Systematic review and meta-analysis: Anti–tumor necrosis factor α therapy and cardiovascular events in rheumatoid arthritis




Control of rheumatoid arthritis (RA) may reduce the risk of cardiovascular events. We sought to systematically assess the association between anti–tumor necrosis factor α (anti-TNFα) therapy in RA and cardiovascular event rates.


Observational cohorts and randomized controlled trials (RCTs) reporting on cardiovascular events (all events, myocardial infarction [MI], congestive heart failure, and cerebrovascular accident [CVA]) in RA patients treated with anti-TNFα therapy compared to traditional disease-modifying antirheumatic drugs were identified from a search of PubMed (1950 to November 2009), EMBase (1980 to November 2009), and conference abstracts. Relative risks (RRs) or hazard ratios and 95% confidence intervals (95% CIs) were extracted. If the incidence was reported, additional data were extracted to calculate an incidence density ratio and its variance.


The systematic review and meta-analysis include 16 and 11 publications, respectively. In cohort studies, anti-TNFα therapy was associated with a reduced risk for all cardiovascular events (pooled adjusted RR 0.46; 95% CI 0.28, 0.77), MI (pooled adjusted RR 0.81; 95% CI 0.68, 0.96), and CVA (pooled adjusted RR 0.69; 95% CI 0.53, 0.89). Meta-analysis of RCTs also produced a point estimate indicating lower risk of cardiovascular events, but this was not statistically significant (pooled RR 0.85; 95% CI 0.28, 2.59).


Anti-TNFα therapy is associated with a reduced risk of all cardiovascular events, MI, and CVA in observational cohorts. There was heterogeneity among cohort studies and possible publication bias. The point estimate of the effect from RCTs is underpowered with wide 95% CIs, and cardiovascular events were secondary outcomes, but RCTs also demonstrated a trend toward decreased risk.


Rheumatoid arthritis (RA) is a chronic inflammatory disease that has been associated with increased cardiovascular event rates (1–3). This is attributed both to traditional cardiovascular risk factors and, more recently, the inflammatory milieu of inadequately treated RA (4). The treatment approach in RA has changed dramatically in the past 15 years, with a focus on treating disease early and achieving remission with disease-modifying antirheumatic drugs (DMARDs) and biologic therapies. Increased attention is also being paid to the management of cardiovascular risk factors in RA patients, owing to broadened recognition of the elevated cardiovascular morbidity and mortality in this population.

Effective treatment of inflammation through adequate RA disease control may reduce the risk of cardiovascular events. A recent systematic review identified the beneficial effect of methotrexate on cardiovascular event rates (5). Anti–tumor necrosis factor α (anti-TNFα) therapy has become a critical therapy for patients failing methotrexate and other standard DMARDs. It is apparent that anti-TNFα therapy may additionally abrogate risk, as it has been demonstrated to have a positive impact on surrogate biomarkers of cardiovascular disease. It has been shown to improve carotid intima-media thickness and flow-mediated dilation, among other measures of endothelial function, as well as reducing circulating levels of C-reactive protein and interleukin-6 (6, 7).

The assembly of large cohorts of treated RA patients internationally allows the opportunity to assess specific treatment outcomes, and additionally provides information on adverse events. A significant amount of evidence about treatment outcomes and side effects of anti-TNFα therapy has also accumulated in randomized controlled trials (RCTs). We wanted to systematically review these two sources of evidence, cohort studies and RCTs, to determine any possible clinical effect of anti-TNFα therapy on the risk of cardiovascular events. Specifically, our question is to examine whether observational cohorts and RCTs have identified an altered risk of cardiovascular events in RA patients treated with anti-TNFα therapy compared to the DMARD-treated RA population.


A systematic review and meta-analysis were performed according to the framework outlined in the Preferred Reporting Themes for Systematic Reviews and Meta-Analyses Group guidelines to increase standardization and quality in reporting (8). A review protocol was developed prior to literature searching (available from the corresponding author).

Data sources and searches.

