We thank Dr. Welsh and colleagues for their thoughtful comments. They suggest that concentrations of some inflammatory markers were higher in our control subjects than might be expected from published examples that report intercellular adhesion molecule 1 (ICAM-1) (1), IL-6, and TNFα concentrations (2). We found that ICAM-1 concentrations (median 119.4, interquartile range [IQR] 98.3–161.8 ng/ml) appeared to be somewhat lower, rather than higher, than those reported by Sattar et al (geometric mean ± SD 369.4 ± 1.39 ng/ml) (1). TNFα concentrations in our control subjects (median 3.4, IQR 2.4–4.8 pg/ml) were similar to those reported by Schaap et al in the Dynamics of Health, Aging, and Body Composition Study in an elderly population of 1,023 men (median 3.21, IQR 2.43–4.07 pg/ml) and 1,154 women (median 3.04, IQR 2.35–3.84 pg/ml) (3). Concentrations of IL-6 in our controls (median 4.2, IQR 1.2–18.2 pg/ml) were indeed somewhat higher than those generally reported in the literature, but were similar to those reported in the top one-third of control subjects in a large epidemiologic study (4) where the range was 2.5–28.1 pg/ml with a geometric mean of 4.2 pg/ml. Differences in median values among studies are likely to be due to relative differences resulting from variation in assay and subject characteristics, rather than noise in the assays, which would bias comparisons between groups in a study towards the null.
We agree that enzyme-linked immunosorbent assays (ELISAs) have inherent methodologic limitations, but these are not limited to multiplex assays. We did not set out to characterize the sensitivity and specificity of the assay, and the range of interassay CVs for the cytokine assays reported in our article (6.8–21%) were those of the manufacturer. Intra-assay reproducibility was good. For example, we found that for IL-6 the average percentage difference between 221 pairs of duplicate samples was 4.0% (95% confidence interval [95% CI] −0.87%, 9.4%); using a Bland-Altman plot analysis (5), the mean absolute difference between 2 duplicate samples was 0.96 (95% CI 0.90, 1.01 pg/ml).
Circulating antibodies such as RF that are present in serum could affect immunodetection assays; therefore, commercially available ELISAs include proprietary blocking reagents to minimize heterophilic interference. Nevertheless, we cannot exclude the possibility of interference by RF in the assay. However, there was no evidence of a systematic effect (an increase in all cytokines) in samples from patients with RF since only 2 (IL-6 and TNFα) of the 8 inflammatory mediators measured using the multiplex ELISA were significantly higher in patients with RF. An increase in these 2 cytokines in patients with RF is not unexpected given the association between RF and more severe disease. Furthermore, when we restricted our main analysis to only patients with RA and further adjusted the statistical model for RF (present or absent), TNFα (P = 0.002) and IL-6 (P = 0.026) remained significantly associated with coronary calcification.
As Welsh and colleagues mention, we did not set out to define the determinants of coronary calcification in control subjects since other studies with larger samples and greater power have done this. The borderline significant negative association between IL-6 concentrations and coronary calcification in control subjects was surprising, and we deliberately did not overinterpret this finding given the small number of control subjects with coronary calcification (n = 35). However, the significance of the positive associations between IL-6 and TNFα concentrations and coronary calcification in patients with RA was not dependent on the findings in control subjects, as shown above. Given the limitations of the study, we agree that the P value for interaction between disease status and mediator and coronary calcification should be interpreted cautiously. However, the relationship between these cytokines and coronary calcification in patients with RA is clear.
Our novel finding that IL-6 and TNFα were associated with coronary artery calcification in RA is concordant with the large body of literature, suggesting that inflammation is associated with atherosclerosis (6).