Abstract
Background:
There is uncertainty regarding patient outcomes with using tumor necrosis factor inhibitors (TNFi) vs. other biologic and targeted-synthetic disease-modifying anti-rheumatic drugs (non-TNFi b/tsDMARDs) in rheumatoid arthritis-associated interstitial lung disease (RA-ILD). Therefore, we compared survival and respiratory hospitalization outcomes following TNFi or non-TNFi b/tsDMARD treatment initiation in RA-ILD.
Methods:
We performed an active-comparator, new-user cohort study with propensity score matching following the Target-Trial-Emulation Framework using national U.S. Veterans Affairs (VA) electronic and administrative health records between 1/1/2006 and 6/30/2021. VA healthcare enrollees fulfilling a validated administrative algorithm for RA-ILD and no prior receipt of ILD directed therapies (e.g., antifibrotics) initiating a TNFi or non-TNFi b/tsDMARD were studied. Propensity score matching was performed using demographics, healthcare utilization, health behaviors, comorbidity burden, RA-related severity factors, and ILD-related severity factors, including baseline forced vital capacity. Study outcomes were respiratory hospitalization, all-cause mortality, and respiratory-related death over up to 3-years of follow-up from VA, Medicare, and National Death Index data. Patients were not involved in the study.
Findings:
We matched 237 TNFi and 237 non-TNFi initiators with a mean age of 68 years and 92% (n=234) being male. The composite outcome of death or respiratory hospitalization did not significantly differ among non-TNFi vs. TNFi (adjusted hazard ratio [aHR] 1·21 [0·92, 1·58]). The individual outcomes of respiratory hospitalization (non-TNFi vs. TNFi aHR 1·27 [0·91, 1·76]), all-cause mortality (aHR 1·15 [0·83, 1·60]), and respiratory mortality (aHR 1·38 [0·79, 2·42]) also did not differ. Secondary, sensitivity, and subgroup analyses supported the primary findings.
Interpretation:
No significant difference in survival or respiratory hospitalization was observed between TNFi or non-TNFi b/tsDMARDs in U.S. veterans with RA-ILD. These data do not support systematic avoidance of TNFi in all people with RA-ILD. Comparative effectiveness trials in RA-ILD are needed given the potential for residual confounding and/or selection bias in observational studies.
Funding:
U.S. Department of Veterans Affairs.
INTRODUCTION
Interstitial lung disease (ILD) clinically affects approximately 10% of rheumatoid arthritis (RA) patients, with another 20-30% having subclinical disease.1,2 The median survival after RA-ILD diagnosis is poor, ranging from 3-8 years.1,3 Despite a growing number of effective disease-modifying anti-rheumatic drugs (DMARDs) for RA, there is a paucity of data on the effectiveness of immunomodulatory therapies for RA-ILD. Further, the potential for pulmonary toxicities accompanying DMARDs complicates RA-ILD treatment decisions.4-6
Tumor necrosis factor inhibitors (TNFi) are the most commonly used biologic DMARDs in RA, but there is concern about their safety in RA-ILD. Many cases of ILD development or exacerbation have been anecdotally reported with TNFi use, but causality is difficult to infer as patients receiving TNFi typically have more severe RA.5 Few comparative observational studies have investigated TNFi safety in RA-ILD. A small, single-center, case-control study in Japan found more ILD presenting or progressing amongst RA patients treated with TNFi compared to abatacept or tocilizumab.7 Other observational comparative studies have not observed significantly increased risks of death,8,9 ILD exacerbation,10 or hospitalization11 with TNFi compared to conventional synthetic (cs)DMARDs,9 rituximab,8 or abatacept.10 Many of these estimates were imprecise, and the safety of TNFi in RA-ILD remains uncertain.
In place of TNFi, providers may select non-TNFi biologic or targeted synthetic (b/ts)DMARDs in RA-ILD patients. Single-arm studies have found most RA-ILD patients had pulmonary function stability after treatment with rituximab,12 abatacept,13 or tocilizumab.14 In a small study of patients on established therapy (i.e., current users), bDMARDS, and specifically non-TNFi agents, were associated with slower worsening of respiratory status compared to nonuse.15 Recently, a multicenter cohort study found that patients with RA-ILD initiating rituximab, azathioprine, and mycophenolate mofetil had stabilization of forced vital capacity (FVC), without meaningful differences between these agents.16 Beyond RA-ILD, tocilizumab is FDA-approved for systemic sclerosis (SSc)-associated ILD based on results of the FaSScinate and FocuSSced trials. Together, these findings have resulted in some organizations issuing caution about TNFi use in patients with RA-ILD and poor respiratory reserve.17 American College of Rheumatology/American College of Chest Physicians ILD Treatment Guidelines conditionally recommend against TNFi specifically for the treatment of ILD in RA-ILD.18
Several methodological limitations of prior studies such as small sample sizes (particularly among control groups), case misclassification, lack of adjustment for ILD severity, prevalent user study designs (rather than new user), comparisons of users vs. non-users, and an inability to assess survival have limited the conclusions that can be drawn. Given persistent uncertainty regarding the safety of TNFi vs. other b/tsDMARDs in RA-ILD, we utilized the Target-Trial-Framework19 to perform a real-world comparative study of TNFi vs. non-TNFi b/tsDMARDs in U.S. veterans with RA-ILD. Based on prior literature, we hypothesized that non-TNFi b/tsDMARDs would be associated with improved survival and fewer respiratory hospitalizations than TNFi.
METHODS
Study design
We performed an active-comparator, new-user, retrospective cohort study of RA-ILD patients initiating TNFi or non-TNFi b/tsDMARDs within the Veterans Health Administration (VA) between January 1, 2006 and December 31, 2018 using the Target-Trial-Emulation Framework19 (Appendix p. 2). The VA is the largest integrated healthcare system in the U.S. with robust administrative and electronic health record (EHR) data as well as established linkages to external data sources, including Medicare and National Death Index (NDI).20 The study was approved by the VA Central IRB (#1619487). Study reporting is in accordance with STROBE and RECORD-PE guidelines .21,22 A graphical depiction of the study design is provided in Figure 1.