We conducted a systematic literature search of PubMed (1950 to November 1, 2009) and EMBase (1980 to November 1, 2009). Pertinent narrative review articles and reference lists of key articles were searched for further relevant publications. Experts with recent publications were contacted to identify any unpublished data. Conference abstracts from 2007 to 2009 for major rheumatology and cardiology annual meetings were also reviewed (including those of the European League Against Rheumatism, American College of Rheumatology, Canadian Rheumatology Association, American College of Cardiology, American Heart Association, and Canadian Cardiovascular Society). In the instance where additional information was required for detail clarification or statistical analysis, the corresponding authors were contacted directly. Three themes were created using keywords and synonyms in titles and abstracts and using medical subject headings. These themes were “rheumatoid arthritis,” “cardiovascular events,” and “anti-TNF therapies” (see Supplementary Appendix A for the complete list of search terms, available in the online version of this article at These 3 themes were combined using the Boolean operator “AND” to identify studies reporting on cardiovascular events occurring in anti-TNFα–treated RA patients. In addition, the “anti-TNF therapies” and “rheumatoid arthritis” themes were combined and we applied the Cochrane collaboration RCT filter to identify all RCTs of anti-TNFα therapy in the treatment of RA (9).

Study selection.

Two authors (CB and B-JM) independently screened articles for inclusion in the full-text review by an initial screen of all titles and abstracts retrieved from the search strategy. No language restrictions were placed. Articles were included if they reported data from an original study of RA patients receiving anti-TNFα therapy that also reported on cardiovascular events. Because our focus was on clinical events, we also excluded studies reporting only on surrogate markers of atherosclerosis. The observed agreement at the first screening stage was 99.7% (κ = 0.89). Any articles identified from the first screen by either reviewer as possibly relevant to the study question were brought forward to the full-text review.

Full-text review was undertaken as the next step. Two study designs were considered for inclusion at this stage: observational cohorts and RCTs. Articles were included in the systematic review if they reported original data on RA patients receiving anti-TNFα therapy compared to patients receiving DMARD therapy and describing cardiovascular events (including all events, myocardial infarction [MI], congestive heart failure [CHF], and cerebrovascular accident [CVA]). We excluded studies with fewer than 26 weeks of observation to ensure that any identified effect was most likely a true effect and not due to chance in a short treatment period. The observed agreement at this second stage was 95.9% (κ = 0.89). Discrepancies were resolved by discussion between the two authors with input from the third author (WG). Studies included in the quantitative meta-analysis must have included sufficient data to allow for determination of a relative risk (RR) equivalent between groups.

Data extraction and quality assessment.

Data were extracted by both authors independently. We recorded the patient demographics, location of the study, and time period of observations; comparative treatment (methotrexate or any DMARD); RA disease activity parameters; and cardiovascular risk factors or preexisting conditions. RRs or hazard ratios (HRs) and their associated 95% confidence intervals (95% CIs) were directly extracted. If the incidence was reported, the number of events in both groups and patient-years of followup were extracted in order to calculate an incidence density ratio and its variance. In situations where covariate adjustment was made in regression models, the most adjusted RR equivalent was selected for data extraction. In cases where an anti-TNFα–treated population was compared to a general RA population and its treatment was not specified, it was assumed that it was receiving DMARD therapy, as this is the standard of care. Information necessary to assess study quality in accordance with checklists proposed by Egger et al for observational cohorts and RCTs was also extracted (10).

Data synthesis and statistical analysis.

Studies that did not present adequate information for calculation of a measure of association are presented as they were in the original publication. We did not combine the two study designs into one analysis, even if they reported on the same outcome. For cohort studies with adequate information for meta-analysis, we determined the pooled adjusted RR equivalent for all cardiovascular events, MI, CVA, and CHF in separate analyses for each outcome. HRs and incidence density ratios were directly considered RRs. Results presented as odds ratios were transformed into RRs using the method described by Zhang and Yu (11). For RCTs with adequate information for meta-analysis, the RR of an event between the anti-TNFα group and the comparator group was calculated. We used Stata, version 10.0 (StataCorp), for meta-analysis, using the log-adjusted RR across observational studies, and the “metan” command to pool the RR for RCTs. For the observational cohorts, we used the DerSimonian and Laird random-effects model when assessing the outcome of “all cardiovascular events,” and used fixed-effects models for assessing the outcomes of MI and CVA. We used a fixed-effects model for the RCTs. Forest plots were created to summarize the RR and log RR estimates and their 95% CIs for the cohort studies and RCTs. The “metacum” command was used to identify the effect of accumulation of evidence over time in the cohort studies.

Heterogeneity across studies was assessed using the Cochrane Q statistic and the I2 statistic. The Yates transformation was applied to one cohort study (Geborek et al [12]) to allow analysis, as there was a group with no events. Begg's test and visual analysis of the funnel plot were used to assess for the possibility of publication bias.


