Additional Considerations on the Comparative Effectiveness of <scp>TNF</scp> ‐α and <scp>IL</scp> ‐6 Inhibitors on Bone Health and Mortality in Rheumatoid Arthritis

We read with great interest the retrospective cohort study by Wang et al. comparing the effects of TNF-α and IL-6 inhibitors on bone health outcomes and all-cause mortality in patients with rheumatoid arthritis (RA) [1]. The authors demonstrated that both inhibitor classes exhibit comparable efficacy in reducing the risks of osteoporosis and fractures, while TNF-α inhibitors may offer advantages in reducing all-cause mortality. The study provides valuable insights into the comparative effectiveness of these biologic agents; however, several limitations warrant further discussion to enhance the robustness of future research. First, the authors note that IL-6 inhibitors were often administered as second-line therapy following TNF-α inhibitor failure, a scenario that may introduce immortal time bias [1]. This bias arises because the “immortal” period between RA diagnosis and initiation of IL-6 inhibitor therapy could conflate treatment effects with underlying disease progression. To address this concern, future studies could employ marginal structural models (MSMs) to account for time-dependent confounding and treatment sequencing [2]. MSMs weight observations according to the probability of receiving a specific treatment at each time point, thereby adjusting for prior treatment failures and fluctuations in disease severity [3]. Alternatively, propensity score stratification based on treatment line (first-line vs. second-line) could help isolate the true effect of IL-6 inhibitors, independent of previous TNF-α inhibitor exposure. Second, the authors appropriately highlight that the TriNetX database, not being population-based, may be susceptible to loss to follow-up and misclassification [1]. These issues could compromise the generalizability of the findings, as the database likely overrepresents patients with higher healthcare engagement. To mitigate these limitations, we recommend implementing enhanced data quality verification procedures, such as cross-referencing ICD-10 codes for fractures and osteoporosis with corresponding imaging reports (e.g., radiographs) to minimize diagnostic misclassification. Additionally, conducting sensitivity analyses—either by excluding individuals with insufficient follow-up (e.g., < 6 months) to evaluate the consistency of the results, or by applying multiple imputation techniques for missing data—would strengthen the reliability of the conclusions. Moreover, we wish to emphasize several additional considerations that may further contextualize the findings and inform subsequent investigations. The dependence on ICD-coded diagnoses for osteoporosis and fractures fails to capture subclinical bone changes. Future studies would benefit from incorporating serial bone mineral density measurements (e.g., via DXA) and biomarkers of bone turnover (e.g., CTX-1, PINP) to better quantify drug-specific effects on bone remodeling [4, 5]. The notable reduction in mortality associated with TNF-α inhibitors also merits exploration into cause-specific mortality (e.g., cardiovascular, infectious, or malignancy-related) to elucidate potential mechanisms. Finally, although propensity score methods adjusted for baseline use of bone-protective agents (e.g., bisphosphonates), prospective monitoring of osteoporosis medications throughout the follow-up period is necessary to disentangle the effects of biologic agents from those of concurrent therapies. The study by Wang et al. [1] represents a meaningful advancement in our understanding of biologic therapies in RA. Addressing these methodological considerations would further reinforce causal inferences. Such refinements will contribute to more robust evidence, ultimately guiding personalized treatment strategies that optimize joint and systemic outcomes—including bone integrity and survival—in patients with RA. We thank the authors for their valuable contribution and hope these suggestions stimulate further scholarly discussion. Yunxi He and Yueyun Qiu wrote the original draft. Yiyan Zhang revised the manuscript and served as the corresponding author for publication. The authors declare no conflicts of interest. The authors have nothing to report.