Figure 1. Graphical depiction of study design.
Study design is graphically depicted illustrating the evaluation of eligibility criteria and covariable assessment prior to biologic or targeted-synthetic disease-modifying anti-rheumatic (b/tsDMARD) drug start date. Study outcomes of death and respiratory hospitalization were ascertained after b/tsDMARD start date. Graphical depiction template adapted from Wang SV & Schneeweiss S, Clin Epidemiol 2022; 14:601-608.
Abbreviations: bDMARD, biologic disease-modifying anti-rheumatic drug; CMS, Centers for Medicare Services; CTD, connective tissue disease; ILD; interstitial lung disease; MS, multiple sclerosis; NDI, National Death Index; RA, rheumatoid arthritis; tsDMARD, targeted-synthetic disease-modifying anti-rheumatic drug; VA, Veterans Health Administration.
Study population
We selected U.S. veterans receiving VA healthcare with RA-ILD using validated algorithms that required ≥2 diagnostic codes for RA and ≥2 diagnostic codes for ILD, separated in time. Such algorithms have >70% positive predictive value (PPV) for RA-ILD.23 In sensitivity analyses, we required further ILD evidence by either a diagnostic code for ILD from a pulmonology encounter or completion of chest CT and pulmonary function tests (PPV >80%23). We identified the first initiation of a TNFi or non-TNFi b/tsDMARD occurring after fulfilling the RA-ILD algorithm, excluding those with an active course for that specific medication in the prior 12 months (Figure 1). Individuals with a history of multiple sclerosis or lymphoproliferative malignancy were excluded, as were individuals with diagnostic codes for other connective tissue diseases or who had received any prior treatment with ILD-directed therapies (azathioprine, mycophenolate, cyclophosphamide, tacrolimus, cyclosporine, nintedanib, pirfenidone). Recognizing that non-TNFi b/tsDMARDs are typically used after TNFi, we also performed analyses among a cohort with ≥1 prior b/tsDMARD use before the eligible medication course (Appendix p. 3-4).
Medications
Initiations of TNFi (etanercept, adalimumab, infliximab, certolizumab, golimumab) and non-TNFi b/tsDMARDs (tocilizumab, sarilumab, abatacept, rituximab, tofacitinib, baricitinib, upadacitinib) were identified from VA pharmacy dispensing data and intravenous medication administration records. Individual dispensing episodes without a 90-day gap between dispensing episodes were assembled into continuous treatment courses, as previously described.24 The date of medication course initiation was designated the cohort entry date. In primary analyses, individuals were considered exposed to the initial therapy for the duration of follow-up (i.e., intention-to-treat approach).
Study outcomes
The primary composite outcome was time to death or respiratory hospitalization over up to a 3-year follow-up period after treatment initiation, assessed between January 1, 2006 and June 30, 2021. Secondary outcomes were all-cause mortality, respiratory-related mortality, and respiratory-related hospitalization assessed individually. Because of anticipated medication switching or discontinuation during follow-up that may result in exposure misclassification, primary and secondary outcomes were additionally assessed over a 1-year maximum follow-up period. Respiratory hospitalizations were identified as the primary discharge diagnosis in VA, non-VA care billed to the VA, and linked Medicare claims. Vital status was determined from VA death records and the NDI. Respiratory-related hospitalizations and death were based on International Classification of Diseases (ICD)-9 (466·xx-519·xx) and ICD-10 (J09·xx-J99·xx) codes. For descriptive purposes, respiratory-related hospitalizations and deaths were sub-categorized as ILD, using diagnostic codes from our case-finding algorithm, or chronic obstructive pulmonary disease (COPD), asthma, lower respiratory tract infection, and other causes according to World Health Organization definitions23 (Appendix p. 5). Recognizing the use of FVC in clinical trials but with variable testing frequency in real-world settings, we descriptively evaluated post-treatment FVC values between 1- and 3-years after cohort entry date.
Covariables
Patient demographics including gender, hospitalization in the prior year, smoking status, body mass index, Elixhauser comorbidity score, RA-related factors, and ILD-related factors were obtained from VA administrative and EHR data prior to cohort entry. RA-related factors included time since first RA diagnostic code, number of prior DMARDs using all available data before cohort entry, use of individual conventional DMARDs (methotrexate, leflunomide, hydroxychloroquine, sulfasalazine) in the prior 365 days, glucocorticoid use in the prior 365 days, number of rheumatology visits in the prior 365 days, elevation in erythrocyte sedimentation rate (ESR) or C-reactive protein (CRP), and positive anti-cyclic citrullinated peptide (CCP) antibody or rheumatoid factor (RF). ILD-related factors included time since first ILD diagnostic code, inhaled glucocorticoid use in the prior 365 days, supplemental oxygen use, and FVC. The most recent FVC value before the cohort entry date was obtained from discrete data elements in VA databases or clinical notes through natural language processing.25 Missing covariables were modeled using the missing indicator method. Recognizing the importance of FVC as a prognostic variable,3 a separate cohort was constructed requiring a non-missing pre-cohort entry date FVC value as additional eligibility criteria.
Statistical analysis
Given strong temporal trends in the selection of TNFi vs. non-TNFi, we constructed calendar time-period specific propensity score (PS) models to facilitate estimation of the average treatment effect of the treated. Logistic regression models predicting treatment assignment were separately executed for 1/1/2006-12/31/2010, 1/1/2011-12/31/2015, and 1/1/2016-12/31/2018 including the aforementioned covariables. TNFi and non-TNFi initiators were 1:1 PS-matched within these periods using nearest-neighbor matching without replacement and a caliper of 0·2-times the pooled standard deviation of the logit of the PS.26 Balance of covariables before and after matching was assessed with standardized differences. Variables with standardized differences ≤0·1 were considered adequately balanced.
Individuals were followed from cohort entry until the first of respiratory hospitalization, death, or end of the follow-up period (maximal follow-up time or study end date). Incidence rates of primary and secondary outcomes were calculated for PS-matched TNFi and non-TNFi initiators. Cox proportional hazards regression models compared the risk of primary and secondary outcomes with adjustment for any unbalanced variables. Based on feasibility queries and anticipated event rates, we were powered to detect a hazard ratio (HR) of 0·67 favoring non-TNFi over TNFi.
