We appreciate Almalag and colleagues’ interest in our systematic review and meta-analysis of the association between methotrexate (MTX) use and incident cardiovascular disease (CVD). Most of their questions arise from their errors during efforts to reproduce our findings. Their first, and relatively minor, error is their characterization of several of the cohort studies as retrospective. Although there is no single definition of what makes a cohort study prospective or retrospective, the key methodologic consideration is whether exposure status was assessed before or after the development of the outcome, which directly affects the likelihood of recall bias or exposure information bias. For example, recall bias would be likely if past MTX use were ascertained by self-report in a cohort after patients had already developed or not developed CVD, and exposure information bias would be possible if investigators physically reviewed and abstracted chart information on past MTX use in a cohort after patients had already developed or not developed CVD. In contrast, the studies in question used established electronic administrative databases in which MTX prescription had been prospectively entered by electronic pharmacy prescriptions, before the development of CVD. Administrative databases are certainly more prone to random misclassification compared to more robust assessment methods, such as home medication inventories, but this does not affect the prospective nature of the exposure-outcome evaluation.

Their second error is the use of a minimally adjusted relative risk (RR) from van Halm et al. Such minimally adjusted RRs are highly susceptible to residual confounding. As specified in our “Methods” section, we used RRs that were multivariate adjusted for sociodemographic indicators, cardiovascular risk factors, the severity of underlying disease, medications for underlying disease, and the use of folate. In this case, we obtained the appropriate multivariate RR directly from the investigators, which is therefore correct as reported: 0.53 (95% confidence interval [CI] 0.24 to 1.19). Notably, Almalag and colleagues’ erroneous inclusion of this minimally adjusted RR is the major cause of the substantially greater heterogeneity in their meta-analysis. When the correct multivariate-adjusted RR is used, as in our meta-analysis, minimal heterogeneity is evident between studies.

We acknowledge that we used risk estimates and 95% CIs from Choi et al and Wolfe et al that were rounded to the nearest first decimal point in the original reports. This rounding led to slightly asymmetric CIs and calculated standard errors for these reports. We have repeated the meta-analysis using our best estimate of the true central RR and 95% CI for Choi et al (RR 0.346, 95% CI 1.86 to 0.653) and the exact values directly obtained from Wolfe et al (RR 0.964, 95% CI 0.718 to 1.295); the overall pooled RR is essentially unchanged: 0.78 (95% CI 0.72 to 0.86, I ² = 15.7%). Also, as specified in our “Methods” section, for Solomon et al, we pooled the inverse RRs for all other rheumatoid arthritis (RA) medications as the reference group, including biologic agents, other cytotoxic agents, noncytotoxic agents, and glucocorticoids. This includes pooling RRs from 6 models, not 4 as incorrectly performed by Almalag et al, resulting in the correct pooled RR of 0.74 (95% CI 0.62 to 0.88) as given in our report. A reanalysis of the data by the original investigators would be expected to be quite similar to the analysis we performed. Prodanowich et al have confirmed that patients with RA and psoriasis were excluded from their analysis, alleviating the concern raised for “double counting” (which would have unlikely had appreciable effects in any case). As we discuss in the following, there was also no justification to exclude patients with psoriasis from our meta-analysis. Interestingly, even with these errors and misconceptions, Almalag et al arrived at results very similar to ours, with a pooled estimate within our estimated CI. This supports the robustness of our results.

Perhaps the greatest misconception by Almalag et al is their consideration that these data are most relevant simply for making decisions for drug choices in patients with RA. This, for instance, leads to their suggestion to exclude patients with psoriasis and to their conclusions that these findings largely inform the selection of first-line therapy for patients with RA. As we described, the aim of our analysis was to evaluate the relation between MTX therapy and CVD risk to elucidate the potential role of direct treatment of chronic systematic inflammation on CVD risk. Such effects would have far-reaching implications for the inflammatory hypothesis of CVD risk and for all patients with persistently elevated inflammatory responses, such as those with high C-reactive protein levels. Our aims, design, inclusion and exclusion criteria, and analyses accordingly focused on obtaining the best unbiased comparisons of risk among any groups of otherwise similar patients with chronic systematic inflammation who were or were not treated with MTX.

On the basis of the design and analytic limitations of the individual studies that contributed to our meta-analysis, the observed lower risk for CVD with MTX use requires confirmation in additional well-designed prospective observational studies as well as large randomized controlled trials to evaluate whether direct treatment of chronic inflammation can reduce CVD risk in diverse populations. Such trials are planned, including the Cardiovascular Inflammation Reduction Trial (CIRT), and their results will substantially increase our understanding of the fascinating field of inflammation and CVD risk.