk of mortality.^4–6 Percent emphysema and airway wall thickness are two well-known CT biomarkers associated with poor clinical outcomes in COPD.^4,6 However, these global metrics fail to capture important regional variations in bronchovascular structures, which may be crucial for understanding chronic inflammation and related respiratory morbidity. Understanding the relationship between airway and vessel regions and inflammation is important, as these are the lung areas where inflammatory cell infiltrates accumulate during immune responses and tertiary lymphoid structures form in advanced COPD.⁷

Quantitative computed tomography (QCT) texture analysis is increasingly utilized to identify structural and functional abnormalities in various lung diseases, including COPD and idiopathic pulmonary fibrosis.^8–11 A recent study identified a novel QCT texture biomarker, obtained from the adaptive multiple feature method (AMFM), for predicting severe COPD exacerbations.^12–14 AMFM is a texture analysis tool used to quantify radiologic patterns within the lung, and it might offer a more accurate assessment of inflammation-related structural variations in bronchovascular bundles compared to traditional metrics. We hypothesize that the local textures of the bronchovascular bundles, along with the surrounding regions, as quantified by AMFM, are associated with systemic inflammation, impaired lung function, and all-cause mortality, independent of CT-based emphysema and airway wall thickening.

We tested our hypothesis using data from two large multicenter cohorts of current and former smokers: the SubPopulations and InteRmediate Outcome Measures in COPD Study (SPIROMICS)¹⁵ and the Genetic Epidemiology of COPD (COPDGene).¹⁶ Some of the preliminary results of this study have been previously reported in the form of an abstract.¹⁷

STUDY DESIGN AND METHODS

Study Cohorts

We analyzed data from two independent cohort studies, SPIROMICS¹⁵ and COPDGene¹⁶. Both studies are being conducted at multiple centers across the United States. Note that although some clinical sites participate in both studies, by design the overlap in participants or first-degree relatives is prohibited. Written consent was provided by all participants, and the protocols were approved by the Institutional Review Boards (IRBs) of each participating study center.

SPIROMICS enrolled 2,981 participants who underwent high-resolution chest computed tomography (CT) scans at total lung capacity (TLC) and residual volume (RV).¹⁸ Participants were enrolled into one of the four strata: individuals who never smoked, individuals who smoked but had no airflow obstruction, individuals with mild-to-moderate COPD, and individuals with severe COPD.¹⁵

COPDGene enrolled 10,305 participants, also scanned at multiple lung volumes: TLC and functional residual capacity (FRC).¹⁶ Participants who did not exhibit airflow obstruction based on spirometry (defined as a forced expiratory volume in 1 second (FEV₁)/forced vital capacity (FVC) ratio of ≥0.70 and FEV₁≥80% of the predicted value) were categorized as having normal spirometry. Others who exhibited airflow obstruction based on spirometry results were categorized based on the global initiative for chronic obstructive lung disease (GOLD) spirometry guideline.¹⁹ Those with an FEV₁/FVC ratio of 0.70 or higher but an FEV₁ less than 80% of the predicted value were classified as having preserved ratio impaired spirometry (PRISm).¹⁶ To ensure consistency with SPIROMICS, PRISm subjects were excluded from the COPDGene participants.

Predictor Variables

Our primary predictors were two QCT texture biomarkers of the lung: bronchovascular bundles (BVB) and CT density gradient (CTDG) textures, obtained using AMFM^12–14. The BVB texture captures the structure of extra-mediastinal airways and vasculature, while the CTDG texture captures changing CT density surrounding the bronchovascular bundles and fissures; both expressed as percentages of lung at TLC.