A total of 5,022 titles and abstracts were identified with the database search strategy described. An additional 7 publications were identified from manual searching and conference proceedings. Of the 5,029 total abstracts, 97 were selected for full-text review (Figure 1). Based on the prespecified eligibility criteria, 16 studies were selected for systematic review; characteristics of these studies are shown in Table 1. In summary, we identified 13 cohort studies (106,202 patients) and 3 RCTs (2,126 patients). Events reported varied between the studies, and included “all cardiovascular events” (in 5 cohorts and 1 RCT), MI (in 6 cohorts and 2 RCTs), CVA (in 4 cohorts), and CHF (in 6 cohorts). Four cohorts reported on “all anti-TNFα use,” 5 cohorts reported specifically on infliximab and etanercept, and 4 cohorts reported on infliximab, etanercept, and adalimumab. There was 1 RCT each for infliximab, etanercept, and adalimumab. Despite golimumab being included in our search, there was no exposure to this agent in the identified studies. The risk measure reported and the covariates and method of adjustment used in the analyses of the observational cohorts are shown in Table 2. After data extraction was performed, only 8 cohort studies and the 3 RCTs were determined to be acceptable for meta-analysis.

Figure 1.

Flow chart of study identification and selection. Anti-TNFα = anti–tumor necrosis factor α; DMARDs = disease-modifying antirheumatic drugs.

Table 1. Characteristics of studies included in the systematic review*
Author, year (ref.)SourceTherapy comparisonOutcomeNDisease duration, yearsFollowup time
  • *

    If demographic values were different between comparison groups, the highest value is recorded in the table. INF = infliximab; MTX = methotrexate; MI = myocardial infarction; ADA = adalimumab; DMARD = disease-modifying antirheumatic drug; ETN = etanercept; NR = not reported; CHF = congestive heart failure; CVA = cerebrovascular accident; anti-TNFα = anti–tumor necrosis factor α; RA = rheumatoid arthritis.

  • Fourteen percent comparator, no treatment.

  • Comparator treatment not specified.

St.Clair et al, 2004 (32)Multinational studyINF vs. MTXMI1,0400.854 weeks
Van de Putte et al, 2004 (33)Multinational studyADA vs. placebo (previous DMARD)MI5441126 weeks
Emery et al, 2008 (34)Multinational studyETN vs. MTXCardiovascular event5420.852 weeks
Geborek et al, 2002 (12)Swedish registryINF, ETN vs. leflunomideMI36914NR
Wolfe and Michaud, 2004 (15)American registryINF, ETN vs. DMARDCHF13,17115NR
Jacobsson et al, 2005 (24)Swedish registry, administrative dataINF, ETN vs. DMARDCardiovascular event, MI, CHF, CVA983114 years
Carmona et al, 2007 (16)Spanish registryAnti-TNFα vs. MTXCardiovascular event, CHF, CVA, mortality1,578NR5 years
Curtis et al, 2007 (18)American insurance claimsINF, ETN vs. MTXCHF2,121NR21 months
Cole et al, 2007 (17)American veteransINF, ETN, ADA vs. RA control groupCHF, mortality203NRNR
Dixon et al, 2007 (14)British registryINF, ETN, ADA vs. DMARDMI10,755121.7 years
Wolfe and Michaud, 2007 (35)American registryINF, ETN, ADA vs. RA control groupMI25,343NRNR
Jacobsson et al, 2008 (36)Swedish registryTNF vs. DMARDCardiovascular event, MI, CVA26,383NRNR
Listing et al, 2008 (13)German registryINF, ETN, ADA vs. DMARDCHF4,2489NR
Naranjo et al, 2008 (37)Multinational studyTNF vs. DMARDCardiovascular event, MI, CVA4,36311NR
Setoguchi et al, 2008 (19)American insurance claimsINF, ETN vs. MTXCHF5,593NR2.5 years
Solomon et al, 2008 (38)American registryTNF vs. DMARDCardiovascular event10,87072 years
Table 2. Risk expression, method of adjustment, and variables adjusted for in observational cohorts*
Author, year (ref.)Risk expressionMethod of adjustmentVariables adjusted for in model
  • *

    IR = incidence rate; RR = relative risk; IRR = incidence rate ratio; BMI = body mass index; HAQ = Health Assessment Questionnaire; DM = diabetes mellitus; CVD = cardiovascular disease; COPD = chronic obstructive pulmonary disease; HR = hazard ratio; DAS = Disease Activity Score; DAS28 = Disease Activity Score in 28 joints; RF = rheumatoid factor; RA = rheumatoid arthritis; DMARDs = disease-modifying antirheumatic drugs; NSAIDs = nonsteroidal antiinflammatory drugs; CRP = C-reactive protein; ESR = erythrocyte sedimentation rate; CKD = chronic kidney disease; OP = osteoporosis; CAD = coronary artery disease.

