This retrospective study highlights a higher risk of all-cause mortality among ABA users compared to TNFi users in the management of patients diagnosed with RA-ILD. Despite adjusting for varying covariates, most models consistently indicated a higher risk of mortality in the ABA cohort compared to the TNFi. We considered the results difficult to elucidate, leading us to reconsider the role of tumor necrosis factor (TNF) in the pathogenesis of RA-ILD. Several evidence and animal studies showed that TNF play a major role in fibrosis progression [15, 16].

Our findings reveal a higher risk of all-cause mortality in ABA users compared to TNFi users, particularly in patients without cardiovascular risk factors; while, this was not statistically significance in patients with cardiovascular risk factors. Notably, a recent study indicates that ABA users are associated with reduced cardiovascular risk compared to TNFi users [13]. Additionally, an observational study revealed a slight reduction in low-density lipoprotein levels in ABA users when compared to TNFi users [17]. Therefore, ABA may potentially offer more favorable outcomes in patients with cardiovascular risk factors. This could explain why our study found a higher risk of all-cause mortality in ABA users without cardiovascular risk factors; whereas, the risk in patients with these factors was similar to that of TNFi users.

Our study demonstrates no significant difference in the risk of requiring ventilation between ABA and TNFi overall. However, subgroup analysis indicates a higher risk of requiring mechanical ventilation in the ABA users with aged 18–64 years. This suggests that age may indeed be a significant factor, with ABA may potentially offering relatively more benefits on older patients. Another subgroup analysis indicates a higher risk of mechanical ventilation in patients with cardiovascular risk factors beyond the disease itself. These findings underscore the importance of careful consideration in using ABA, especially in younger to middle-aged patients with cardiovascular risk factors and concurrent diseases that heighten the risk of respiratory distress.

ABA was approved for the management of RA in 2005, and its mechanism of action is significantly distinct from TNFi [18]. Previous studies revealed ABA provides preferable results in disease control among the ACPA-positive patients [19, 20]. Furthermore, ABA has been associated with a reduction in ACPA and rheumatoid factor (RF) titers, both known risk factors for RA-ILD [21].

Several studies have identified key risk factors for the progression and poor prognosis of RA-ILD, including male sex, high ACPA level, higher DAS-28-ESR score, and UIP pattern [8, 22,23,24]. Furthermore, risk factors for the development of RA-ILD include older age, male sex, smoking, and high titers of ACPA or RF [9]. Given these factors, along with ABA’s potential benefits for ACPA-positive patients and its role in reducing antibody production, ABA may offer more advantages than TNFi in patients with adverse prognostic factors.

In a previous retrospective cohort study, ABA demonstrated a higher clinical remission rate and drug survival rate among ACPA-positive patients compared to ACPA-negative ones in patients diagnosed with RA [25]. However, another study found no preferable effects on ACPA-positive patients [26]. In a previous large cohort study in Japan, ABA reached the higher drug retention rate compared to TNFi [27]. With limited evidences, a recent consensus recommends considering ABA as a first-line treatment option for RA-ILD [28].

Previous studies have been primarily focused on the effects of positive RF or ACPA on the efficacy of ABA. Consequently, our research included a subgroup analysis to explore the potential impact of ACPA autoantibodies on the management of RA-ILD. The findings revealed a higher risk of all-cause mortality with ABA compared to TNFi among the ACPA-positive patients with RA-ILD. However, our study could not evaluate the severity of ILD, a factor that may influence clinicians' choice of treatment. Recent consensus and evidence have suggested a preference for ABA over TNFi in RA-ILD cases, especially considering that positive-ACPA status is associated with a poorer prognosis and disease progression. Therefore, the results of our analysis should be interpreted with caution.

A recent study based on the US MarketScan database have shown that, in patients with RA and chronic obstructive pulmonary disease, no significant difference in lung infection rates between those treated with abatacept and other biologics was observed [29]. One study found that the infection rate in the abatacept group was lower compared to the adalimumab group, although this difference did not reach statistical significance. However, when compared to infliximab, a statistically significant lower infection rate was observed in the abatacept group [30]. A recent post marketing analysis also concluded that there is no significant difference in infection rates between ABA and other biologics [31].

Additionally, data from a Japanese database study indicated that older RA patients treated with abatacept had a lower infection risk compared to those treated with adalimumab [32]. However, this trend was not observed in patients under 65 years of age. This suggests that the lower infection incidence associated with ABA may be related to patient age. In our study, the average age of patients after propensity score matching was around 62 years, which may explain the differences in outcomes observed.