The AMFM textures were derived from a Bayesian classifier trained on three-dimensional volume of interest (VOIs) from the TLC scans annotated by a consensus of board-certified chest radiologists and pulmonologists. After annotation, first-order and geometric descriptors were extracted from these VOIs, refined through feature selection, and used to train the classifier. This enabled quantification of six texture patterns: normal texture, increased CT density (ICTD), emphysema, reticulation with ICTD, BVB, and CTDG.^12–14 Of these six textures, BVB and CTDG were selected to characterize the structure of bronchovascular bundles and their surroundings. For more details on the development of AMFM textures, please refer to the cited studies.^12–14

Outcomes

Outcome variables indicating systemic inflammation included nine plasma proteomic biomarkers obtained from the Meso Scale Discovery (MSD) assay (Malmö, Sweden): tumor necrosis factor (TNF)-α; interleukin (IL)-2, −6, −8, −10; interferon (IFN)-γ; CCL11 (eotaxin); CCL26 (eotaxin 3); CCL17. ²⁰ Additional biomarkers of systemic inflammation included complete blood counts (CBCs) that were performed in each study (at enrollment in SPIROMICS and at Phase II visit in COPDGene) by the clinical center laboratories. CBCs included white blood cell (WBC) count, platelet count, percentages of neutrophils, lymphocytes, monocytes, and eosinophils, and neutrophil-to-lymphocyte ratio (NLR).^21,22

We also studied the association of BVB and CTDG textures with lung function measures including postbronchodilator FEV₁ in liters (L) and FEV₁/FVC. Additionally, we examined the association between QCT textures and respiratory quality of life quantified using the total St. George’s Respiratory Questionnaire (SGRQ) score, which varies from 0 to 100, where 0 signifies no symptom burden and a good quality of life. We also investigated the relationship of QCT texture with six-minute walk distance (6MWD) in meters (m) and the modified Medical Research Council (mMRC) dyspnea scale. This scale increases from 0 to 4 as a function of shortness of breath. We also examined the associations between both QCT textures and FEV₁ decline over five years. Since SPIROMICS had five-year data available, we chose this as the cutoff for the analysis. The change in FEV₁ was calculated by subtracting the baseline FEV₁ from the follow-up FEV₁ at five years. This difference was then divided by the interval between visits to obtain the annualized change in mL / year. A combined texture feature, derived by summing BVB and CTDG textures, was also calculated to investigate its association with the annualized change in FEV₁. Additionally, we leveraged longitudinal data to directly evaluate changes in FEV₁ across both cohorts. Lastly, we employed all-cause mortality data censored at 10 years from SPIROMICS (Dec’21) and COPDGene (Sept’23).

Statistical Analysis

All the plasma biomarker concentrations, except for the already normally distributed CBC data, were log-transformed to ensure normality. We constructed univariate regression models to investigate relationships between QCT textures and each blood plasma biomarker. We applied a false discovery rate (FDR) correction of 10% to the univariate P values. We used this analysis to retain the plasma biomarkers that were significantly linked to BVB and CTDG textures. We further conducted multivariable analysis between the QCT textures and systemic inflammation markers including both proteomics and blood count markers. The models were adjusted for age, sex, race, BMI, current smoking status, smoking pack-years, FEV₁, FVC, percent emphysema defined as proportion of voxels below −950 Hounsfield Units (HU) on a TLC scan, the square root of the wall area of a hypothetical airway with a 10 mm lumen perimeter (Pi10), and CT scanner type.

We also conducted multivariable regression analysis to examine the associations of QCT textures with lung function and clinical outcomes in COPD. For evaluating the associations with lung function measures, the multivariable models were adjusted for the same covariates mentioned above except FEV₁ and FVC. Furthermore, we examined the distribution of both BVB and CTDG across GOLD stages and assessed differences using the Kruskal-Wallis test. For assessing the association of QCT textures with total SGRQ score, 6MWD (m), and mMRC dyspnea scale, we added FEV₁ and FVC as additional adjustment variables. An independent analysis was also conducted to examine the correlation between both BVB and CTDG textures with Pi10.

