This was a retrospective comparative cohort study. This design was informed by the previous literature. The risk of COPD was compared between patients newly diagnosed with RA and the individuals without RA.

The dataset utilized in this study was collected on October 9, 2023, from the TriNetX Research Network. TriNetX is a federated health research network that aggregates longitudinal electronic health records from 68 healthcare organizations. TriNetX, LLC adheres to the Health Insurance Portability and Accountability Act (HIPAA), a U.S. federal law designed to protect the privacy and security of healthcare data, along with other relevant data privacy regulations applicable to the contributing healthcare organizations. TriNetX is ISO 27001:2013 certified and maintains an Information Security Management System (ISMS) to ensure the protection of healthcare data and compliance with the HIPAA Security Rule. All data displayed on the TriNetX Platform in aggregate form or included in any patient-level data set generated by the platform is de-identified according to the de-identification standards outlined in Section §164.514(a) of the HIPAA Privacy Rule. The de-identification process is verified through a formal determination by a qualified expert as defined in Section §164.514(b)(1) of the HIPAA Privacy Rule. Consequently, as this study involved only de-identified patient records and did not include the collection, use, or transmission of individually identifiable data, it was exempt from Institutional Review Board (IRB) approval. Additionally, this study was conducted in accordance with the principles outlined in the Declaration of Helsinki. The use of TriNetX in this research was approved by the Institutional Review Board of Chung Shan Medical University Hospital (CSMUH), under the approval number CS2-21176.

TriNetX is a platform that de-identifies and aggregates electronic health record (EHR) data, most of which are large academic medical institutions with both inpatient and outpatient facilities at multiple locations, across 50 states in the US. TriNetX Analytics provides web-based and secure access to patient EHR data from hospitals, primary care, and specialty treatment providers, accessible data from the platform include demographics, diagnoses, medications, laboratory values, and procedures. The contributing EHR systems used United Medical Language System biomedical ontologies for coding. TriNetX maps the data to a standard and controlled set of clinical terminologies, for example, mapping disease terms from SNOMED-CT to ICD-10, drug terms from NDCs to RxNorm. The data is then transformed into a proprietary data schema. This transformation process includes an extensive data quality assessment that includes data cleaning (10).

To ensure sufficient data coverage for baseline covariates, we required that all included patients had at least one recorded encounter within 6 months prior to their index date. Patients without records in this look-back window were excluded. Additionally, most TriNetX sites have longitudinal EHR data dating back to January 1, 2000, which ensured that patients diagnosed between 2010 and 2021 had an equal opportunity to contribute covariate data for the baseline period.

Patients with RA and non-RA were identified from the TriNetX network between 2010 and 2021. Exclusion criteria and the steps of cohort refinement and propensity score matching are shown in the diagram ( Figure 1 ).

We identified patients diagnosed with rheumatoid arthritis (ICD-10: M05-M05.9, M06.0, M06.8, and M06.9) (11) from 2010 to 2021, excluding prevalent RA cases with visits prior to 2010 to ensure incident cases only. The index date was defined as the first recorded RA diagnosis. To establish a clean cohort for assessing COPD risk, we excluded patients with: (1) pre-existing COPD to focus on incident cases; (2) other systemic autoimmune diseases (SLE, systemic sclerosis, Sjögren syndrome, polymyositis, or dermatopolymyositis) due to their overlapping pulmonary manifestations with RA that could confound results; and (3) deaths within 3 months post-index date to avoid immortal time bias. After these exclusions, we selected 136,821 RA patients with confirmed disease activity (evidenced by an RA revisit within 3 months post-diagnosis) for analysis ( Figure 1 ).

Individuals who were never diagnosed with RA at any time on or before 2021 were included. These patients had medical visits between 2010 and 2021. The index date for the non-RA cohort was defined as the first recorded medical visit within this period. After excluding any diagnosis of COPD, SLE, systemic sclerosis, Sjögren syndrome, polymyositis, dermatopolymyositis, or death prior to the index date and within 3 months after the index date, we selected 8,108,944 non-RA patients who had any medical encounter within 3 months after the index date ( Figure 1 ).

The study investigated the 5-year risk of newly onset COPD (ICD-10: J41-J44) among RA patients and non-RA patients. For each participant, follow-up began at the index date and ended at the earlier of the occurrence of the study outcomes or censoring at the last recorded fact within the time window in the patient’s record.

To reduce potential misclassification of non-RA individuals due to incomplete electronic health record (EHR) data, we implemented several design strategies. First, we required that all patients—both RA and non-RA—had at least one medical encounter within 3 months after the index date, ensuring that all included individuals were actively engaged in care during the cohort entry period. Second, patients who had any diagnosis of RA, COPD, systemic lupus erythematosus, systemic sclerosis, Sjögren syndrome, polymyositis, or dermatomyositis either prior to or within 3 months after the index date were excluded to improve the specificity of cohort assignment. Lastly, for all patients, follow-up was censored at the last recorded clinical encounter to mitigate misclassification bias due to loss to follow-up or incomplete data.