Geborek et al, 2002 (12)IRNoneNone
Wolfe and Michaud, 2004 (15)Percent incidenceNoneNone
Jacobsson et al, 2005 (24)IRNot specifiedAge, sex
Carmona et al, 2007 (16)IRNoneNone
Curtis et al, 2007 (18)IR, RRNoneNone
Cole et al, 2007 (17)IRNoneNone
Dixon et al, 2007 (14)IR, IRRPoisson regressionAge, sex, disease severity, BMI, social deprivation, smoking, comorbidity, baseline drugs
Wolfe and Michaud, 2007 (35)RRConditional logistic regressionBaseline HAQ score, prednisone use, hypertension, DM, smoking
Jacobsson et al, 2008 (36)RRCox regressionAge; sex; previous hospitalization for CVD, DM, or COPD; marital status; hospital stays; joint surgery
Listing et al, 2008 (13)HRCox proportional hazards models, multivariate Cox regression analysisAge, sex, previous CVD, BMI, functional capacity, DAS at followup
Naranjo et al, 2008 (37)HRCox proportional hazards regression modelsAge, sex, DAS28, HAQ score, RF, extraarticular disease, hypertension, hyperlipidemia, DM, smoking, obesity
Setoguchi et al, 2008 (19)HR, IRCox proportional hazards regression, multivariate Cox proportional hazardsAge, sex, race, CVD, RA surgery, extraarticular disease, other DMARDs and NSAIDs, arthrocentesis, intraarticular injection, CRP level, ESR, CKD, COPD, DM, hyperlipidemia, cancer, depression/anxiety, hypothyroidism, OP, acute infections, anemia
Solomon et al, 2008 (38)HRCox proportional hazards regression modelsHypertension, treatment for hyperlipidemia, DM, prior CAD, cigarettes, DAS, HAQ score, disease duration, seropositive, nodules

Study quality.

Cohort studies and RCTs were assessed according to components suggested by Egger et al (10). Tables 1 and 2 collectively show the information required to assess study quality. For the majority of the cohort studies, patient sampling and outcome assessments were as objective as possible given the inherent limitations in cohort design studies. Two cohort studies were thought to be high quality (Listing et al, 2008 [13], and Dixon et al, 2007 [14]) and 11 were medium quality. The major limitation in the medium-quality studies was the lack of information provided to determine representativeness and time point in the disease course of the cohorts. For all of the cohort studies except one, treatment was not standardized or randomized after entry into the cohort, as expected with real-world observational studies. Five cohort studies did not adjust for RA or cardiovascular risk factors at baseline, and among these, only 3 met the prespecified inclusion criteria to be included in the meta-analysis. All of the RCTs were high-quality studies, with all 3 studies having the key elements of blinding, concealment of allocation, and transparent reporting of patient flow through the trials.

Cohort study results.

Forest plots of the treatment effect of anti-TNFα therapy on all cardiovascular events, MI, and CVA are shown in Figure 2. In summary, anti-TNFα therapy is associated with a reduced risk for all cardiovascular events (pooled adjusted RR 0.46; 95% CI 0.28, 0.77), MI (pooled adjusted RR 0.81; 95% CI 0.68, 0.96), and CVA (pooled adjusted RR 0.69; 95% CI 0.53, 0.89). Analysis of each of these end points revealed heterogeneity of findings across studies, with I2 = 89.3% (heterogeneity P < 0.001) for all cardiovascular events, I2 = 39.9% (P = 0.139) for MI, and I2 = 39.3% (P = 0.176) for CVA. Analysis for publication bias was also done for each end point. Visual inspection suggests publication bias for all cardiovascular events, with a significant value for Begg's test (P = 0.05). There were few cohorts reporting increased cardiovascular events with anti-TNFα therapy. The funnel plots for MI and CVA were more symmetric, with Begg's test values of P = 0.851 and P = 0.174, respectively. The small number of studies published for each end point does limit the interpretation of the funnel plot visually and statistically.

Figure 2.

Adjusted relative risk of cardiovascular events in rheumatoid arthritis patients treated with anti–tumor necrosis factor α in observational cohorts of A, all cardiovascular events, B, myocardial infarction, and C, cerebrovascular accident. ES = effect size; 95% CI = 95% confidence interval.