While the primary analyses used an intention-to-treat approach, as-treated analyses were completed with censoring for treatment discontinuation, including a 90-day exposure extension. In a sensitivity analysis to balance follow-up length, the matched pair was censored when either patient ended follow-up. To address the correlated nature of matched sets, we performed post hoc analyses using Cox models with shared frailty and robust variance estimation for the matched sets. Alternative modeling of prior b/tsDMARD use in PS models was tested and produced similar results (data not shown). Inverse probability of treatment weighting (IPTW) was performed with stabilized weights to estimate the average treatment effect. Subgroup analyses were performed within the matched primary cohort separately comparing matched sets of rituximab or other non-TNFi/non-rituximab, given rituximab is often selected in RA-ILD and had been previously compared with TNFi,8,18 as well as matched sets with age greater or less than 65 years. The proportional hazards assumption was tested with negative log-log plots and Schoenfeld residuals, which did not indicate violations (data not shown). To estimate the degree to which residual or unmeasured confounding may influence results, E values were calculated to measure the minimum associations of the unmeasured confounder with the treatment and outcome that are needed to explain away the estimated HR and confidence intervals (CI) to a range of possible “true” values, using the formula for HR with common outcomes.27 All analyses were completed using Stata v18 (StataCorp) within the VA Informatics and Computing Infrastructure.
Role of the funder and patient involvement
The U.S. Department of Veterans Affairs funded the study and had no role in the design, conduct, or reporting of the study. Patients were not involved in the design or conduct of the study.
RESULTS
Among 1,047 b/tsDMARD initiators with prevalent RA-ILD (n=705 TNFi; n=342 non-TNFi), we PS-matched 237 TNFi to 237 non-TNFi initiators in the primary cohort (Appendix p. 3). After PS-matching, mean age was 68 years, the cohort was male predominant (n=434; 92%), and had frequent smoking history (Table 1). Most patients were seropositive (n=389; 82%) and used glucocorticoids (n=347; 73%) while the minority used supplemental oxygen (n=124; 26%) or had a baseline FVC <60% predicted (n=80; 16·9%). Patient characteristics were balanced between TNFi and non-TNFi initiators after PS-matching, with the exception of slight imbalances in BMI, elevated ESR/CRP, and inhaled glucocorticoid use (Figure 2, Table 1). Patient characteristics were similarly balanced in the additional study cohorts (Appendix p. 6-7). In the primary cohort, the most common TNFi were adalimumab (n=120; 50·6%) and etanercept (n=87; 36·7%) while the most common non-TNFi were rituximab (n=126; 53·2%) and abatacept (n=66; 27·8%) (Appendix p. 8).
TABLE 1.
Patient characteristics before and after propensity-score matching in the primary cohort.
| Before matching | After 1:1 match | |||||
|---|---|---|---|---|---|---|
| TNFi (n=705) |
Non-TNFi (n=342) |
TNFi (n=237) |
Non-TNFi (n=237) |
|||
| Mean (SD) or N (%) |
Mean (SD) or N (%) |
Std Diff† | Mean (SD) or N (%) |
Mean (SD) or N (%) |
Std Diff† | |
| Demographics | ||||||
| Age, years | 67·7 (9·0) | 67·9 (9·1) | 0·017 | 68·1 (9·3) | 67·9 (8·8) | 0·023 |
| Gender | ||||||
| Male | 663 (94·0) | 304 (88·9) | 0·185 | 218 (92·0) | 216 (91·1) | 0·030 |
| Female | 42 (6·0) | 38 (11·1) | 19 (8·0) | 21 (8·9) | ||
| Race | ||||||
| White | 574 (81·4) | 268 (78·4) | 0·079 | 189 (79·7) | 187 (78·9) | 0·020 |
| Non-White | 131 (18·6) | 74 (21·6) | 48 (20·3) | 50 (21·1) | ||
| Health status | ||||||
| Smoking status | ||||||
| Never | 87 (12·3) | 45 (13·2) | 0·080 | 30 (12·7) | 30 (12·7) | 0·009 |
| Former | 229 (32·5) | 119 (34·8) | 87 (36·7) | 86 (36·3) | ||
| Current | 358 (50·8) | 167 (48·8) | 113 (47·7) | 114 (48·1) | ||
| Unknown/missing | 31 (4·4) | 11 (3·2) | 7 (3·0) | 7 (3·0) | ||
| BMI category | ||||||
| <25 kg/m2 | 135 (19·2) | 75 (21·9) | 0·083 | 54 (22·8) | 50 (21·1) | 0·121 |
| 25 to <30 kg/m2 | 295 (41·8) | 141 (41·2) | 94 (39·7) | 95 (40·1) | ||
| 30 to <35 kg/m2 | 174 (24·7) | 77 (22·5) | 61 (25·7) | 55 (23·2) | ||
