Rheumatoid arthritis, quantitative parenchymal lung features and mortality among smokers

Abstract

Objectives

There have been limited investigations of the prevalence and mortality impact of quantitative CT (QCT) parenchymal lung features in RA. We examined the cross-sectional prevalence and mortality associations of QCT features, comparing RA and non-RA participants.

Methods

We identified participants with and without RA in COPDGene, a multicentre cohort study of current or former smokers. Using a k-nearest neighbour quantifier, high resolution CT chest scans were scored for percentage of normal lung, interstitial changes and emphysema. We examined associations between QCT features and RA using multivariable linear regression. After dichotomizing participants at the 75th percentile for each QCT feature among non-RA participants, we investigated mortality associations by RA/non-RA status and quartile 4 vs quartiles 1–3 of QCT features using Cox regression. We assessed for statistical interactions between RA and QCT features.

Results

We identified 82 RA cases and 8820 non-RA comparators. In multivariable linear regression, RA was associated with higher percentage of interstitial changes (β = 1.7 [0.5], P = 0.0008) but not emphysema (β = 1.3 [1.7], P = 0.44). Participants with RA and >75th percentile of emphysema had significantly higher mortality than non-RA participants (hazard ratio [HR] 5.86; 95% CI: 3.75, 9.13) as well as RA participants (HR 5.56; 95% CI: 2.71, 11.38) with ≤75th percentile of emphysema. There were statistical interactions between RA and emphysema for mortality (multiplicative P = 0.014; attributable proportion 0.53; 95% CI: 0.30, 0.70).

Conclusion

Using machine learning-derived QCT data in a cohort of smokers, RA was associated with higher percentage of interstitial changes. The combination of RA and emphysema conferred >5-fold higher mortality.

Rheumatology key messages.

• RA was associated with increased percentage of interstitial changes on quantitative chest CT.

• The combination of RA and emphysematous changes (>75th percentile) conferred >5-fold increased mortality.

• Quantitative CT methods can identify RA patients at increased risk of mortality.

Introduction

Rheumatoid arthritis (RA) is an autoimmune inflammatory condition that affects up to 1% of the general population [1]. Although the hallmark clinical feature of RA is an inflammatory arthritis leading to joint damage, a significant proportion of patients have ‘extra-articular’ disease manifestations in other organ systems including the lungs [2, 3]. RA has been associated with obstructive and interstitial lung diseases [4], and several respiratory risk factors, including smoking [5], for RA have also been identified, suggesting that the lung plays a central role in RA disease pathogenesis [6]. However, these overlapping risk factors have also presented a challenge for investigation of RA-related lung disease, as there has been relatively limited research investigating the burden of RA-related lung diseases and their mortality impacts in population cohorts that adjusted for cigarette smoking and other respiratory risk factors.

CT chest imaging plays a key role in the evaluation of parenchymal lung diseases that may be associated with RA. However, due to the subjective nature of visual scan interpretation, CT chest imaging can be subject to variable ascertainment and insensitive to subtle abnormalities and changes over time. As a result, there has been significant enthusiasm for the development of automated methods that quantitatively evaluate pulmonary abnormalities on CT chest imaging, and several methods have been developed [7]. These quantitative CT (QCT) techniques have demonstrated the ability to predict prognosis in patients with idiopathic pulmonary fibrosis [8, 9] and have also been able to detect quantitative progression of interstitial abnormalities [10]. Recent studies have demonstrated the ability of QCT to detect subclinical RA-associated interstitial lung disease (RA-ILD) [11] and also to predict mortality in patients with clinically apparent RA-ILD [12]. However, there has been limited investigation of mortality associations of QCT methods in RA patients without clinical ILD or in comparison with patients without RA.

There were two main goals of this study: first, to use machine learning-derived QCT data from a cohort of current or former smokers to investigate the cross-sectional prevalence of interstitial and emphysematous lung disease in RA and compare with non-RA participants accounting for smoking, which is a risk factor for both RA and lung disease; and second, to investigate the associations of QCT findings with mortality in RA patients. We hypothesized that RA would be associated with a higher percentage of both interstitial and emphysematous changes detected by QCT methods after adjustment for smoking. We further hypothesized that QCT scoring of interstitial and emphysematous lung changes would identify RA participants at higher risk of mortality.