Cumulative meta-analysis was performed on all of the cohort studies reporting MI as an outcome to assess whether findings varied over time of publication, and this analysis revealed no significant changes across years in the association with anti-TNFα therapy.

Meta-analysis of studies reporting on CHF as an outcome was not undertaken due to an insufficient number of studies available, inconsistent definitions for this end point, and highly inconsistent results across studies, leaving uncertainty regarding the direction of association. We nonetheless provide a narrative summary of the findings of the 6 studies that reported on this end point. Wolfe and Michaud reported a 1.2% (95% CI −1.9, −0.5) reduction in the rate for all heart failure events in anti-TNFα–treated patients compared to DMARD-treated patients, without a significant effect on incident cases (15). Carmona et al identified a rate ratio of 0.14 (95% CI 0.06, 0.32) for cases of CHF in persons exposed to anti-TNFα compared to persons exposed to methotrexate (16). Listing and colleagues identified that 16 of 25 incident cases of CHF occurred in patients exposed to anti-TNFα agents compared to 20 of 25 matched controls on DMARDs, for an RR of 0.8 (95% CI 0.56, 1.14) (13). A study by Cole et al, meanwhile, revealed essentially no difference in incident CHF between anti-TNFα users compared to DMARD users (7 of 103 versus 8 of 100 and RR 0.85; 95% CI 0.32, 2.26) (17). Two other studies found an elevated risk of CHF with anti-TNFα therapy, with Curtis et al identifying a 4.4-fold increased risk and Setoguchi et al identifying an adjusted HR of 2.1 (95% CI 1.0, 4.3) (18, 19).

RCT results.

Meta-analysis of the 3 RCTs revealed a pooled RR for cardiovascular event rates with anti-TNFα therapy compared to DMARD therapy of 0.85 (95% CI 0.28, 2.59) (Figure 3). This 95% CI is quite broad and it encompasses both the point estimate found with meta-analysis of the cohort studies and 1.0. The studies were found to be homogeneous (I2 = 0%, P = 0.90). The finding from the RCTs is not statistically significant but the point estimate is in the same direction of effect as that from the cohort studies, acknowledging small numbers and the fact that the RCTs were underpowered to assess differences in cardiovascular events.

Figure 3.

Relative risk of cardiovascular events in rheumatoid arthritis patients receiving anti–tumor necrosis factor α therapy in randomized controlled trials of a duration of >26 weeks. ES = effect size; 95% CI = 95% confidence interval.


This systematic review and meta-analysis supports the hypothesis, with caveats, that treatment with anti-TNFα agents is associated with a reduced risk of all cardiovascular events, MI, and CVA in patients with RA. Given that cardiovascular disease accounts for 35–50% of excess mortality in RA patients, this is a finding of great potential importance (20). A reduced risk of cardiovascular events was demonstrated across all of the cohort studies identified, regardless of the cardiovascular outcome reported, and after sensitivity analysis. Furthermore, cumulative analysis demonstrated that the effect has remained stable over time. Although only a small number of RCTs met the inclusion criteria for this systematic review and meta-analysis, the direction of the point estimate of effect was also in the direction of decreased risk, albeit not statistically significant.

Should one therefore conclude that our cohort study and RCT findings are consistent or discrepant? Statistically they are not inconsistent, and therefore it could be argued that they are demonstrating the same effect. However, the cohort effect estimate is significant while the RCT finding is not, with the latter estimate being closer to 1.0. Concato et al have previously discussed the relevance and value of generally congruent findings (as seen here) between well-designed observational cohort studies and RCTs (21). They propose that such scenarios are common, and provide added robustness to conclusions regarding the effect of specific exposures on outcomes of interest. However, we should consider the possible contribution of the “healthy user effect” to our results (22). In the context of RA, the “healthy user” would be a patient who is generally healthier, younger, and more likely to receive health services such as aggressive RA management (including anti-TNFα therapy) and potentially other cardiovascular risk–modifying therapies. This would imply that there are material differences between the patients that are receiving anti-TNFα therapy versus those who are not, which may explain why we observed a stronger association between anti-TNFα therapy and a reduction in cardiovascular events in the cohort studies (that cannot control for unmeasured confounders) relative to the RCTs. Discussion of such differences has been invoked to explain the stark differences in the effect of hormone replacement therapy in observational versus RCTs of that medication (23). However, unlike the notable hormone replacement therapy story, our findings were more consistent across study designs and also various user strata. The association between anti-TNFα therapy and fewer cardiovascular events was consistent across countries known to have different health care delivery structures (15, 24). The effect was also consistent across socioeconomic groups, as the patients receiving Medicaid appeared to derive the same benefit from anti-TNFα therapy (19). The effect was consistent across ages, as decreased events were seen in both young and elderly patients (14, 16). Finally, as demonstrated in our cumulative analysis, the effect was consistent across the time of publication in our cohort studies. As anti-TNFα therapy was prescribed more widely and to more diverse patient populations, we continued to see an association between its use and decreased cardiovascular events.