We examined the association between change in FEV₁ and QCT textures (BVB, CTDG, and their sum), adjusting for the same variables mentioned above. We also analyzed these associations after stratifying subjects into GOLD 0 and GOLD 1–4 subgroups. This analysis was repeated using linear mixed models, incorporating random intercepts for participants and CT scanner type. Unadjusted Cox Proportional Hazards models were constructed to estimate hazard ratios for both BVB and CTDG textures, followed by adjusted models that included clinical covariates, emphysema, and Pi10. Furthermore, sensitivity analyses were performed using additional imaging biomarkers such as parametric response mapping (PRM)-derived functional small airways disease (fSAD) and the ratio of blood vessel volume (cross-sectional-area<5mm²) (BV5) and total pulmonary vessel volume (TBV) in both cohorts. A P value less than 0.05 was considered significant. All analyses were performed in R version 4.1.1.

RESULTS

Study Design

SPIROMICS enrolled 2,981 individuals between November 12, 2010, and July 31, 2015, of whom eight withdrew consent, 855 had missing clinical information and QCT biomarker data at baseline (Figure-1). We analyzed data from the remaining 2,118 participants at enrollment to investigate the associations of QCT textures with lung function and respiratory morbidity (Figure-1). Out of these 2,118 participants, proteomics biomarkers from the MSD assay were available for 851 participants, and CBC was available for 2,348 participants (Figure-1). For studying the change in FEV₁ at five-year follow-up, spirometry data was available only for 1,014 individuals from SPIROMICS (Figure-1). All-cause mortality data at 10 years was available for 2,374 individuals (Figure-1).

Figure 1:

Flow Diagram.

COPDGene enrolled 10,305 participants, of whom 3,228 had missing clinical information and QCT texture information, yielding data from 7,077 participants at enrollment (Figure-1). In COPDGene, MSD data was available for 1,503 participants, and CBC for 471 participants (Figure-1). Spirometry at five-year follow-up was available for 4,094 individuals, while all-cause mortality data was available for 7,996 participants (Figure-1).

Participant Characteristics

The mean age at baseline was 63.0 (SD:9) years in SPIROMICS and 60.0 (SD:9) years in COPDGene (Table-1). The gender distribution between males and females was well-balanced in both cohorts, with no significant differences observed. However, the COPDGene cohort had a significantly higher percentage of individuals who were actively smoking at baseline (50%) (Table-1). The mean percent BVB texture was 12.09% (SD:1.84%) in SPIROMICS and 12.08% (SD:1.99%) in COPDGene (Table-1). Similarly, percent CTDG had mean value of 3.2% in both cohorts (Table-1). Both BVB and CTDG highlighted on TLC scans, showcasing varying degrees of texture in two participants from both SPIROMICS and COPDGene is illustrated (Figure-2). Distribution statistics of plasma biomarkers and participant characteristics at five-year follow-up from both cohorts are presented in the supplementary section (Tables-E1–E3).

Table 1:

Baseline characteristics of study participants from SPIROMICS and COPDGene.

	SPIROMICS (n = 2,118)	COPDGene (n = 7,077)	P value
Age, years	63 (9)	60 (9)	<0.001
Gender			0.13
Male	1,108 (52%)	3,837 (54%)
Female	1,010 (48%)	3,240 (46%)
Race			<0.001
White	1,624 (77%)	5,054 (71%)
Non-White	494 (23%)	2,023 (29%)
Current smoking status			<0.001
No	1,320 (62%)	3,566 (50%)
Yes	798 (38%)	3,511 (50%)
Smoking pack years *	46 (28)	44 (25)	<0.001
BMI, kg/m2	27.9 (5.3)	28.4 (5.8)	0.15
FEV1, liters	2.16 (0.91)	2.30 (0.95)	<0.001
FVC, liters	3.46 (1.02)	3.42 (1.00)	0.2
FEV1/FVC ratio	0.62 (0.17)	0.66 (0.17)	<0.001
GOLD stages			<0.001
Never smokers	154 (7.3%)	91 (1.3%)
0	670 (32%)	3,430 (48%)
1	288 (14%)	626 (8.8%)
2	567 (27%)	1,572 (22%)
3	319 (15%)	926 (13%)
4	120 (5.7%)	432 (6.1%)
Total SGRQ score	31 (21)	26 (23)	<0.001
Six-minute walk distance, meters	402 (116)	422 (123)	<0.001
mMRC Dyspnea score			<0.001
0	751 (35%)	3,258 (46%)
1	863 (41%)	1,008 (14%)
2	304 (14%)	908 (13%)
3	155 (7.3%)	1,258 (18%)
4	45 (2.1%)	645 (9.1%)
Emphysema (%)	7.2 (9.9)	7 (10.1)	0.4
Pi10	3.74 (0.11)	3.67 (0.13)	<0.001
AMFM: Bronchovascular (BVB)	12.09 (1.84)	12.08 (1.99)	0.6
AMFM: CT density gradients(CTDG)	3.2 (3.3)	3.2 (3.4)	0.023