Methods

Study population and design

The study was conducted using data from COPDGene, a multicentre prospective cohort study of current or former smokers, which has been previously described [13–15]. Between 2007 and 2011, COPDGene enrolled self-identified non-Hispanic White and Black people with at least 10 pack-years of smoking history at 21 clinical centres in the USA. The study included smokers with and without obstructive lung disease. At baseline visits, all participants completed health questionnaires and had high-resolution CT (HRCT) chest imaging, spirometry and genotyping. Participants with known diagnosis of bronchiectasis or ILD, or clinical evidence of diffuse bronchiectasis or ILD on baseline CT imaging were ineligible. We also excluded a small number of non-smokers who were included in COPDGene as healthy controls. We performed a cross-sectional investigation of the associations between RA and QCT features as well as a cohort study investigating the mortality risk of RA and QCT features. This substudy was approved by the Mass General Brigham Institutional Review Board (protocol 2020P000558) and adheres to the Declaration of Helsinki. Signed informed consent was obtained at COPDGene enrolment.

RA cases and non-RA comparators

We identified participants with RA using self-reported history of physician-diagnosed RA combined with medication history. Because self-reported RA status alone has a positive predictive value ranging between 5% and 16% [16], we additionally required the use of at least one DMARD reported on baseline medication questionnaires. We included DMARD medications in a validation study for RA case ascertainment in other cohorts (Supplementary Table S1, available at Rheumatology online) [17]. The combination of self-report RA status and DMARD use has demonstrated high positive predictive value (88%) in a previous study [18]. The non-RA comparator group included all COPDGene participants who did not have an RA history or use a DMARD. Participants who were using one or more DMARDs but did not report a RA history and those who reported an RA history without DMARD use were excluded from the non-RA comparator group.

Quantitative CT evaluation

Baseline HRCTs were assessed using quantitative methods that have been previously described [19, 20]. Briefly, regions of interest in the lung parenchyma of each scan were categorized into normal lung, interstitial changes, or emphysema using a k-nearest neighbour classifier based on the local tissue density and distance from the pleural surface. Interstitial changes were subclassified into reticular, subpleural line, linear scar, honeycombing, centrilobular nodule, nodular and ground glass. Emphysematous changes were subclassified into centrilobular and panlobular emphysema. Each feature was summed and standardized to total lung volume.

Chronic obstructive pulmonary disease

We categorized participants into chronic obstructive pulmonary disease (COPD) severity grades using the Global Initiative for Chronic Lung Disease 2023 classification system [21]. We used each participant’s post-bronchodilator forced expiratory volume in 1 s (FEV₁) and forced vital capacity (FVC) for classification. We also identified the subgroup of participants with preserved ratio impaired spirometry (PRISm), defined as FEV₁/FVC ≥0.70 and FEV₁ <80% predicted.

Mortality

Mortality outcomes were ascertained with biannual contacts as part of longitudinal study follow-up as well as periodic searches of the social security death index, updated through 1 August 2022.

Covariates

Covariates included age at COPDGene baseline visit, sex and self-reported race. BMI was calculated from height and weight measured at the COPDGene baseline examination. Smoking status (current or former) and pack-years of smoking exposure were determined from baseline questionnaires. Because the promoter variant of MUC5B (rs35705950) is the strongest known genetic risk factor for RA-ILD [22] and has been associated with RA in some cohorts [23], we included rs35705950 genotype as a covariate. Genotyping was performed using the TaqMan real-time polymerase chain reaction assay (Thermo Fisher Scientific, Waltham, MA, USA). For analyses that included the rs35705950 genotype, we additionally adjusted for the first six principal components of ancestry to reduce the possibility of confounding by genetic ancestry.

Missing data

Of the participants with available QCT data, 51 were missing postbronchodilator pulmonary function test (PFT) information. Additionally, 521 did not have available genotyping data of rs35705950 and were excluded from the analyses that included the MUC5B promoter variant.

Statistical analysis

The baseline characteristics of RA cases and non-RA comparators were summarized using descriptive statistics. We compared the percentages of normal lung, interstitial lung and emphysematous lung between RA cases and non-RA comparators using Student’s t-test. We also constructed linear regression models that examined the association of RA status with QCT features. The β-coefficients derived from these models represent the percentage difference in QCT feature comparing RA and non-RA participants. The multivariable models were adjusted for age, sex, smoking status (either current or former), pack-years and BMI. We performed a secondary analysis additionally adjusting for the MUC5B promoter variant. We also used multivariable logistic regression to estimate a possible association of RA with COPD, defined as GOLD class 2–4, adjusted for age, sex, smoking status, pack-years and BMI. In separate analyses, we either included or excluded GOLD class 1 with GOLD class 0 in the definition of ‘no COPD’.