| ≥35 kg/m2 | 100 (14·2) | 48 (14·0) | 28 (11·8) | 37 (15.6) | ||
| Unknown | 1 (0·1) | 1 (0·3) | 0 | 0 | ||
| Elixhauser score | 18·3 (7·7) | 20·8 (8·3) | 0·321 | 20·1 (8·8) | 19.6 (8·2) | 0·055 |
| Hospitalization in prior 365 days | 160 (22·7) | 120 (35·1) | 0·275 | 67 (28·3) | 68 (28·7) | 0·009 |
| RA-related factors | ||||||
| RA duration, years* | 5·3 (4·8) | 7·4 (5·3) | 0·425 | 6·9 (5·5) | 6.5 (5·3) | 0·077 |
| Number of rheumatologist visits in prior 365 days | 3·1 (2·3) | 3·7 (2·3) | 0·239 | 3·4 (2·5) | 3.4 (2·0) | 0·004 |
| Elevated ESR or CRP | ||||||
| No | 197 (27·9) | 91 (26·6) | 0·168 | 55 (23·2) | 68 (28·7) | 0·135 |
| Yes | 399 (56·6) | 216 (63·2) | 150 (63·3) | 143 (60·3) | ||
| Missing | 109 (15.5) | 35 (10·2) | 32 (13·5) | 26 (11·0) | ||
| Anti-CCP or RF positivity | ||||||
| Negative | 63 (8·9) | 28 (8·2) | 0·037 | 18 (7·6) | 24 (10·1) | 0·092 |
| Positive | 562 (79·7) | 278 (81·3) | 198 (83·5) | 191 (80·6) | ||
| Unknown/Missing | 80 (11·3) | 36 (10·5) | 21 (8·9) | 22 (9·3) | ||
| Number of prior DMARDS (ever) | 2·4 (1·4) | 3.2 (1·8) | 0·574 | 2.9 (1·4) | 2.7 (1·6) | 0·095 |
| Medication use in past 365 days | ||||||
| Methotrexate | 255 (36·2) | 108 (31·6) | 0·095 | 78 (32·9) | 79 (33·3) | 0·009 |
| Leflunomide | 201 (28·5) | 115 (33·6) | 0·110 | 81 (34·2) | 70 (29·5) | 0·100 |
| Hydroxychloroquine | 262 (37·2) | 138 (40·4) | 0·064 | 97 (40·9) | 93 (39·2) | 0·034 |
| Sulfasalazine | 185 (26·2) | 70 (20·5) | 0·138 | 48 (20·3) | 52 (21·9) | 0·041 |
| Oral glucocorticoid | 465 (66·0) | 262 (76·6) | 0·238 | 173 (73·0) | 174 (73·4) | 0·010 |
| ILD-related factors | ||||||
| ILD duration, years* | 3·1 (3·2) | 3·0 (2·9) | 0·049 | 3·1 (2·9) | 3·0 (3·0) | 0·030 |
| Home oxygen use | 158 (22·4) | 92 (26·9) | 0·104 | 62 (26·2) | 62 (26·2) | 0·000 |
| Inhaled glucocorticoid use | 126 (17·9) | 80 (23·4) | 0·136 | 40 (16·9) | 52 (21·9) | 0·128 |
| Baseline FVC % pred. | ||||||
| >80% | 236 (33·5) | 115 (33·6) | 0·220 | 79 (33·3) | 78 (32·9) | 0·079 |
| 60–80% | 214 (30·4) | 117 (34·2) | 71 (30·0) | 78 (32·9) | ||
| <60% | 98 (13·9) | 61 (17·8) | 40 (16·9) | 40 (16·9) | ||
| Missing | 157 (22·3) | 49 (14·3) | 47 (19·8) | 41 (17·3) | ||
Duration is time from first diagnostic code
Standardized differences ≤0·1 were considered evidence of adequate balance and calculated according to the Yang and Dalton method.
Abbreviations: Anti-CCP, anti-cyclic citrullinated protein antibody; BMI, body mass index; CRP, c-reactive protein; DMARDs, disease-modifying anti-rheumatic drugs; ESR, erythrocyte sedimentation rate; FVC, forced vital capacity; ILD, interstitial lung disease; RA, rheumatoid arthritis; RF, rheumatoid factor; Std diff, standardized difference; TNFi, tumor necrosis factor.
Figure 2. Distribution of propensity scores and variable balance before and after matching in primary cohort.
Panel A, Kernal density plots of propensity score (PS) distributions before and after matching in the primary cohort. Panel B, Standardized differences before and after propensity score matching in the primary cohort. Variables with standardized difference >0·1 are bolded.
Abbreviations: CCP, cyclic-citrullinated peptide; CRP, C-reactive protein; DMARDs, disease-modifying anti-rheumatic drugs; Elev, elevated; ESR, erythrocyte sedimentation rate; FVC, forced vital capacity; HCQ, hydroxychloroquine; LEF, leflunomide; MTX, methotrexate; Pos, positive; RF, rheumatoid factor; SSZ, sulfasalazine; Supp, supplemental.
In the primary cohort, 112 non-TNFi and 102 TNFi initiators experienced the composite outcome of death or respiratory hospitalization over 3-years of follow-up (incidence rate [IR] 22·2 vs. 18·8/100 person-years [PY], respectively) (Table 2). There was no significant difference in the composite outcome in non-TNFi vs. TNFi (aHR 1·21 [95% CI 0·92, 1·58]) (Table 2, Figure 3A). Because of early separation of survival curves, we repeated the analysis excluding events during the first 30 days which produced overlapping results (aHR 1·13 [0·86, 1·49]). Over a 1-year follow-up period, results were similar with IRs of 25·2 and 17·9/100PY in non-TNFi and TNFi, respectively (Table 2). There was no significant difference in the composite outcome at 1-year (aHR 1·41 [0·93, 2·13]).
TABLE 2.
