At baseline, a total of 6769 participants diagnosed with RA were assessed. After excluding 70 with prevalent ILD, 2150 with missing data on lipid traits, 319 with missing INFLA-score-related data, and one with missing data on medication records, a total of 4229 participants were included in the analysis (Supplementary Fig. S1). Baseline characteristics of the study participants are summarized in Table 1. The baseline characteristics of participants grouped by sex are presented in Supplementary Table S3. Of the included participants, 2,918 (69.000%) were women, and the mean age was 59.135 ± 7.193 years. A total of 152 cases (3.594%) of incident ILD were recorded, and the mean follow-up time was 14.736 ± 3.263 years. Compared to participants who eventually developed ILD, those without ILD were more likely to be younger, female, and have no history of smoking or alcohol use, and to have a higher education level at baseline. Participants who developed ILD were more likely to use statin (29.605% vs. 20.309%, p = 0.007) and to receive any RA medication, including corticosteroids and DMARDs (67.105% vs. 44.837%, p < 0.001), compared to those without ILD.

Associations between each lipid trait and the incidence of ILD in patients with RA are shown in Table 2. ApoA demonstrated a statistically significant association with ILD risk in the model adjusted for age and sex (HR 0.366, 95% CI 0.182–0.737, p = 0.005). Although the effect was slightly attenuated after further adjusting for BMI, ethnicity, education level, socioeconomic status, smoking status, use of statin, and use of RA medication, the protective effect remained significant, with each standard deviation (SD) increase in ApoA associated with a 61.5% risk reduction in ILD (HR 0.385, 95% CI 0.189–0.783, p = 0.008). HDL-C also showed a significant protective effect in the minimally adjusted model (HR 0.494, 95% CI 0.291–0.836, p = 0.009). The protective effect remained significant in the fully adjusted model, with each SD increase in HDL-C associated with a 51.3% risk reduction in ILD (HR 0.487, 95% CI 0.280–0.847, p = 0.011). Additionally, an increased level of triglycerides was associated with a significantly higher risk of incident ILD in the minimally adjusted model (HR 1.162, 95% CI 1.008–1.340, p = 0.039), though this effect was attenuated to non-significant after adjusting for covariates. Conversely, lipoprotein A demonstrated a significant negative association in the fully adjusted model (HR 0.996, 95% CI 0.992–1.000, p = 0.033), but this association was not observed in the minimally adjusted model. No other lipid traits, including ApoB and LDL-C, demonstrated statistically significant associations with ILD risk in either model specification, suggesting that no association was observed between these lipid traits and ILD risk in patients with RA.

There was limited evidence for a sex difference in the association between lipid traits and the risk of incident ILD in patients with RA. Sex-specific associations between lipid traits and ILD are presented in Fig. 1A. When stratified by sex, a higher level of ApoA was significantly associated with a lower risk of ILD in male patients with RA (HR 0.243, 95% CI 0.079–0.750, p = 0.014), whereas no significant association was detected in female patients with RA (HR 0.525, 95% CI 0.206–1.335, p = 0.176). Furthermore, a higher level of HDL-C was significantly associated with a lower risk of incident ILD in male patients (HR 0.362, 95% CI 0.146–0.894, p = 0.028), whereas this association was not observed in female patients with RA (HR 0.594, 95% CI 0.290–1.216, p = 0.154). No association was found between other lipid traits (including ApoB, LDL-C, and lipoprotein A) and ILD in both male and female patients with RA. These results indicated that the association between lipid traits (particularly for ApoA and HDL-C) and incident ILD showed a clear and significant separation in male and female patients with RA.

Forest plot for multivariate-adjusted hazard ratios for incident ILD in patients with RA per standard deviation increase in each lipid trait, stratified by sex (A) and smoking history (B). Adjusted for age, sex, body mass index, ethnicity, education level, socioeconomic status, smoking status, use of RA medications, and use of statin. ILD interstitial lung disease, ApoA apolipoprotein A, ApoB apolipoprotein B, HDL-C high-density lipoprotein cholesterol, LDL-C low-density lipoprotein cholesterol, HR hazard ratio, CI confidence interval

When stratified by smoking status (ever- vs. never-smokers), high levels of ApoA and HDL-C were associated with a reduced ILD risk among ever-smokers (Fig. 1B; ApoA: HR 0.346, 95% CI 0.152–0.788, p = 0.012; HDL-C: HR 0.421, 95% CI 0.220–0.807, p = 0.009, respectively), while a high level of triglycerides was associated with an increased risk in the same group (HR 1.235, 95% CI 1.052–1.452, p = 0.010). Lipoprotein A showed a modest protective effect in ever-smokers (HR 0.996, 95% CI 0.992–1.000, p = 0.045). No statistically significant associations were observed for any lipid traits among never-smokers. Our data suggested that the relationship between lipid levels and ILD risk tended to be more prominent in the RA population with a smoking history.