To categorize participants by QCT features, we calculated quartiles for each feature among the non-RA comparators and then applied those cut-offs to the RA cases. Because our a priori hypothesis was that RA patients would have increased QCT lung changes compared with the non-RA comparators, we chose to derive the cut points for an increased percentage of QCT features in the non-RA population. This created groups both with and without RA and compared the highest quartile to the lower three quartiles of each QCT feature.

For mortality analyses, we constructed Kaplan–Meier curves with cumulative incidence of mortality based on RA and QCT feature. We also constructed Cox regression models and calculated hazard ratios (HR) for mortality based on cross-classified RA and QCT features. We constructed unadjusted models as well as models adjusted for age, sex, smoking status, pack-years and BMI. Participants were censored at time of death or end of study (1 August 2022, the last date that vital status was verified), whichever came first. We also assessed for interactions between RA and QCT features. To assess for multiplicative interaction, we performed additional models that included an interaction term between RA/non-RA status and continuous QCT feature. To assess for additive interaction using attributable proportion, we dichotomized at the >75th percentile/≤75th percentile of QCT features and assessed for additive interaction between RA/non-RA status and QCT feature.

Due to the large number of participants in the dataset who self-reported RA but were not on DMARD therapy, we repeated our main analyses including all participants who self-reported RA status as RA cases.

We used Schoenfeld residuals to assess the proportional hazards assumptions for all Cox regression models. There were no violations of the proportional hazards assumption except for the covariates BMI and age in the mortality models. Therefore, we constructed models that additionally included interaction terms between follow-up time and BMI or age. In these models, the point estimates of the HR for mortality for the RA group (primary exposure of interest) were unchanged, so we reported the original results. We considered a two-sided P-value of 0.05 to be statistically significant. We performed our analyses using SAS v9.4 (SAS Institute, Cary, NC, USA).

Results

Study sample

Among 10 371 COPDGene participants, we identified 82 RA cases and 8820 non-RA comparators with available baseline QCT data (Fig. 1). RA cases were identified through self-report of physician-diagnosed RA and use of DMARD (Supplementary Table S1, available at Rheumatology online). Participants who self-reported RA but not DMARD use (n = 656) or who were using a DMARD but did not self-report an RA diagnosis (n = 45) were excluded. A total of three RA participants and 516 non-RA comparators did not have QCT scoring and were also excluded.

Figure 1.

Flow diagram of analysed study sample of RA and non-RA comparators in COPDGene. BR: bronchiectasis; ILD: interstitial lung disease; QCT: quantitative computed tomography

Baseline characteristics

The baseline characteristics of RA cases and non-RA comparators are detailed in Table 1. Participants with RA were older (mean 64.0 vs 59.5 years) and a higher proportion were female (65.9% vs 45.9%) and White race (79.3% vs 68.3%). RA cases also had higher mean BMI (30.0 vs 28.7 kg/m²) and a lower proportion of current smokers (32.9 vs 52.5%). The prevalence of the MUC5B rs35705950 (either GT or TT genotype) was 19.7% in RA cases and 14.3% in non-RA comparators.

Table 1.

Demographics, lifestyle factors, pulmonary disease and genetics of RA patients and non-RA comparators at baseline (n = 8902)

	RA cases (n = 82)	Non-RA comparators (n = 8820)
Demographics
Age at enrolment, mean (s.d.), years	64.0 (8.7)	59.5 (9.1)
Female, n (%)	54 (65.9)	4061 (45.9)
White race, n (%)	65 (79.3)	6025 (68.3)
Black race, n (%)	17 (20.7)	2795 (31.7)
Lifestyle
Current smoker, n (%)	27 (32.9)	4628 (52.5)
Past smoker, n (%)	55 (67.1)	4192 (47.5)
Pack-years, mean (s.d.)	42.2 (18.0)	43.9 (24.7)
BMI, mean (s.d.), kg/m2	30.0 (7.9)	28.7 (6.2)
MUC5B promoter variant (rs35705950) genotype, n (%)
GG	61/76 (80.3)	7120/8305 (85.7)
GT	15/76 (19.7)	1122/8305 (13.5)
TT	0/76 (0.0)	63/8305 (0.8)
COPDa, n (%)
GOLD class 0 (no COPD)	44 (53.7)	4892/8769 (55.8)
PRISm	13 (15.9)	1057/8769 (12.1)
GOLD class 1	3 (3.7)	715/8769 (8.2)
GOLD class 2	17 (20.7)	1651/8769 (18.8)
GOLD class 3	13 (15.9)	993/8769 (11.3)
GOLD class 4	5 (6.1)	518/8769 (5.9)