Study outcomes in the primary cohort
| Events | PY | IR (95% CI) /100PY |
HR (95% CI)* | aHR (95% CI)† | |
|---|---|---|---|---|---|
| Study outcomes through up to 3 years | |||||
| Composite of death or respiratory hospitalization | |||||
| TNFi | 102 | 554·0 | 18·8 (15·4, 22·8) | Ref | Ref |
| Non-TNFi | 112 | 503·8 | 22·2 (18·5, 26·8) | 1·18 (0·90, 1·55) | 1·21 (0·92, 1·58) |
| Respiratory hospitalization | |||||
| TNFi | 67 | 554·0 | 12·3 (9·7, 15·6) | Ref | Ref |
| Non-TNFi | 80 | 503·8 | 15·9 (12·8, 19·8) | 1·28 (0·93, 1·77) | 1·27 (0·91, 1·76) |
| Death | |||||
| TNFi | 69 | 616·2 | 11·2 (8·8, 14·2) | Ref | Ref |
| Non-TNFi | 73 | 597·9 | 12·2 (9·7, 15·4) | 1·09 (0·79, 1·52) | 1·15 (0·83, 1·60) |
| Respiratory death | |||||
| TNFi | 22 | 616·2 | 3·6 (2·4, 5·4) | Ref | Ref |
| Non-TNFi | 28 | 597·9 | 4·7 (3·2, 6·8) | 1·31 (0·75, 2·29) | 1·38 (0·79, 2·42) |
| Study outcomes through up to 1 year | |||||
| Composite of death or respiratory hospitalization | |||||
| TNFi | 39 | 218·5 | 17·9 (13·0, 24·4) | Ref | Ref |
| Non-TNFi | 52 | 206·5 | 25·2 (19·2, 33·0) | 1·41 (0·93, 2·13) | 1·41 (0·93, 2·13) |
| Respiratory hospitalization | |||||
| TNFi | 28 | 218·5 | 12·8 (8·8, 18·6) | Ref | Ref |
| Non-TNFi | 42 | 206·5 | 20·3 (15·0, 27·5) | 1·58 (0·98, 2·55) | 1·53 (0·95, 2·48) |
| Death | |||||
| TNFi | 19 | 229·9 | 8·3 (5·3, 13·0) | Ref | Ref |
| Non-TNFi | 22 | 226·0 | 9·7 (6·4, 14·8) | 1·18 (0·64, 2·18) | 1·25 (0·68, 2·32) |
| Respiratory death | |||||
| TNFi | 8 | 229·9 | 3·5 (1·7, 7·0) | Ref | Ref |
| Non-TNFi | 8 | 226·0 | 3·5 (1·8, 7·1) | 1·02 (0·38, 2·71) | 1·06 (0·40, 2·83) |
N=474 (237 TNFi; 237 non-TNFi)
HR is non-TNFi vs. TNFi
Adjusted for variables with Std Diff >0·1 (BMI category, ESR or CRP elevation, inhaled glucocorticoid use)
Abbreviations: aHR, adjusted hazard ratio; CI, confidence interval; IR, incident rate; HR, hazard ratio; PY, person-years; Ref, referent; TNFi, tumor necrosis factor inhibitor
Figure 3. Kaplan-Meier curves among RA-ILD patients initiating TNFi or non-TNFi b/tsDMARDs in the primary cohort.
Kaplan-Meier curves show outcome free survival amongst propensity score (PS)-matched RA-ILD patients (n=474) initiating either TNFi or non-TNFi b/tsDMARDs. Panel A shows the composite outcome of death or respiratory hospitalization, panel B shows respiratory hospitalization, panel C shows all-cause mortality, and panel D shows respiratory mortality.
Abbreviations: b/tsDMARDs, biologic or targeted-synthetic disease-modifying anti-rheumatic drugs; RA-ILD, rheumatoid arthritis-associated interstitial lung disease; TNFi, tumor necrosis factor inhibitor.
Respiratory hospitalizations occurred among 80 non-TNFi and 67 TNFi initiators over 3-years of follow-up (IR 15·9 vs. 12·3/100PY). ILD was responsible for 21·8% (n=32) of hospitalizations and lower respiratory tract infections were the most common cause at 33·3% (n=49) (Appendix p. 9). There was no significant difference in respiratory hospitalization risk in non-TNFi vs. TNFi (aHR 1·27 [0·91, 1·76]) (Table 2, Figure 3B). Similar findings were observed over 1-year of follow-up (aHR 1·53 [0·95, 2·48]).
Over 1,214 person-years of follow-up, 73 deaths occurred in non-TNFi vs. 69 in TNFi. There was no difference in all-cause mortality risk between non-TNFi vs. TNFi (aHR 1·15 [0·83, 1·60]) (Table 2, Figure 3C). Of the 142 deaths, 50 (35·2%) were respiratory-related, with ILD being the most common respiratory cause of death (n=30) (Appendix p. 9). There were also no significant differences in respiratory mortality risk in non-TNFi vs. TNFi (aHR 1·38 [0·79, 2·42]) (Table 2, Figure 3D). Over a 1-year of follow-up period, mortality risk did not differ between non-TNFi vs. TNFi initiators (all-cause: aHR 1·25 [0·68, 2·32]; respiratory: aHR 1·06 [0·40, 2·83]) (Table 2).
Mean (SD) duration of medication use before discontinuation, an event, or censoring was 1·2 (1·0) years for TNFi and non-TNFi. Time-to-discontinuation did not significantly differ for non-TNFi vs. TNFi (3-year: aHR 0·95 [0·78, 1·16]; 1-year: aHR 0·91 [0·71, 1·16]). Between 1-3 years after treatment initiation, FVC values were available for 93 non-TNFi and 87 TNFi initiators. Mean FVC percent predicted during this period was similar amongst non-TNFi and TNFi (mean [SD] 75·1 [17·9] vs. 76·3 [17·8]).
We assessed the robustness of findings by evaluating outcomes in additional cohorts over a 3-year follow-up period. Among patients with ≥1 previous b/tsDMARD use (n=165 TNFi; n=165 non-TNFi), 94·9% (n=313) and 10·9% (n=36) had used a TNFi or non-TNFi previously, respectively. In this cohort, the point estimates favored non-TNFi across primary and secondary outcomes (range aHR 0·77 to 0·88), although differences were not statistically significant (Figure 4, Appendix p. 10). In a cohort requiring a pre-treatment FVC value (n=201 TNFi; n=201 non-TNFi), the mean (SD) FVC percent predicted was 76·2 (17·3) in TNFi vs. 76·1 (18·4) in non-TNFi. Findings in this cohort were similar to the primary cohort (Figure 4, Appendix p. 10). Requiring additional ILD criteria for eligibility (n=185 TNFi; n=185 non-TNFi) also produced results similar to the primary cohort (Figure 4, Appendix p. 10).
Figure 4. Forest plot of results from primary analysis, secondary analyses, and sensitivity analyses over up to 3-years of follow-up.
Adjusted hazard ratios depicting risk of study outcomes in non-TNFi vs. TNFi initiators for the composite outcome of death or respiratory hospitalization as well as the individual outcomes of respiratory hospitalization and all-cause mortality over up to 3-years of follow-up. Standardized differences are provided in the Appendix p. 6-7 and a full listing of these results is provided in the Appendix p. 10-14.