To account for potential confounders, our study included covariates assessed on the index date or within 6 months prior to the index date. These covariates included demographic details (age, sex, race, and socioeconomic status), baseline medical utilization (including inpatient encounters and emergency visits), and lifestyle factors such as nicotine dependence (ICD-10: F17) as an indicator of smoking habits. Additionally, we considered comorbidities including hypertensive diseases, dyslipoproteinemia, diabetes mellitus, neoplasms, overweight or obesity, sleep disorders, ischemic heart diseases, chronic kidney disease, and cerebrovascular diseases. Furthermore, baseline medications were considered, including systemic corticosteroids, systemic nonsteroidal anti-inflammatory drugs (NSAIDs), topical corticosteroids, topical products for joint and muscular pain, antipruritics, psycholeptics, psychoanaleptics, systemic antibacterials, systemic antimycotics, beta-blocking agents, calcium channel blockers, diuretics, and RA-related medications. Laboratory results such as rheumatoid factor (RF), anti-cyclic citrullinated peptide antibody (Anti-CCP), C-reactive protein (CRP), and erythrocyte sedimentation rate (ESR) were also included in the analysis.

Patient cohorts were propensity-score matched for baseline covariates, including demographics, medical utilization (inpatient encounters and emergency visits), lifestyle factors, and comorbidities. Propensity scores were estimated using multivariable logistic regression, with the dependent variable indicating RA status. Matching was performed using 1:1 greedy nearest-neighbor matching without replacement, applying a caliper width of 0.1 standard deviations of the logit of the propensity score to ensure similarity between matched individuals.

To minimize potential overadjustment, co-medications and laboratory values were excluded from the matching process, but were included in downstream multivariable analyses. The matching algorithm was executed on a centrally pooled covariate matrix aggregated from all participating healthcare organizations in the TriNetX network. To reduce potential bias introduced by record ordering in the matrix, rows were randomized before matching. After matching, 136,820 pairs of RA and non-RA patients were selected for comparative analysis ( Figure 1 ).

Time-to-event analyses were conducted using Cox proportional hazards regression to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the development of COPD. The proportional hazards assumption was tested using the generalized Schoenfeld residuals approach. Kaplan–Meier estimators were used to visualize cumulative incidence of COPD, with censoring applied at each patient’s last recorded clinical encounter.

Laboratory covariates with missing data were handled using complete case analysis (listwise deletion); no imputation was performed. All statistical tests were two-sided, with statistical significance set at p < 0.05. Analyses were conducted on the TriNetX Analytics platform, which applies standard statistical routines including R’s Survival package (v3.2-3) (12).

We also conducted subgroup analyses to evaluate heterogeneity of COPD risk across strata defined by sex, age, and race. Among RA patients, additional stratified analyses were performed by rheumatoid factor level (25–49.9, 50–99.9, ≥100 vs. <25), anti-CCP level (20–39.9, 40–59.9, ≥60 vs. <20), C-reactive protein (CRP) level (1–14.9, ≥15 vs. <1), erythrocyte sedimentation rate (ESR) level (>20 vs. ≤20), and baseline use (yes vs. no) of DMARDs, systemic corticosteroids, and NSAIDs.

To ensure the reproducibility and transparency of the study, we provide a detailed description of the specific analytical modules and workflow used within the TriNetX Analytics Platform

Baseline demographic and clinical characteristics of the propensity-score-matched (PSM) RA and non-RA cohorts were compared using the “Compare Characteristics” module within the TriNetX platform ( Table 1 ). This function applies standardized statistical routines to generate descriptive summaries and event estimates. Cumulative incidence of COPD at 1, 3, and 5 years was calculated using Kaplan-Meier survival analysis, and hazard ratios (HRs) with 95% confidence intervals were derived using Cox proportional hazards models. The models were adjusted for propensity score matching and stratified by key demographic subgroups (e.g., sex, age, race). The number of events and total sample size are presented in Table 1 .

To assess the incidence of COPD and estimate hazard ratios between matched cohorts, the “Compare Outcomes” module was applied. The workflow was as follows: “Compare Outcomes”→”How do outcomes compare between cohorts?”→”Index Event.”Kaplan–Meier survival analysis and Cox proportional hazards modeling were performed. The hazard ratios and 95% confidence intervals were calculated using the survival package (v3.2-3) embedded in the TriNetX Analytics Platform ( Figures 2 , 3 ).

The results presented in Figure 3 were derived from a single multivariable Cox proportional hazards model. This model included adjustment for demographics (age, sex, race, and socioeconomic status), baseline medical utilization (inpatient and emergency visits), smoking status, comorbidities, co-medications, and laboratory values. Patients with missing laboratory data were excluded via complete case analysis; no imputation was performed.