^aBased on GOLD 2023 classification using postbronchodilator FEV₁ and FVC. COPD: chronic obstructive pulmonary disease; GOLD: Global Initiative for Chronic Obstructive Lung Disease; ILA: interstitial lung abnormalities; PRISm: Preserved Ratio Impaired Spirometry (subset of GOLD class 0); SGRQ: Saint George’s Respiratory Questionnaire.

COPD associations

There was a similar proportion of RA cases and non-RA comparators in each GOLD class (Table 1). We did not observe an association between RA status and COPD (Supplementary Table S2, available at Rheumatology online). RA had a multivariable odds ratio for COPD of 0.96 (95% CI: 0.60, 1.54) compared with non-RA, adjusted for age, sex, smoking status and pack-years.

QCT features

RA participants had a lower percentage of normal lung (85.8% vs 91.0%, P = 0.0001) and higher percentage of lung with interstitial abnormalities (7.0% vs 4.8%, P < 0.0001, Table 2). There was a trend toward increased emphysematous changes that did not reach statistical significance (2.6% vs 1.9%, P = 0.09).

Table 2.

Quantitative CT features of RA patients and non-RA comparators (n = 8902)

Feature, median (IQR), % of lung volume	RA cases (n = 82)	Non-RA comparators (n = 8820)	P-value
Normal	85.75 (77.34–92.36)	91.96 (82.72–94.69)	0.0001
Interstitial	7.00 (3.97–9.70)	4.76 (3.00–7.75)	<0.0001
Reticular	6.05 (3.56–8.76)	4.11 (2.59–6.81)	<0.0001
Subpleural line	0.40 (0.22–0.64)	0.31 (0.18–0.51)	0.02
Linear scar	0.17 (0.09–0.30)	0.13 (0.06–0.24)	0.006
Honeycombing	0.05 (0.02–0.11)	0.03 (0.009–0.07)	<0.0001
Centrilobular nodule	0.005 (0.00–0.03)	0.008 (0.000–0.05)	0.47
Nodular	0.008 (0.002–0.03)	0.004 (0.001–0.02)	0.007
Ground glass	0.005 (0.001–0.01)	0.003 (0.00–0.007)	0.01
Emphysema	2.59 (0.77–12.78)	1.92 (0.47–8.81)	0.09
Centrilobular	2.59 (0.77–12.76)	1.92 (0.47–8.77)	0.09
Panlobular	0.00 (0.00–0.02)	0.00 (0.00–0.005)	0.69

IQR: interquartile range.

In multivariable linear regression models adjusted for age, sex, smoking status, pack-years and BMI, RA was associated with higher interstitial percentage than non-RA (adjusted β = 1.71; 95% CI: 0.71, 2.71; P = 0.0008) (Table 3). Among the interstitial subtypes, RA was associated with higher percentage of reticular changes (adjusted β = 1.55; 95% CI: 0.65, 2.45; P = 0.0008), subpleural line (adjusted β = 0.08; 95% CI: 0.02, 0.15; P = 0.009) and honeycombing (adjusted β = 0.06; 95% CI: 0.02, 0.10; P = 0.003). There were no statistical differences in percentage of normal lung (adjusted β = −3.04; 95% CI: −6.44, 0.36; P = 0.08) or emphysematous lung (adjusted β = 1.32; 95% CI: −1.99, 4.64; P = 0.43) after multivariable adjustment. Multivariable models that did not include adjustment for smoking and pack-years are detailed in Supplementary Table S3, available at Rheumatology online, in case smoking may have mediated the relationship.

Table 3.