Abbreviations: aHR, adjusted hazard ratio; b/tsDMARD, biologic/targeted-synthetic disease-modifying anti-rheumatic drug; CI, confidence interval; FVC, forced vital capacity; ILD, interstitial lung disease; RTX, rituximab; TNFi, tumor necrosis factor inhibitor
Within the primary cohort we performed sensitivity and subgroup analyses evaluating 3-year outcomes. IPTW and shared frailty models produced results consistent with our primary analysis (Figure 4, Appendix p. 12). The 95% CI were minimally changed with robust variance estimation (aHR 1·21 [0·91, 1·59]). As-treated analyses and censoring the matched pairs both produced significantly higher rates of the composite outcome among non-TNFi initiators. Risk of death, respiratory death, and respiratory hospitalization did not significantly differ between TNFi and non-TNFi in as-treated analyses. In sub-group analyses comparing matched sets of rituximab vs. TNFi (n=252), there were no significant differences in study outcomes (Figure 4, Appendix p. 13). Comparisons of other non-TNFi with TNFi (n=222) produced results even closer to the null. Among patients >65 years of age (n=196), non-TNFi treatment had a higher risk of the composite outcome, all-cause mortality, and respiratory mortality than TNFi (Figure 4, Appendix p. 14). Among those ≤65 years of age (n=278), there were no significant differences.
E values were calculated across a range of possible “true estimates” to quantify the potential for unmeasured confounding to influence results.27 E values ranged from 1·5 (“true” HR of 1·0) to 3·1 (“true” HR of 0·5) (Appendix p. 15).
DISCUSSION
Given concerns about TNFi use in RA-ILD,5,7,8,10 we conducted an active-comparator, new-user study following the Target-Trial-Framework19 comparing TNFi and non-TNFi b/tsDMARDs in U.S. veterans with RA-ILD. Using national VA data and PS-matching, we found no significant differences in survival or respiratory-related hospitalization rates between RA-ILD patients initiating TNFi vs. non-TNFi b/tsDMARDs. Several secondary, sensitivity, and subgroup analyses were performed that generally supported these findings, though some variability in estimates was observed. Our results do not suggest that systematic avoidance of TNFi is required in all RA-ILD patients. However, given disease heterogeneity and imprecision of our estimates, there may be subpopulations of RA-ILD patients that benefit from specific b/tsDMARD treatment strategies. Our findings highlight the need for randomized controlled trials (RCTs) of DMARDs in RA-ILD to address substantial evidence gaps prohibiting strong recommendations in clinical practice guidelines18 and high-quality, evidence-based treatment.
There have been few comparative studies of treatment with TNFi or alternative DMARDs in RA-ILD. Although findings have been mixed, non-significantly higher rates of adverse outcomes have generally been observed with TNFi.7-11 We did not find significantly different outcomes among RA-ILD patients treated with TNFi vs. non-TNFi, refuting our hypothesis based on these prior studies. Prior studies have been limited by likely misclassification of RA-ILD status, limited confounding control, or lacked important outcomes, such as mortality. We addressed these limitations through the use of an active-comparator, new-user pharmacoepidemiologic design that adhered to Target-Trial-Framework,19 and validated RA-ILD algorithms were used to select the cohorts, with PPVs exceeding 70-80%.23 Further, we used calendar time specific PS-matching to balance the treatment groups, importantly including FVC in PS models. We captured clinically relevant outcomes of death, respiratory death, and respiratory hospitalization. Several secondary and sensitivity analyses supported our primary findings and calculated E values illustrated that marked confounding (e.g., E value >3) would need to be present to produce the effect sizes observed in prior studies finding higher risks with TNFi treatment.7-10
Our study results do not support systematically avoiding TNFi in all RA-ILD patients. This is important as TNFi are the most commonly prescribed bDMARD in RA and are highly efficacious for improving articular disease activity and functional status, independent predictors of survival in RA-ILD.3 The heterogeneity of RA-ILD related to distinct histo-radiologic patterns and varying disease severity must be considered when interpreting and applying study findings. ILD pattern was not available in our study, and the mean pre-treatment FVC of 76% indicates the population had a milder disease severity than some RA-ILD cohorts, such as a population from ILD referral centers.16 While this may raise concerns about the generalizability of findings to more severe ILD populations, approximately 50% of patients had a FVC <80% predicted and epidemiologic studies have observed that most RA-ILD patients have a relatively stable or slowly declining FVC trajectory.28 Because ILD typically develops later in the RA disease course,29 treatment histories may also vary among patients. In a secondary analysis among RA-ILD patients who had previously used b/tsDMARDs, we found non-significantly improved outcomes for non-TNFi. This may reflect different prescription patterns between biologic naïve vs. experienced, predisposition to poorer treatment outcomes when receiving a second similar agent, or could indicate there are specific RA-ILD populations for whom b/tsDMARD treatment choices should be modified. Exploratory subgroup analyses favored TNFi among individuals over 65 years, when TNFi were compared to rituximab, and when analyzed using an as-treated approach. As-treated analyses tend to bias towards more extreme findings, and further studies are needed to rigorously evaluate the underlying reasons for these hypothesis generating findings from subgroup analyses.
Our study was focused primarily on the safety of TNFi and non-TNFi therapies in RA-ILD rather than their efficacy for treating ILD. Efficacy studies typically include additional ILD outcomes measures such as longitudinal FVC, dyspnea measures, and chest CT.16,30 To create a more homogenous population, our eligibility criteria excluded patients who previously received ILD-directed therapies. Alternative eligibility criteria may be needed to evaluate efficacy for RA-ILD. Because RA-ILD is characterized by both pulmonary and articular manifestations, providers may have selected immunomodulatory therapies for their anticipated benefits on articular or pulmonary manifestations. For example, rituximab may have been preferentially selected in progressive RA-ILD while TNFi may have been preferentially selected for articular manifestations. This channeling complicates real-world studies and illustrates the importance of pursuing RCTs of DMARDs in RA-ILD with refined eligibility criteria and comprehensive articular and pulmonary outcome measures.