Data are reported as mean (SD) or n (%). SPIROMICS: Subpopulations and Intermediate Outcome Measures In COPD Study; COPDGene: Genetic Epidemiology of COPD; BMI: body-mass index; FEV₁: forced expiratory volume in 1 second; FVC: forced vital capacity; GOLD: global initiative for chronic obstructive lung disease; SGRQ: St. George’s respiratory questionnaire; mMRC: modified medical research council; Emphysema (%): percent LAA < −950 Hounsfield units (HU); Pi10: square root of the wall area of a hypothetical airway with a lumen perimeter of 10 mm; AMFM: adaptive multiple feature method;

^*Pack years defined as the number of packs of cigarettes smoked in a day multiplied by the number of years an individual has smoked. P values were generated using Wilcoxon’s rank sum test or Pearson’s Chi-squared test.

Figure 2: Visualization of the bronchovascular bundles (BVB) and CT density gradients (CTDG) textures on two different participants from the SPIROMICS and COPDGene cohorts.

BVB texture (shown in pink) captures the airways and vessels, while the CTDG texture (shown in gold) surrounds them. The percentage of adaptive multiple feature method (AMFM) derived normal texture for both participants is also tabulated. BVB, CTDG, and normal textures are only three of the six texture features; therefore, the sum does not add to 100. The SPIROMICS subject is a 55-year-old female categorized as GOLD 3 with a postbronchodilator FEV₁ of 0.795, while the COPDGene subject is a 57.5-year-old male categorized as GOLD 0 with a postbronchodilator FEV₁ of 2.736.

QCT Textures and Systemic Inflammation

We first investigated FDR-corrected univariate associations of QCT texture with proteomics biomarkers from MSD assay and CBC data with the BVB texture (Table-E4,E6). None of the cytokines in the MSD assay were significantly associated with BVB texture with 0.1 threshold for FDR-adjusted P value (Table-E4). Consequently, no cytokines from the MSD panel were included in the multivariable regression analysis. However, FDR-corrected univariate analysis using the CBC data revealed five significant associations with BVB texture in both cohorts (Table-E6). After adjustment, platelet count (SPIROMICS:β=−2.100, 95%CI: −3.668, −0.580, P=0.007;COPDGene:β=−4.927, 95%CI: −7.962, −1.893, P=0.002), neutrophil count (SPIROMICS:β=0.059, 95%CI: 0.019, 0.100, P=0.004; COPDGene:β=0.150, 95%CI: 0.059, 0.241, P=0.001), monocyte count (SPIROMICS:β=0.007, 95%CI: 0.003, 0.012, P=0.001; COPDGene:β=0.016, 95%CI: 0.008, 0.025, P<0.001), and neutrophil-to-lymphocyte ratio (NLR) (SPIROMICS:β=0.086, 95%CI: 0.048, 0.125, P<0.001; COPDGene:β=0.200, 95%CI: 0.080, 0.319, P=0.001), remained significantly associated with BVB texture (Figure-3).

Figure 3: Association of AMFM textures with inflammatory biomarkers.