To obtain laboratory values at specified time points after the index event, we used the “Advanced Explore Cohort” module. The workflow was: “Index Event”→”Explore Characteristics”→”Explore Outcomes,” with appropriate time filters applied to extract laboratory data at 1, 3, and 5 years post-RA diagnosis. Only patients with complete laboratory results at each time point were included in the descriptive analysis.

All data extractions and statistical analyses were conducted exclusively within the TriNetX Analytics Platform between October 9 and October 20, 2023. Propensity score matching, censoring, and statistical testing were automatically executed by the platform based on the defined cohorts and time windows. In accordance with the TriNetX data use policy, no patient-level data were downloaded or exported during this study.

This study initially identified 181,747 patients diagnosed with RA and 8,673,811 individuals never diagnosed with RA from 2010 to 2021. After excluding ineligible participants and performing propensity score matching, there were 136,820 pairs of RA and non-RA patients for analysis ( Figure 1 ). Before propensity score matching, in the RA cohort, the mean age at the index date was 58.3 ± 15.6 years, and 73.9% were female. Most baseline characteristics were imbalanced between the RA cohort and the non-RA cohort, as defined by a standardized mean difference (SMD) > 0.1. After propensity score matching, the baseline demographics, medical utilization (inpatient encounters and emergency visits), lifestyle factors, and comorbidities were balanced between the RA cohort and the non-RA cohort ( Table 2 ).

Table 1 details the risk of COPD within 5 years after the index date in the RA and non-RA cohorts. For the 136,820 pairs of propensity score matching RA and non-RA cohorts, there were 6,761 COPD cases in the RA cohort and 5,832 COPD cases in the non-RA cohort during the 5-year observation period. In the RA cohort, the cumulative probability of COPD was 1.60% (95% CI = 1.53%-1.67%) at 1 year, 4.61% (95% CI = 4.49%-4.74%) at 3 years, and 7.36% (95% CI = 7.18%-7.53%) at 5 years. In the non-RA cohort, the cumulative probability of COPD was 1.29% (95% CI = 1.22%-1.35%) at 1 year, 3.90% (95% CI = 3.79%-4.02%) at 3 years, and 5.97% (95% CI = 5.82%-6.12%) at 5 years. The hazard ratio of COPD was 1.228 (95% CI = 1.186-1.272) in the RA cohort compared with the non-RA cohort. For subgroup analysis, the hazard ratio of COPD in the RA cohort was 1.300 (95% CI = 1.219-1.388) in the male subgroup, 1.258 (95% CI = 1.205-1.313) in the female subgroup, 1.560 (95% CI = 1.464-1.662) in the age <60 years subgroup, 1.302 (95% CI = 1.256-1.350) in the age ≥60 years subgroup, 1.313 (95% CI = 1.258-1.370) in the White race subgroup, and 1.132 (95% CI = 1.039-1.233) in the non-White race subgroup. Furthermore, the Kaplan-Meier method showed that the RA cohort was associated with a significantly higher risk of cumulative incidence of COPD in the full cohort (Log-rank p<0.001; Figure 2A ), male subgroup (Log-rank p<0.001; Figure 2B ), female subgroup (Log-rank p<0.001; Figure 2C ), age <60 years subgroup (Log-rank p<0.001; Figure 2D ), age ≥60 years subgroup (Log-rank p<0.001; Figure 2E ), White race subgroup (Log-rank p<0.001; Figure 2F ), and non-White race subgroup (Log-rank p=0.005; Figure 2G ).

Figure 3 shows the hazard ratio of COPD for various COPD-associated factors, including sex, age, race, rheumatoid factor, anti-CCP, C-reactive protein, erythrocyte sedimentation rate, DMARDs, systemic corticosteroids, and systemic NSAIDs. These hazard ratios are adjusted for demographics (age, sex, race, and socioeconomic status), medical utilization (inpatient encounters and emergency visits), smoking status, comorbidities, co-medications, and laboratory examination data. Compared with females, males had a 15.9% (95% CI = 8.8%-23.5%) increased risk of COPD. Compared with RA patients aged <60 years, those aged ≥60 years had a 32.8% (95% CI = 25.9%-40.1%) increased risk of COPD. Compared with a rheumatoid factor <25, a rheumatoid factor ≥100 increased the risk of COPD by 23.2% (95% CI = 2.4%-48.2%). Compared with an erythrocyte sedimentation rate ≤20, an erythrocyte sedimentation rate >20 increased the risk of COPD by 22.1% (95% CI = 11.9%-33.2%). RA patients receiving DMARDs treatment had a decreased COPD risk (HR = 0.824, 95% CI = 0.776-0.875) and systemic NSAIDs treatment (HR = 0.855, 95% CI = 0.789-0.926). However, systemic corticosteroid treatment increased the risk of COPD (HR = 1.120, 95% CI = 1.054-1.190).