There are limitations to the study. RA-ILD was determined using administrative algorithms which may result in misclassification and a limited ability to phenotype RA-ILD. Pre-treatment FVC was included, but other factors such as FVC trajectory, ILD pattern, or measures of severity from chest CT were not available. Despite the use of PS methods, there may be residual selection bias or unmeasured confounding, although E values suggest the latter is unlikely to have meaningfully impacted these results. While clinically relevant outcomes were included, we did not have replete post-treatment FVC measures to measure change from baseline FVC, nor other indicators of ILD severity such as chest CT scoring or dyspnea measures. Similarly, RA-related measures such as disease activity and functional status were not available. Specific respiratory causes of hospitalization and death were only descriptively evaluated due to the difficulty in distinguishing between different causes in observational data (e.g., ILD flare vs. lower respiratory tract infection). Even with a large observational data source, the resulting sample sizes after PS-matching were modest and estimates were imprecise. This limited statistical power precluded analyses of individual TNFi and non-TNFi b/tsDMARDs, the performance of additional subgroup comparisons by relevant patient characteristics or treatment strategies (e.g., monotherapy vs. combination), and detection of small differences between therapies. The study population was male predominant with frequent smoking history. While representative of the VA population and established RA-ILD risk factors, these characteristics may limit generalizability to other populations. Finally, patients were not involved in the design or conduct of the study.
In conclusion, we found no significant differences in survival or respiratory-related hospitalization between RA-ILD patients initiating TNFi or non-TNFi b/tsDMARDs. These findings do not support the systematic avoidance of TNFi in the management of RA-ILD patients. RCTs of DMARDs in RA-ILD are critical to generate the evidence needed to inform treatment decisions.
Supplementary Material
RESEARCH IN CONTEXT.
Evidence before this study
There is a paucity of high-quality evidence to guide the selection of disease-modifying anti-rheumatic drugs (DMARDs) in rheumatoid arthritis-associated interstitial lung disease (RA-ILD). A search of PubMed (from inception to December 2023) using the terms “rheumatoid arthritis”, “interstitial lung disease”, “tumor necrosis factor inhibitor”, and “anti-tumor necrosis factor” identified a select number of comparative observational studies suggesting that tumor necrosis factor inhibitor (TNFi) use in RA-ILD may be associated with poorer outcomes compared to other DMARDs, though findings have been mixed and prone to selection and confounding bias.
Added value of this study
This study utilized the Target-Trial-Emulation Framework and pharmacoepidemiologic best practices to compare patient outcomes in RA-ILD following the initiation of TNFi vs. non-TNFi biologic and targeted-synthetic DMARDs (b/tsDMARDs). No significant differences in survival or respiratory-related hospitalization were observed between RA-ILD patients initiating TNFi or non-TNFi b/tsDMARD. Several secondary, sensitivity, and subgroup analyses were performed demonstrating the robustness of these findings.
Implications of all the available evidence
These findings do not suggest that TNFi need to be systematically avoided in all patients with RA-ILD. Given substantial disease heterogeneity in RA-ILD, there may exist subgroups who would benefit from specific b/tsDMARDs. Higher quality evidence for the treatment of RA-ILD needs to be generated in clinical trials to inform these clinical decisions.
ACKNOWLEDGEMENTS
Work supported by Center of Excellence for Suicide Prevention, Joint Department of Veterans Affairs and Department of Defense Mortality Data Repository – National Death Index. Support for VA/CMS data provided by the Department of Veterans Affairs, VA Health Services Research and Development Services, VA Information Resource Center (SDR 02-237 and 98-004).
Funding:
The primary funder of this study was the U.S. Department of Veterans Affairs (IK2 CX002203 to BRE). BRE is supported by a VA CSR&D (IK2 CX002203) and the Rheumatology Research Foundation. JFB is supported by a CSR&D Merit Award (CX001703) and RR&D Merit Award (RX003644). TMJ is supported by the Rheumatology Research Foundation. SMM is supported by the National Institutes of General Medical Sciences (P20 GM130423). JRC is supported by the National Institutes of Arthritis and Musculoskeletal and Skin Diseases (P30 AR072583). TRM is supported by grants from the VA (BLR&D Merit I01 BX004660), National Institutes of Health (2U54GM115458), U.S. Department of Defense (PR200793), and the Rheumatology Research Foundation.
Footnotes
Disclosures: BRE has consulted with and received research support from Boehringer-Ingelheim. TRM has consulted for Horizon Therapeutics, Pfizer, UCB, and Sanofi and receives research support from Horizon. JFB has consulted for CorEvitas, Cumberland Pharma, Formation Bio and has received research support from Horizon. MG has received research support from GlaxoSmithKline, Janssen, and Pfizer. JC has received research grants and consulting monies from Amgen, Abbvie, BMS, Lilly, Novartis, Pfizer, and Sanofi.
The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States government.
DATA SHARING
Data will be made available upon reasonable request to the corresponding author (Bryant.england@unmc.edu) after obtainment of necessary regulatory approvals, including data use agreements and institutional review board approvals. Restrictions apply to the availability of these data in order to protect the privacy of patients.