In previous studies focusing on the benefits and harms of TNFi in RA-ILD, the British Society for Rheumatology Biologics Register (BSRBR) study revealed that while the all-cause mortality rate was not higher, the mortality rate due to RA-ILD was indeed higher in the TNFi group than traditional DMARDs [33]. This finding is particularly noteworthy given that only 3% of RA patients had baseline ILD and the events were fewer; whereas, contemporary research observed a lifetime risk of developing ILD to be 7.7% [34]. This discrepancy may introduce baseline characteristic mismatches and selection biases, potentially leading to an overestimation of ILD-related mortality in the TNFi group. In the German RABBIT registry research, patients treated with TNFi (n = 114) and ABA (n = 60) were followed for a median duration of 2.6 years. Our Kaplan–Meier analysis indicated a widening mortality gap beyond 2 years. The RABBIT study found adjusted hazard ratios (aHRs) of less than 1 for all DMARD comparisons involving TNFi, although these results lacked statistical significance. Notably, when ABA and TNFi were directly compared, no significant statistical differences were observed. Statistical significance was only detected when using B-cell biological DMARDs as the reference [35]. A previous study from BSRBR also demonstrated the benefit of Rituximab compared to TNFi [36]. These findings suggest that longer follow-up periods may be necessary to fully elucidate the comparative benefits and harms of ABA and TNFi in the treatment of RA-ILD. In a database study using the Optum Clinformatics Data Mart, the incidence rate of ILD was found to be higher in RA-ILD patients treated with ABA (4.46 per 1000 person-years) compared to those treated with adalimumab (3.43 per 1000 person-years). Although it is not possible to directly equate a higher incidence of RA-ILD with a higher mortality risk, these findings suggest that the clinical use of ABA should be approached with caution compared to adalimumab. This may align with our results, which also suggest a potentially higher risk associated with ABA treatment in RA-ILD patients [37].

In our study, the different results from previous studies may need to be clarified. In our subgroup analysis, positive-ACPA and cardiovascular risk significantly influenced the outcomes. In a direct comparison with TNFi, a Japanese study showed that ABA was associated with a lower risk of ILD progression. Compared to our study, their patient population was generally older. Our subgroup analysis also revealed that patients over 65 years old did not experience worse outcomes with ABA compared to TNFi [2]. Another database study that included a cohort of RA-ILD patients had fewer ABA users, focusing on ILD progression as an outcome and the events about inpatient exacerbation are few [38]. In contrast, our study included a larger number of ABA-treated patients and assessed a broader range of outcomes, including all-cause mortality, mechanical ventilation, and critical services, beyond just ILD progression. It is important to note that many studies on ABA did not include a TNFi control group, often evaluating the improvement or stabilization rates in a single-arm design. This methodological difference can make direct comparisons with our study challenging [39,40,41,42]. A recent large observational multicenter study there is still near one fourth patients treated with ABA presented progression of RA-ILD [43]. This highlights the complexity and challenges in managing RA-ILD with ABA.

The reasons for these differing outcomes could include variations in patient demographics, disease severity, and the specific endpoints assessed. Our study utilized a real-world database with a diverse patient population, potentially capturing a wider range of clinical scenarios. Additionally, the mechanisms by which ABA and TNFi influence RA-ILD may differ, particularly in patients with varying comorbidities and disease profiles. These factors combined could explain the observed differences in outcomes, underscoring the need for further research to elucidate these findings.

Our study has several strengths. Firstly, the use of the TriNetX database, one of the largest electronic medical record databases, enhances the statistical power and representativeness of our findings. Secondly, we conducted a comprehensive assessment of outcomes, including all-cause mortality, mechanical ventilation, and critical care service utilization, which provides a thorough evaluation of potential complications in RA-ILD. Thirdly, our detailed subgroup analysis, considering factors such as age, cardiovascular risk and ACPA status, offers deeper insights into the drug's performance across different patient populations.

However, our study also has several limitations. Firstly, despite adjusting for numerous covariates, potential confounding factors may still influence the results in this retrospective study. Secondly, we could not evaluate the severity and type of ILD on lung function or imaging due to insufficient data. We have endeavored to use matched corticosteroids, mycophenolic acid, and other medications that are relatively more likely to be effective for RA-ILD to ensure that the severity of both groups is as similar as possible. Thirdly, the study lacks information on DAS-28 or other scores to evaluate disease activity. Fourthly, as most of our study participants are Caucasian, applying these findings to other ethnic groups may have potential limitations. Fifthly, the inability to retrieve the cause of mortality or mechanical ventilation from the database, this may lead us several challenge in interpreting results and further correlating them with clinical situations. Although the reasons for mechanical ventilation were not entirely due to the exacerbation or acute episode of ILD, we hope that this information can provide a comprehensive comparison of the overall impact on patients in both groups, not just ILD. Sixthly, previous evidence demonstrated ABA may be a better choice in severe RA-ILD, this may lead selection bias in this study. We try to utilize the laboratory and image examination data to partially represent the disease activity or type of ILD. However, we found that there was insufficient data available for analysis in the setting. Therefore, we made every effort to match the covariates as accurately as possible to mitigate the bias.