Linear regression of quantitative CT features comparing RA patients and non-RA comparators (n = 8902)

Characteristic	Unadjusted β (RA vs non-RA) (95% CI)	P-value	Adjusted a β (RA vs non-RA) (95% CI)	P-value
Normal	−4.67 (−8.34, −0.99)	0.01	−3.04 (−6.44, 0.36)	0.08
Interstitial	2.15 (1.12, 3.17)	<0.0001	1.71 (0.71, 2.71)	0.0008
Reticular	1.95 (1.02, 2.88)	<0.0001	1.55 (0.65, 2.45)	0.0008
Subpleural line	0.11 (0.05, 0.18)	0.0005	0.08 (0.02, 0.15)	0.009
Linear scar	0.04 (0.01, 0.08)	0.01	0.03 (−0.004, 0.06)	0.08
Honeycombing	0.07 (0.03, 0.11)	0.0005	0.06 (0.02, 0.10)	0.003
Centrilobular nodule	−0.02 (−0.04, 0.01)	0.28	−0.006 (−0.03, 0.02)	0.69
Nodular	−0.01 (−0.08, 0.05)	0.70	−0.005 (−0.07, 0.06)	0.88
Ground glass	0.0002 (−0.03, 0.03)	0.99	0.002 (−0.03, 0.03)	0.93
Emphysema	2.52 (−1.11, 6.14)	0.17	1.32 (−1.99, 4.64)	0.43
Centrilobular	2.27 (−1.20, 5.75)	0.20	1.05 (−2.12, 4.22)	0.51
Panlobular	0.24 (−0.14, 0.63)	0.22	0.28 (−0.11, 0.66)	0.16

^aAdjusted for age, sex, smoking status (current/past), pack-years and BMI.

Results were similar after additional adjustment for the MUC5B promoter variant rs35705950 (Supplementary Table S4, available at Rheumatology online). RA was associated with higher percentage of interstitial QCT features (β = 1.46; 95% CI: 0.45, 2.47; P = 0.005), reticular features (β = 1.32; 95% CI: 0.40, 2.31; P = 0.005), subpleural line (β = 0.07; 95% CI: 0.002, 0.13; P = 0.04) and honeycombing (β = 0.06; 95% CI: 0.02, 0.10; P = 0.002).

RA, QCT features and mortality risk

Among the RA cases, there were 32 deaths and the median follow-up was 10.7 years. Adjudicated cause of death was available for 17 RA cases and is detailed in Supplementary Table S5, available at Rheumatology online. Among the non-RA comparators, there were 2259 deaths and median follow-up was 10.7 years. We used quartile cut-offs from the non-RA comparators to categorize participants by QCT features. The 10-year mortality for RA participants with >75th percentile of interstitial features was 41.7% (95% CI: 24.8, 58.6%) compared with 30.3% (95% CI: 28.2, 32.5%) for the non-RA comparators with >75th percentile of interstitial features. The RA participants with >75th percentile of emphysema had 10-year mortality of 79.7% (95% CI: 63.4, 95.9%), compared with 46.0% (95% CI: 43.8, 48.3%) for non-RA comparators with >75th percentile of emphysema. Kaplan–Meier curves stratified by RA and QCT status are shown in Fig. 2.

Figure 2.

Cumulative mortality stratified by RA/non-RA comparator status and quantitative CT features in COPDGene (n = 8902). (A) Mortality by RA and QCT (normal lung). ^aFourth quartile has the highest amount of normal lung as determined by quantitative CT. Multiplicative interaction between RA and continuous percentage normal lung was observed (HR 0.98; 95% CI: 0.96, 0.99; P = 0.0045). (B) Mortality by RA and QCT (interstitial). No statistically significant multiplicative or additive interactions observed. (C) Mortality by RA and QCT (emphysema). Multiplicative interaction between RA and continuous emphysema percentage was observed with HR 1.02 (95% CI: 1.00, 1.03; P = 0.0143). Additive interaction was noted between RA and >75th percentile emphysema with attributable proportion 0.53 (95% CI: 0.30, 0.70; P < 0.0001). HR: hazard ratio

In Cox regression models adjusted for age, sex, smoking status, pack-year and BMI, the RA cases with >75th percentile of interstitial lung changes had higher mortality compared with non-RA participants with ≤75th percentile of normal lung (HR 1.91; 95% CI: 1.13, 3.24). The HR when comparing RA with and without >75th percentile of normal lung changes was 1.25 (95% CI: 0.62, 2.52). There were no statistically significant multiplicative or additive interactions between QCT interstitial features and RA on mortality outcomes. Kaplan–Meier curves and Cox regression analyses for subtypes of interstitial features are detailed in Fig. 3.