Multivariable associations of bronchovascular bundles (BVB) and CT density gradients (CTDG) textures, and inflammatory biomarkers collected using Meso Scale Discovery (MSD) platform and Complete Blood Count (CBC). All the associations shown have a P value less than 0.05. BVB: Bronchovascular bundles; CTDG: CT density gradients; SPIROMICS: Subpopulations and Intermediate Outcome Measures In COPD Study; COPDGene: Genetic Epidemiology of COPD; CI = confidence interval; TNF - α: tumor necrosis factor - α;

Based on univariate analysis, CTDG was significantly associated with six proteomics biomarkers from the MSD panel and four significant associations from the CBC panel in SPIROMICS and COPDGene after FDR correction (Table-E5,E7). Adjusted analysis revealed that out of all these biomarkers, CTDG remained significantly associated with TNF-α (SPIROMICS:β= 0.014, 95%CI: 0.001, 0.028, P = 0.04; COPDGene:β=0.017, 95%CI: 0.000, 0.034, P=0.05), NLR (SPIROMICS:β=0.065, 95%CI: 0.035, 0.096, P<0.001; COPDGene:β=0.144, 95%CI: 0.056, 0.233, P=0.001), and neutrophil count (SPIROMICS:β=0.041, 95%CI: 0.009, 0.073, P=0.01; COPDGene:β=0.098, 95%CI: 0.030, 0.165, P=0.001) in both cohorts (Figure-3).

Lung Function and Respiratory Morbidity

Multivariable regression analysis revealed that BVB texture was associated with postbronchodilator FEV₁ (SPIROMICS:β=−0.056, 95%CI: −0.072, −0.040, P<0.001; COPDGene:β=−0.022 95%CI: −0.029, −0.015, P<0.001) (Table-2). Additionally, BVB texture demonstrated a significant association with six-minute walk distance in both SPIROMICS and COPDGene post adjustment (Table-2). The distribution of BVB was also significantly different across never-smokers and all GOLD stages (0 through 4) (Figure-E1). Dyspnea score was only significantly associated with BVB in COPDGene cohort (β=0.028 95%CI: 0.004, 0.053, P=0.02) (Table-2). CTDG demonstrated significant associations with measures of lung functions:postbronchodilator FEV₁ (SPIROMICS:β=−0.029, 95%CI: −0.041, −0.016, P<0.001; COPDGene:β=−0.027, 95%CI: −0.034, −0.021, P<0.001) and postbronchodilator FEV₁ / FVC (SPIROMICS:β=−0.007, 95%CI: −0.009, −0.005, P<0.001; COPDGene:β=−0.008, 95%CI: −0.009, −0.007, P<0.001) independent of all the confounders. An upward trend in CTDG was observed with increasing COPD severity (Figure-E1). CTDG was also significantly associated with all the studied indicators of respiratory morbidity after adjusting for the clinical covariates and imaging biomarkers including emphysema and Pi10 (Table-2). Correlation analyses revealed that both these textures were not highly correlated with Pi10 (Figure-E2).

Table 2:

Multivariable linear regression analysis for assessing associations of bronchovascular bundles (BVB) texture and CT density gradients (CTDG) texture with lung function and respiratory morbidity in SPIROMICS and COPDGene (Estimate, 95% CI, P Value).

	BVB				CTDG
	SPIROMICS		COPDGene		SPIROMICS		COPDGene
	β (95% CI)	P Value	β (95% CI)	P Value	β (95% CI)	P Value	β (95% CI)	P Value
FEV1, (liters) *	−0.056 (−0.072, −0.040)	P < 0.001	−0.022 (−0.029, −0.015)	P < 0.001	−0.029 (−0.041, −0.016)	P < 0.001	−0.027 (−0.034, −0.021)	P < 0.001
FEV1/FVC ratio *	−0.003 (−0.005, 0.000)	0.06	0.000 (−0.001, 0.001)	0.95	−0.007 (−0.009, −0.005)	P < 0.001	−0.008 (−0.009, −0.007)	P < 0.001
Total SGRQ score **	0.098 (−0.335, 0.532)	0.66	0.216 (−0.005, 0.437)	0.06	0.926 (0.586, 1.267)	P < 0.001	0.994 (0.789, 1.200)	P < 0.001
Six-minute walk distance, (meters) **	−2.773 (−5.394, −0.151)	0.04	−2.624 (−3.861, −1.386)	P < 0.001	−5.062 (−7.127, −2.998)	P < 0.001	−6.350 (−7.500, −5.200)	P < 0.001
mMRC dyspnea score **	0.023 (−0.029, 0.074)	0.38	0.028 (0.004, 0.053)	0.02	0.059 (0.019, 0.099)	0.004	0.08 (0.058, 0.102)	P < 0.001