When stratified by age, high levels of ApoA and HDL-C were associated with a reduced ILD risk among younger participants aged < 60 years (Supplementary Figure S2; ApoA: HR 0.199, 95% CI 0.042–0.955, p = 0.044; HDL-C: HR 0.276, 95% CI 0.082–0.935, p = 0.039, respectively), while these associations were not observed in participants aged ≥ 60 years. Additionally, elevated triglycerides significantly increased ILD risk in younger patients with RA (Supplementary Figure S2, HR 1.352, 95% CI 1.115–1.641, p = 0.002), while no significant association was observed among patients aged ≥ 60 years (HR 0.984, 95% CI 0.801–1.209, p = 0.879). Lipoprotein A showed a modest protective effect only in the older group (HR 0.996, 95% CI 0.991–1.000, p = 0.044). These findings suggested that the relationship between lipid traits and ILD risk differs by age, with a more pronounced role for triglycerides, ApoA, and HDL-C in younger RA patients.

The restricted cubic spline analysis revealed a clear linear relationship between INFLA-score and incident ILD (linear model AIC: 2391.92, nonlinear model AIC: 2396.67, overall association p < 0.001, nonlinear p = 0.741). A steep increase in ILD risk was strongly associated with a higher INFLA-score (Fig. 2).

Association between chronic low-grade systemic inflammation (INFLA-score) and risk of incident ILD in patients with RA. The association between INFLA-score and risk of incident ILD was analyzed using linear regression with restricted cubic splines with 4 knots (− 4, 0, 4, and 8). The model was adjusted for age, sex, body mass index, ethnicity, education level, socioeconomic status, smoking status, use of RA medications, and use of statin. CI confidence interval

The GSEM model was employed to evaluate the potential mediating effect of the INFLA-score on the relationship between HDL-C and incident ILD (Fig. 3, Supplementary Table S4). The analysis revealed a statistically significant mediation effect of HDL-C on ILD risk through the INFLA-score (Estimate: − 0.260, p < 0.001), indicating that higher HDL-C levels were associated with a lower ILD risk via the mediation effect of reducing the chronic low-grade inflammation (INFLA-score). The model also showed a significantly protective total effect of HDL-C on the risk of ILD (Estimate: − 0.294, p = 0.014), while the direct effect of HDL-C on ILD, which represented a pathway independent of inflammation, was not statistically significant (Estimate: − 0.034, p = 0.790). The mediation analysis revealed that the INFLA-score accounted for approximately 88.5% of the total association between HDL-C and ILD risk (p = 0.023). The results indicated that the protective effect of HDL-C on the incidence of ILD in patients with RA is primarily mediated by suppressing the chronic systemic low-grade inflammation.

Mediation effect of chronic low-grade inflammation (INFLA-score) on the association between lipid traits and incident ILD in patients with RA. Mediation analysis was adjusted for adjusted for age, sex, body mass index, ethnicity, education level, socioeconomic status, smoking status, use of RA medications, and use of statin. Bootstrapping with 500 replications was used to estimate the coefficient and significant levels. ILD interstitial lung disease, ApoA apolipoprotein A, HDL-C high-density lipoprotein cholesterol

Similarly, the GSEM model identified the INFLA-score as a significant mediator in the relationship between ApoA and ILD risk (Fig. 3, Supplementary Table S4). A statistically significant indirect effect was observed (Estimate: − 0.245, p < 0.001), indicating that the protective association of ApoA operates predominantly by attenuating chronic low-grade inflammation. The total effect of ApoA on ILD was also significant (Estimate: − 0.385, p = 0.007), whereas the direct effect of ApoA on ILD was insignificant (Estimate: − 0.140, p = 0.371). The INFLA score mediated a 63.6% of the protective effect of ApoA against ILD (p = 0.027), explaining its role in reducing the low-grade inflammation.

In sensitivity analyses, similar results were obtained when we repeated the analyses after adjusting for age only (Supplementary Table S5). Both increased levels of ApoA and HDL-C were significant protective factors against incident ILD in patients with RA. When stratified by follow-up period, the HR and its statistical significance varied markedly across the censoring intervals of 5, 10, and 15 years (Supplementary Table S6). Notably, the protective effects of higher levels of ApoA and HDL-C became more pronounced with longer follow-up. ApoA showed a significant HR of 0.312 (95% CI 0.130–0.746, p = 0.009) at 10 years and 0.385 (95% CI 0.189–0.783, p = 0.008) at 15 years. A similar pattern was observed for HDL-C, with a significant protective effect during the 10- and 15-year follow-up periods, but no effect was significant during the first 5 years. This pattern suggested that the relationship between ApoA or HDL-C and ILD risk may be time-dependent, with protective associations becoming more evident over longer observation windows.