Figure 3.

Cumulative mortality stratified by RA/comparator and quantitative CT interstitial features in COPDGene (n = 8902). (A) Mortality by RA and QCT (reticular). (B) Mortality by RA and QCT (subpleural line). (C) Mortality by RA and QCT (linear scar). (D) Mortality by RA and QCT (honeycombing). (E) Mortality by RA and QCT (centrilobular nodules). (F) Mortality by RA and QCT (nodular). (G) Mortality by RA and QCT (ground glass). HR: hazard ratio

The combination of emphysema and RA conferred significantly higher risk of mortality in multivariable models. RA participants with >75th percentile of emphysema had higher mortality compared with RA participants with ≤75th percentile of emphysema (HR 5.86; 95% CI: 3.75, 9.13) and non-RA patients with ≤75th percentile of emphysema (HR 5.56; 95% CI: 2.71, 11.38). There was a multiplicative interaction between RA/non-RA status and continuous QCT-derived emphysema percentage on mortality with HR 1.012 (95% CI: 1.00, 1.03; P = 0.014). There was also an additive interaction between RA/non-RA status and the >75th percentile/≤75th percentile of emphysema, with attributable proportion of 0.53 (95% CI: 0.30, 0.70; P < 0.0001). Similar mortality trends were observed for both centrilobular and panlobular emphysema features (Supplementary Fig. S1, available at Rheumatology online). For example, the multivariable HR for mortality was 5.86 (95% CI: 3.75, 9.13) for RA participants and >75th percentile of centrilobular emphysema compared with non-RA comparators with ≤75th percentile of centrilobular emphysema.

In the sensitivity analysis that relaxed the stringent RA case definition to include all participants who self-reported RA, we noted that RA was associated with interstitial features and that the combination of RA with the fourth quartile of both interstitial changes and emphysema was associated with increased mortality (Supplementary Tables S6 and S7, available at Rheumatology online). These findings were attenuated compared with the main analysis. A linear regression model including the relaxed RA case definition and not adjusted for smoking status and pack-years is included in Supplementary Table S8, available at Rheumatology online.

Analyses of correlations between QCT features and spirometry measures as well as correlation between QCT-derived emphysema and honeycombing are detailed in the Supplementary Figs S2–S9, available at Rheumatology online.

Discussion

Using machine learning-derived QCT data in a large, prospective cohort of smokers, we found that RA was associated with a higher percentage of interstitial changes, and this finding persisted after adjustment for smoking and other lifestyle factors. RA was also associated with specific subtypes of interstitial changes, including reticular changes, subpleural lines and honeycombing. The combination of RA and emphysema conferred >5-fold higher mortality. These findings suggest that QCT methods can be used to identify parenchymal lung disease in RA patients and that QCT abnormalities have prognostic significance in RA patients. Using these methods, we found a striking association of RA and emphysema with increased mortality not explained by smoking, suggesting close monitoring is needed for these patients. Additional research is needed to investigate the implementation of QCT screening and monitoring strategies in high-risk RA patients, including those with smoking history.

Our findings add to the existing literature investigating the burden and mortality impact of obstructive and interstitial lung diseases in RA. We noted an association between RA and subclinical interstitial changes on QCT when compared with non-RA comparators. This finding is consistent with prior studies that reported associations between RA and subclinical interstitial lung disease using quantitative methods. One investigation performed comparing RA cases from the ESCAPE-RA cohort with controls from the MESA cohort found that RA patients had an increased percentage of high attenuation areas on cardiac CT imaging compared with population controls [24]. However, this study did not use dedicated full lung HRCT imaging and some may have had clinical ILD, in contrast to our study where participants with known ILD were excluded. Another investigation performed using the ESCAPE-RA cohort used similar methods and found that high attenuation areas were associated with several known RA-ILD risk factors [25]. A recent prospective investigation of RA patients found associations between the MUC5B promoter variant and subclinical ILD diagnosed via quantitative fibrosis score, but did not quantify emphysematous changes, include non-RA comparators, or investigate mortality [11]. However, these studies did not have a non-RA comparator group or investigate mortality outcomes.