BVB: Bronchovascular bundles; CTDG: CT density gradients; SPIROMICS: Subpopulations and Intermediate Outcome Measures In COPD Study;COPDGene: Genetic Epidemiology of COPD; CI: confidence interval; FEV₁: forced expiratory volume in 1 second; FVC: forced vital capacity; SGRQ: St. George’s respiratory questionnaire; mMRC: modified medical research council;

^*Models controlled for age, gender, race, BMI, current smoking status, smoking pack-years, percent emphysema defined as percent LAA < −950 Hounsfield units (HU), square root of the wall area of a hypothetical airway with a lumen perimeter of 10 mm (Pi10), and CT scanner type.

^**Models controlled for age, gender, race, BMI, current smoking status, smoking pack-years, FEV₁, FVC, percent emphysema, Pi10, and CT scanner type.

QCT Textures and Decline in FEV₁

In COPDGene, CTDG texture showed a significant association with the change in FEV₁ for all participants (β=−0.967, 95%CI: −1.840, −0.094; P=0.03) (Table-3). We stratified the overall population based on varying degrees of airflow obstruction: GOLD 0 and GOLD 1–4 individuals. After adjusting for the covariates, we found that each additional 1% increase in CTDG texture was associated with a 1.346 mL / year decline in FEV₁ for GOLD 1–4 participants (P=0.009) (Table-3). In SPIROMICS, neither BVB nor CTDG textures showed a significant association with changes in FEV₁, regardless of whether the population was stratified by GOLD stages (Table-3). Additionally, the sum of BVB and CTDG texture demonstrated significant association with decline in FEV₁ in COPDGene (GOLD 1–4), but no associations were observed in SPIROMICS (Table E8). A similar trend was observed with linear mixed models (Table E9).

Table 3:

Association between bronchovascular bundles (BVB) texture and CT density gradients (CTDG) texture, and change in FEV₁ (mL / year) stratified by baseline COPD GOLD stage (Estimate, 95% CI, P Value).

		BVB				CTDG
	Population	SPIROMICS		COPDGene		SPIROMICS		COPDGene
		β (95% CI)	P Value	β (95% CI)	P Value	β (95% CI)	P Value	β (95% CI)	P Value
Change in FEV1 (mL/year)	Overall	−0.246 (−2.301, 1.809)	0.81	−0.449 (−1.307, 0.408)	0.3	−0.760 (−2.557, 1.037)	0.41	−0.967 (−1.840, −0.094)	0.03
	GOLD 0	0.534 (−3.315, 4.384)	0.78	−0.008 (−1.285, 1.270)	P > 0.99	1.287 (−2.982, 5.556)	0.55	–0.175 (−2.661, 2.311)	0.89
	GOLD (1–4)	0.134 (−2.502, 2.771)	0.92	−0.910 (−2.125, 0.305)	0.14	−0.467 (−2.645, 1.711)	0.67	−1.346 (−2.359, −0.333)	0.009

COPDGene: Genetic Epidemiology of COPD; CI: confidence interval; FEV₁: forced expiratory volume in 1 second; GOLD: global initiative for chronic obstructive lung disease; Models are controlled for age, gender, race, BMI, current smoking status, smoking pack-years, FEV₁, FVC, percent emphysema defined as percent LAA < −950 Hounsfield units (HU), square root of the wall area of a hypothetical airway with a lumen perimeter of 10 mm (Pi10), and CT scanner type.