Our review does not identify the mechanism responsible for a potential reduced risk for cardiovascular events, whether it be through improved disease control itself and improved inflammatory parameters, a resulting reduction in the use of potentially cardiotoxic medications such as antiinflammatory agents and corticosteroids, or a direct effect of the anti-TNFα agent, but should be considered in the context of basic science studies showing improved surrogate measures of atherosclerosis and inflammation with anti-TNFα therapy. Anti-TNFα therapy is associated with a reduction in C-reactive protein level, a known independent risk factor for the development of cardiovascular disease (25, 26). Treatment of ankylosing spondylitis, another inflammatory condition, with etanercept demonstrated a considerable improvement in the lipid profile of patients after 3 months of treatment (27). Through a series of vascular biology studies, Gonzalez-Juanatey and colleagues have demonstrated improvements in endothelial function, as measured through either flow-mediated dilation or carotid intima-media thickness, in patients administered biologic agents, including anti-TNFα therapy (28, 29). Recent recommendations from the European League Against Rheumatism reinforce the importance of early aggressive treatment of inflammatory arthritis with anti-TNFα and methotrexate treatment for the specific purpose of lowering cardiovascular risk in two ways (30). The first is the direct effect of anti-TNFα therapy on decreasing inflammation, and the second is that improving joint inflammation and function may lead to increased levels of physical activity, which will subsequently decrease the incidence of other cardiovascular risk factors such as diabetes mellitus and hypertension.

Limitations of our study include the assumption that the cohorts were receiving the prescribed treatment that allowed their inclusion into their respective treatment groups. Information is not available regarding the proportion of patients that were receiving DMARD therapy, or in what dose. We also had to assume that both treatment groups were treated to the same remission goals. We were unable to analyze for differences in the treatment effects of patients receiving anti-TNFα monotherapy as compared to anti-TNFα and DMARD combination therapy, as this information was not provided in the literature. Meta-regression was not possible due to the limited information provided on important disease covariates and cardiovascular risk factors. Cohort studies also varied in the amount of statistical adjustment performed in modeling for the outcomes of interest. We also acknowledge that the results of the cohort studies' meta-analysis are limited by significant heterogeneity. This is at least partially explained by the large sample size overpowering the heterogeneity test. Visual inspection of the effect sizes from each study consistently demonstrated a decrease in events with anti-TNFα therapy, so meta-analysis was still thought to be justified by a uniform direction of effect.

The limitations with regard to cardiovascular events from RCTs of anti-TNFα therapy are that the studies were primarily conducted to identify a short-term treatment effect of anti-TNFα therapy, with adverse events as secondary end points. RCTs consist of a highly selected population that is likely to be at low risk for cardiovascular events. The advantage of including RCTs in our analysis is that adverse events are adjudicated with strict criteria, such that outcome ascertainment is firm. They also afford the ability to directly compare the effect of anti-TNFα therapy to a standardized DMARD treatment.

A final caveat to our review findings is that certolizumab was not considered in our study. Certolizumab was just approved for broad clinical use at the start of our review, and was not included a priori in our search strategy. Only one study of certolizumab met our inclusion criteria (31); including this study in our meta-analysis did not have an impact on the results.

In summary, we identified that RA treatment with anti-TNFα therapy is likely associated with a reduced risk of cardiovascular events, a finding that is supported by clear biologic plausibility. We encourage further observation of the effect of anti-TNFα therapy benefits in RA, especially in studies (both cohort and RCTs) that include a longer duration of followup. It is possible that with longer observation, a more pronounced treatment effect on reducing cardiovascular events will be discovered, as many of the people who are currently treated with anti-TNFα therapies are relatively young and may not face cardiovascular morbidity for several years to come. Studies providing more detailed information on potential confounders would be of great value in adding to this body of literature.


All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Barnabe had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Barnabe, Martin, Ghali.

Acquisition of data. Barnabe, Martin.

Analysis and interpretation of data. Barnabe, Martin, Ghali.