Some recent studies have investigated relationships of RA with obstructive lung disease, but there has been limited research focused specifically on emphysema. One recent meta-analysis of four retrospective studies found that RA is associated with increased risk of COPD [26]. In a research screening study performed in a subset of RA patients without known lung disease in the BRASS registry, 36% had undiagnosed emphysema and 47% of emphysema cases occurred in non-smokers [27]. Population-based cohort studies have also noted an association between RA with obstructive PFT abnormalities in the UK Biobank [28] and COPD in the Nurses’ Health Study [29]. RA patients have also reported an increased prevalence of obstructive lung disease in comparison with controls with other autoimmune diseases [30]. Other studies have noted associations between RA autoantibodies, obstructive lung disease and subsequent risk of RA, suggesting a possible pathogenic link between lung disease and development of articular RA [31–33].

Although RA-ILD has been associated with increased mortality in multiple studies [34–39], there has been limited investigation of the mortality impact of obstructive lung disease or emphysema in RA patients. One study compared 594 patients with RA to 596 non-RA controls and reported 2-fold increased mortality in the RA patients with obstructive lung disease compared with non-RA comparators [40]. A population-based study performed in Denmark found that RA patients with COPD had 59.3% mortality at 10 years [41], similar to results that we report here. Another investigation performed in the Nurses’ Health Study found that respiratory disease, including obstructive lung disease, was a major driver of mortality in RA patients [42]. Notably, in our study, the respiratory mortality was higher in RA participants with emphysema than in those with interstitial lung changes. This may be due to selection factors, as clinically apparent ILD was an exclusion criterion for COPDGene, whereas there were no restrictions on participants with emphysema and the cohort included participants with severe obstructive lung disease. Additionally, the mechanisms that explain this mortality interaction remain unclear. The effect of RA treatment with immunosuppressive medications or RA disease activity on emphysema pathogenesis, incidence, progression and mortality are important avenues of future research. As we were unable to investigate longitudinal medication use and RA-related factors such as disease activity, future studies are needed to further investigate this novel finding.

Our study provides support for the use of QCT methods in the evaluation of RA-related lung diseases in research studies. Prior research has established that QCT methods have a robust sensitivity in detecting interstitial lung abnormalities identified by visual inspection [19] and also identify subjects at increased mortality risk, even in the absence of visually defined interstitial lung abnormalities [43]. Other studies have also applied similar methods to the detection of ILD or interstitial features in RA patients [44]. A study by Oh and colleagues demonstrated the ability of QCT to predict mortality in patients with RA-ILD [12]. Another investigation found that RA patients had a higher percentage of high attenuation areas compared with non-RA controls [24]. QCT methods have also been used to identify associations between QCT-derived emphysema and mortality [45] and also used to predict changes in pulmonary physiological parameters and QCT progression over time [10]. To our knowledge, this is the first investigation of QCT-derived emphysema in RA.

Our findings also provide further epidemiological evidence suggesting an association between lung inflammation and RA. The ‘mucosal origins’ hypothesis of RA suggests that mucosal sites, including the respiratory mucosa, serve as important locations where immune dysregulation, dysbiosis and RA-related autoantibody generation occur [6]. Several lines of evidence support this hypothesis, including investigations that have noted high prevalence of interstitial lung abnormalities in subjects that are positive for RA-related autoantibodies, but lack signs of joint inflammation [46, 47]. Additionally, elevated levels of RA-related autoantibodies have been detected in patients with idiopathic pulmonary fibrosis [48]. Further investigations to understand the associations between emphysema and RA are needed.

Our study has several strengths. First, we used data from a prospective, longitudinal cohort that collected HRCT and PFTs for research purposes and obtained extensive data on respiratory risk factors including smoking exposure. We specifically adjusted for known RA-ILD risk factors including cigarette smoking, BMI and the MUC5B promoter variant. The quantitative CT measurements were obtained using validated and established methods. Study recruitment and the application of QCT analysis were agnostic to RA status. Furthermore, similar methods have been used successfully in pulmonary evaluation of RA patients [12, 49]. Finally, the study had a long duration of follow-up, and mortality outcomes were ascertained using robust methods.