All-cause Mortality

After censoring, 76% of participants were alive in SPIROMICS, while 79% of individuals were alive in COPDGene. Cox Proportional Hazards models revealed that BVB texture was significantly associated with all-cause mortality in both SPIROMICS and COPDGene. Specifically, the unadjusted hazard ratios for BVB texture were 1.079 (95%CI: 1.032, 1.127; P<0.001) in SPIROMICS and 1.100 (95%CI: 1.075, 1.126; P<0.001) in COPDGene (Table-4). After adjusting for confounding factors including age, gender, race, BMI, current smoking status, smoking pack-years, FEV₁, FVC, percent emphysema, and Pi10, these associations remained significant, with hazard ratios of 1.084 (95%CI: 1.035, 1.135; P<0.001) and 1.106 (95%CI: 1.080, 1.131; P<0.001), respectively (Table-4).

Table 4:

Association between bronchovascular bundles (BVB) texture and CT density gradients (CTDG) texture, and all-cause mortality censored at 10 years. (Hazard ratio, 95% CI, P Value).

	BVB				CTDG
	SPIROMICS		COPDGene		SPIROMICS		COPDGene
	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value
Unadjusted	1.079 (1.032, 1.127)	P < 0.001	1.100 (1.075, 1.126)	P < 0.001	1.123 (1.103, 1.142)	P < 0.001	1.158 (1.148, 1.169)	P < 0.001
Adjusted	1.084 (1.035, 1.135)	P < 0.001	1.106 (1.080, 1.131)	P < 0.001	1.033 (1.003, 1.064)	0.03	1.079 (1.061, 1.096)	P < 0.001

BVB: Bronchovascular bundles; CTDG: CT density gradients; SPIROMICS: Subpopulations and Intermediate Outcome Measures In COPD Study; COPDGene: Genetic Epidemiology of COPD; HR: hazard ratio; CI: confidence interval; Adjusted Cox proportional hazards model are controlled for age, gender, race, BMI, current smoking status, smoking pack-years, FEV₁, FVC, percent emphysema defined as percent LAA < −950 Hounsfield units (HU), and square root of the wall area of a hypothetical airway with a lumen perimeter of 10 mm (Pi10).

Similarly, CT density gradients (CTDG) texture also showed a significant association with all-cause mortality. In the adjusted models with Pi10 and emphysema, CTDG texture remained significantly associated with all-cause mortality in both cohorts. In SPIROMICS, the hazard ratio was 1.033 (95%CI: 1.003, 1.064; P=0.03), while in COPDGene, the hazard ratio was 1.079 (95%CI: 1.061, 1.096; P<0.001) (Table-4). Additional sensitivity analyses demonstrated that adjusting for PRM-fSAD did not alter these associations. (Table-E10).

DISCUSSION

In this study of two large cohorts of individuals with smoking-related COPD, we found that QCT textures of the bronchovascular bundles and regions surrounding them were associated with plasma biomarkers of systemic inflammation, lung function, respiratory morbidity, and all-cause mortality. The BVB texture captures the density distribution and morphology of the airways and the pulmonary blood vessels, while the CTDG texture characterizes changing CT densities in the vicinity of the bronchovascular bundles and interlobar fissures.¹⁴ The association of these QCT biomarkers with different blood biomarkers of inflammation suggests that they may capture different elements of the heterogeneity of COPD pathogenesis. These findings are noteworthy because airway inflammation and the subsequent remodeling of airways and vasculature are widely recognized as key indicators of continued COPD progression, which aligns with GOLD stage assessment of disease severity as well.^1,2 By directly analyzing the connections between these systemic biomarkers and QCT textures, we build upon previous studies that have linked those biomarkers and adverse COPD outcomes.^23,24

The BVB texture was found to be negatively associated with platelet count, a relationship that may be explained by the non-linear association between platelets and respiratory morbidity in COPD subjects.²⁵ We also noted a positive association of BVB texture with the neutrophil and monocyte counts, which are cells known to contribute to regional tissue destruction in COPD.^26–28 Additionally, the positive correlation of both BVB and CTDG texture with the neutrophil-to-lymphocyte ratio (NLR), a straightforward and reliable biomarker of inflammation,^29,30 reinforces our argument that these textures might offer a visual representation of inflammation in those lung regions, which to date have not figured prominently in COPD imaging. Additionally, the positive association of CTDG with TNF-α is noteworthy, since it is a potent inflammatory cytokine speculated to be a mediator of cigarette smoke-induced diseases like COPD.^31,32