REFERENCES
- 1.Bongartz T, Nannini C, Medina-Velasquez YF, et al. Incidence and mortality of interstitial lung disease in rheumatoid arthritis: a population-based study. Arthritis Rheum 2010; 62(6): 1583–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Gabbay E, Tarala R, Will R, et al. Interstitial lung disease in recent onset rheumatoid arthritis. Am J Respir Crit Care Med 1997; 156(2 Pt 1): 528–35. [DOI] [PubMed] [Google Scholar]
- 3.Brooks R, Baker JF, Yang Y, et al. The impact of disease severity measures on survival in U.S. veterans with rheumatoid arthritis-associated interstitial lung disease. Rheumatology (Oxford) 2022; 61(12): 4667–77. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Hadjinicolaou AV, Nisar MK, Bhagat S, Parfrey H, Chilvers ER, Ostor AJ. Non-infectious pulmonary complications of newer biological agents for rheumatic diseases--a systematic literature review. Rheumatology (Oxford) 2011; 50(12): 2297–305. [DOI] [PubMed] [Google Scholar]
- 5.Roubille C, Haraoui B. Interstitial lung diseases induced or exacerbated by DMARDS and biologic agents in rheumatoid arthritis: a systematic literature review. Semin Arthritis Rheum 2014; 43(5): 613–26. [DOI] [PubMed] [Google Scholar]
- 6.Jani M, Hirani N, Matteson EL, Dixon WG. The safety of biologic therapies in RA-associated interstitial lung disease. Nat Rev Rheumatol 2014; 10(5): 284–94. [DOI] [PubMed] [Google Scholar]
- 7.Nakashita T, Ando K, Kaneko N, Takahashi K, Motojima S. Potential risk of TNF inhibitors on the progression of interstitial lung disease in patients with rheumatoid arthritis. BMJ Open 2014; 4(8): e005615. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Druce KL, Iqbal K, Watson KD, Symmons DPM, Hyrich KL, Kelly C. Mortality in patients with interstitial lung disease treated with rituximab or TNFi as a first biologic. RMD Open 2017; 3(1): e000473. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Dixon WG, Hyrich KL, Watson KD, et al. Influence of anti-TNF therapy on mortality in patients with rheumatoid arthritis-associated interstitial lung disease: results from the British Society for Rheumatology Biologics Register. Ann Rheum Dis 2010; 69(6): 1086–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Kang EH, Jin Y, Desai RJ, Liu J, Sparks JA, Kim SC. Risk of exacerbation of pulmonary comorbidities in patients with rheumatoid arthritis after initiation of abatacept versus TNF inhibitors: A cohort study. Semin Arthritis Rheum 2020; 50(3): 401–8. [DOI] [PubMed] [Google Scholar]
- 11.Curtis JR, Sarsour K, Napalkov P, Costa LA, Schulman KL. Incidence and complications of interstitial lung disease in users of tocilizumab, rituximab, abatacept and anti-tumor necrosis factor alpha agents, a retrospective cohort study. Arthritis Res Ther 2015; 17: 319. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Md Yusof MY, Kabia A, Darby M, et al. Effect of rituximab on the progression of rheumatoid arthritis-related interstitial lung disease: 10 years' experience at a single centre. Rheumatology (Oxford) 2017; 56(8): 1348–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Fernández-Díaz C, Castañeda S, Melero-González RB, et al. Abatacept in interstitial lung disease associated with rheumatoid arthritis: national multicenter study of 263 patients. Rheumatology 2020; 59(12): 3906–16. [DOI] [PubMed] [Google Scholar]
- 14.Manfredi A, Cassone G, Furini F, et al. Tocilizumab therapy in rheumatoid arthritis with interstitial lung disease: a multicenter retrospective study. Intern Med J 2019. [DOI] [PubMed] [Google Scholar]
- 15.Mena-Vázquez N, Godoy-Navarrete FJ, Manrique-Arija S, et al. Non-anti-TNF biologic agents are associated with slower worsening of interstitial lung disease secondary to rheumatoid arthritis. Clinical Rheumatology 2021; 40(1): 133–42. [DOI] [PubMed] [Google Scholar]
- 16.Matson SM, Baqir M, Moua T, et al. Treatment Outcomes for Rheumatoid Arthritis-Associated Interstitial Lung Disease: A Real-World, Multisite Study of the Impact of Immunosuppression on Pulmonary Function Trajectory. Chest 2023; 163(4): 861–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Holroyd CR, Seth R, Bukhari M, et al. The British Society for Rheumatology biologic DMARD safety guidelines in inflammatory arthritis. Rheumatology 2018; 58(2): e3–e42. [DOI] [PubMed] [Google Scholar]
- 18.Johnson SR, Bernstein EJ, Bolster MB, et al. 2023 American College of Rheumatology (ACR)/American College of Chest Physicians (CHEST) Guideline for the Treatment of Interstitial Lung Disease in People with Systemic Autoimmune Rheumatic Diseases. Arthritis Rheumatol 2024. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Hernan MA, Robins JM. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available. Am J Epidemiol 2016; 183(8): 758–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Johnson TM, Mahabir CA, Yang Y, et al. Aortic Stenosis Risk in Rheumatoid Arthritis. JAMA Intern Med 2023; 183(9): 973–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Langan SM, Schmidt SA, Wing K, et al. The reporting of studies conducted using observational routinely collected health data statement for pharmacoepidemiology (RECORD-PE). BMJ 2018; 363: k3532. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Vandenbroucke JP, von Elm E, Altman DG, et al. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. Ann Intern Med 2007; 147(8): W163–94. [DOI] [PubMed] [Google Scholar]
- 23.England BR, Roul P, Mahajan TD, et al. Performance of Administrative Algorithms to Identify Interstitial Lung Disease in Rheumatoid Arthritis. Arthritis Care Res (Hoboken) 2020; 72(10): 1392–403. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Sauer BC, Teng CC, Tang D, et al. Persistence With Conventional Triple Therapy Versus a Tumor Necrosis Factor Inhibitor and Methotrexate in US Veterans With Rheumatoid Arthritis. Arthritis Care Res (Hoboken) 2017; 69(3): 313–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.England BR, Roul P, Yang Y, et al. Extracting forced vital capacity from the electronic health record through natural language processing in rheumatoid arthritis-associated interstitial lung disease. Pharmacoepidemiol Drug Saf 2023. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Austin PC. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behav Res 2011; 46(3): 399–424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.VanderWeele TJ, Ding P. Sensitivity Analysis in Observational Research: Introducing the E-Value. Ann Intern Med 2017; 167(4): 268–74. [DOI] [PubMed] [Google Scholar]
- 28.Chang SH, Lee JS, Ha Y-J, et al. Lung function trajectory of rheumatoid arthritis–associated interstitial lung disease. Rheumatology 2023; 62(9): 3014–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Mohning MP, Amigues I, Demoruelle MK, et al. Duration of rheumatoid arthritis and the risk of developing interstitial lung disease. ERJ Open Research 2021; 7(1): 00633–2020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Solomon JJ, Danoff SK, Woodhead FA, et al. Safety, tolerability, and efficacy of pirfenidone in patients with rheumatoid arthritis-associated interstitial lung disease: a randomised, double-blind, placebo-controlled, phase 2 study. Lancet Respir Med 2023; 11(1): 87–96. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data will be made available upon reasonable request to the corresponding author (Bryant.england@unmc.edu) after obtainment of necessary regulatory approvals, including data use agreements and institutional review board approvals. Restrictions apply to the availability of these data in order to protect the privacy of patients.