Our study also has certain limitations to consider. First, COPDGene was limited to White and Black smokers, which may limit the generalizability of our findings to other populations. However, smoking is a known risk factor for RA [50] and future screening strategies for RA-related lung disease would likely focus on high-risk groups, such as smokers. Second, we identified RA cases using a combination of self-reported RA status and DMARD medication use and were unable to confirm the diagnosis using standard classification criteria. However, similar methods have demonstrated positive predictive value of 88% when used for case-finding in other cohorts [17, 18]. Because COPDGene was not designed to investigate RA, we have limited information on prior RA treatment history, RA disease course and other RA-related factors that may impact respiratory disease in RA. Third, there is a possibility that the quantitative machine learning method may misclassify lung features, including misclassification among and between subtypes (i.e. honeycombing and emphysema). Despite the large size of the cohort, the number of cases with RA and number of mortality outcomes in this group were small, which may have limited our power to detect associations with RA in certain subgroups. We were also limited in our ability to investigate change in QCT features over time since many of the RA patients did not participate in follow-up study visits. For this reason, we investigated all-cause mortality rather than cause-specific mortality. Future studies investigating changes in QCT features over time in RA and non-RA participants are needed.

In conclusion, using machine learning-derived QCT data in a cohort of smokers, we found that RA was associated with higher percentage of interstitial changes, even after adjustment for smoking and other lifestyle factors. The combination of RA and emphysema conferred >5-fold higher mortality. Further studies investigating screening strategies in RA patients at high risk for lung disease are needed.

Supplementary material

Supplementary material is available at Rheumatology online.

Data availability

Data are available upon reasonable request and with appropriate institutional review board approval.

Funding

G.C.M. is supported by National Institute of Arthritis and Musculoskeletal and Skin Diseases (T32 AR007530 and T32 AR055885) and the Rheumatology Research Foundation Scientist Development Award. T.J.D. is supported by the National Institutes of Health/National Heart, Lung, and Blood Institute (R01HL155522). J.A.S. is supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases (grant numbers R01 AR080659, R01 AR077607, P30 AR070253 and P30 AR072577), the R. Bruce and Joan M. Mickey Research Scholar Fund, and the Llura Gund Award funded by the Gordon and Llura Gund Foundation. M.H.C. was supported by R01HL153248, R01HL149861 and R01HL135142. The COPDGene was supported by Award Number U01 HL089897 and Award Number U01 HL089856 from the National Heart, Lung, and Blood Institute. COPDGene is also supported by the COPD Foundation through contributions made to an Industry Advisory Board that has included AstraZeneca, Bayer Pharmaceuticals, Boehringer Ingelheim, Genentech, GlaxoSmithKline, Novartis, Pfizer and Sunovion. The funders had no role in the decision to publish or in the preparation of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of Harvard University, its affiliated academic healthcare centres or the National Institutes of Health.

Disclosure statement: K.Y. reports consulting fees to OM1 unrelated to this work. P.A.J. reports honoraria from Bristol Myers Squibb, Boehringer Ingelheim and AstraZeneca, and grant support from Novartis, unrelated to this work. M.M. reports institutional grant support from Bayer. M.H.C. has received grant funding from Bayer, unrelated to this work. T.J.D. has received support from Bayer and Bristol Myers Squibb, consulting fees from Boehringer Ingelheim and L.E.K. Consulting, and has been part of a clinical trial funded by Genentech. P.F.D. reports grant support from Genentech and Bristol Myers Squibb, royalties or licenses from Up To Date, Inc., and participation on a Federal Drug Administration advisory board. Z.S.W. reports grant support from BMS and Sanofi, royalties or licenses from COVID-19 Symptom Severity Index, consulting fees from Sanofi, Horizon, MedPace, Viela Bio, Zenas, Shionogi and PPD, and participation in data safety monitoring board or advisory board for Sanofi, Horizon, Novartis, Visterra/Otsuka and Shionogi, unrelated to this work. G.M.H. reports consulting fees from Boehringer-Ingelheim, the Gerson Lehrman Group and Chugai Pharmaceuticals, unrelated to this work. E.K.S. has received grant support from Bayer. R.S.J.E. received grant support from Boehringer Ingelheim, and he is a co-founder and equity holder of Quantitative Imaging Solutions. G.R.W. reports grants from Boehringer Ingelheim, consultancy for Pulmonx, Janssen Pharmaceuticals, Novartis and Vertex, and is founder and co-owner of Quantitative Imaging Solutions. J.A.S. has received research support from Bristol Myers Squibb and performed consultancy for AbbVie, Amgen, Boehringer Ingelheim, Bristol Myers Squibb, Gilead, Inova Diagnostics, Janssen, Optum, Pfizer and ReCor unrelated to this work. The remaining authors have declared no conflicts of interest.