Another important result of this study was that we observed an association of QCT textures with all-cause mortality, in both SPIROMICS and COPDGene cohorts, even when accounting for percent CT emphysema and Pi10, and PRM-fSAD.^4,6 A concurrent association of these QCT textures with systemic inflammation may suggest an increased risk of mortality in individuals with elevated levels of plasma biomarkers of inflammation. Based on our findings, these AMFM textures provide anatomic information that complements the well-known identification by percent emphysema, airway wall thickening, increased blood vessel density, and loss of terminal bronchioles, all pathological changes in COPD believed to result from inflammation.³³ We speculate that BVB texture might be capturing local airway wall thickening, potentially in a more direct manner than Pi10, and any inflammation-induced neovascularization. On the other hand, CTDG texture captures the changes in the regions around these bronchovascular bundles.

Our work has several potential limitations, including the inability as a descriptive study to establish causation. Both our QCT measures and plasma biomarkers were acquired at a single time-point. While this design precludes any statement about repeatability, it is offset by the spatial consistency between results of the SPIROMICS and COPDGene cohorts, a significant strength. Additionally, CBC data in COPDGene was acquired at phase 2 (approximately five years after enrollment) as opposed to the rest of the data that was collected at baseline. Furthermore, in our analysis of FEV₁ decline, we observed a discrepancy in the association between CTDG and FEV₁ change, with CTDG showing associations in COPDGene only. This is likely due to the higher annualized FEV₁ decline and larger sample size in COPDGene compared to SPIROMICS.

In future studies, we will investigate whether these associations are maintained during progression, and importantly, whether baseline QCT measures predict future clinical outcomes other than severe exacerbations. We acknowledge that the inflammatory response in COPD is intricate and is unlikely to be captured fully by our restricted panel of systemic biomarkers. Finally, we did not analyze biomarkers that would directly represent the lung environment, such as from bronchoalveolar lavage fluid or sputum, a potential extension for which the collection of biospecimens in SPIROMICS is ideally suited.

INTERPRETATION

In conclusion, both BVB and CTDG textures have the potential to serve as visual indicators of inflammation in individuals with or at risk for COPD with significant associations with respiratory morbidity and all-cause mortality. If validated, they could join with the detection of small-airways disease by PRM, a precursor of emphysema ³⁴, to provide non-invasive means to test novel disease-modifying therapies in far smaller groups of research participants than is feasible for studies employing spirometric outcomes.

Abbreviation List

6MWD

Six-minute walk distance

AMFM

Adaptive multiple feature method

BMI

Body mass index

BV5

Peripheral pulmonary blood vessel volume (vessels < 5 mm² in cross-sectional area)

BVB

Bronchovascular bundles

CBC

Complete blood count

COPD

Chronic obstructive pulmonary disease

COPDGene

The Genetic Epidemiology of COPD

CTDG

CT density gradient

FDR

False discovery rate

FEV₁

Forced expiratory volume in 1 second

FVC

Forced vital capacity

fSAD

Functional small airways disease

GOLD

Global initiative for chronic obstructive lung disease

IFN-γ

Interferon-γ

IL

Interleukin

mMRC

Modified medical research council

MSD

Meso scale discovery

NLR

Neutrophil-to-lymphocyte ratio

Pi10

The square root of the wall area of a hypothetical airway with a 10 mm lumen perimeter

PRM

Parametric response mapping

QCT

Quantitative computed tomography

SGRQ

St. George’s Respiratory Questionnaire

SPIROMICS

SubPopulations and InteRmediate Outcome Measures in COPD Study

TBV

Total pulmonary blood vessel volume

TNF-α

Tumor necrosis factor – α

WBC

White blood cell