Cardiovascular safety of tocilizumab: A systematic review and network meta-analysis

Objectives Our objective was to compare the cardiovascular safety of tocilizumab and other biological disease-modifying antirheumatic drugs (bDMARD) in rheumatoid arthritis using a network meta-analysis (NMA). Methods A systematic literature search through May 2018 identified randomized controlled trials (RCT) or observational studies (cohort only) reporting cardiovascular outcomes of tocilizumab (TCZ) and/or abatacept (ABA) and/or rituximab (RTX) and/or tumor necrosis factor inhibitors (TNFi) in rheumatoid arthritis patients. The composite primary outcome was the rate of major adverse cardiovascular outcomes (MACE, myocardial infarction (MI), peripheral artery disease (PAD) and cardiac heart failure (CHF)). Results 19 studies were included in the NMA, including 11 RCTs and 8 cohort studies. We found less events with RTX (5.41 [1.70;17.26]. We found no difference between TCZ and other treatments. Concerning MI, we found no difference between TCZ and csDMARD (4.23 [0.22;80.64]), no difference between TCZ and TNFi (2.00 [0.18;21.84]). There was no difference between TCZ and csDMARD (1.51[0.02;103.50] and between TCZ and TNFi (1.00 [0.06;15.85]) for stroke event. With cohorts and RCT NMA, we found no difference between TCZ and other treatments for MACE (0.66 [0.42;1.03] with ABA, 1.04 [0.60;1.81] with RTX, 0.78[0.53;1.16] and 0.91 [0.54;1.51] with csDMARD), but the risk of myocardial infarction was lower with TCZ compared to ABA (0.67 [0.47;0.97]). We lacked data to compare TCZ and other bDMARD for stoke and MI. Not enough data was available to perform a NMA for CHF and PAD. Conclusions Despite an increase in cholesterol levels, TCZ has safe cardiovascular outcomes compared to other bDMARD.


Results
19 studies were included in the NMA, including 11 RCTs and 8 cohort studies. We found less events with RTX (5.41 [1.70;17.26]. We found no difference between TCZ and other treatments. Concerning MI, we found no difference between TCZ and csDMARD ( We lacked data to compare TCZ and other bDMARD for stoke and MI. Not enough data was available to perform a NMA for CHF and PAD. Introduction guidelines (S1 File) [19]. Our protocol was recorded on PROSPERO under the registration number CRD42018097180.

Literature search
This systematic review was performed by searching in PubMed/Medline, Science Direct, and Web of Science databases, and in Cochrane and Wiley Online libraries. We also searched abstracts in the American College of Rheumatology (ACR) and European Ligue Against Rheumatism (EULAR) annual meetings databases, from January 2003 until May 2018. The search equations are available in supporting information (S2 File).

Eligibility criteria
Our eligibility criteria for qualitative analysis were: RCT or observational (cohort only) studies, wrote in French or in English, assessing cardiovascular outcomes among rheumatoid arthritis patients treated by TCZ and/or ABA and/or rituximab (RTX) and/or TNFi.

Study selection
Studies were selected on the basis of their titles and abstracts, then on their full text, by two independent reviewers (BC, JM). Disagreements were resolved through discussion with a third reviewer (MV), when necessary.

Data extraction and quality assessment
Two reviewers (BC and JM) assessed independently the quality of selected studies using the Cochrane Risk Of Bias tool (RoB 2.0) [20] for randomized studies and the Cochrane Risk Of Bias In Non-randomized Studies-of Interventions (ROBINS-I tool) [21] for non-randomized. The two reviewers (BC and JM) also performed data extraction independently using a predetermined form. If necessary, disagreements were resolved by the third reviewer (MV). We extracted from each study: authors and publication year, country, bDMARD used in the study, number of previous bDMARD and previous csDMARD, number of patients, patientyear, follow-up period (year), number of MACE (major adverse cardiovascular events), myocardial infarction (MI), stroke, cardiac heart failure (CHF) and peripheral artery disease (PAD),

Outcomes measures
The composite primary outcome was the rate of MACE which included stroke, MI, CHF and PAD (obliterating arteriopathy of the limbs, kidney and mesenteric artery diseases and aortic diseases. The secondary outcomes were rates of each cardiovascular event).

Data analysis
A frequentist network meta-analysis based on a random effects model [22] was conducted for all outcomes to compute relative risk (RR) and their 95% confidence interval (95%CI). NMA allows to synthesize information from numerous studies addressing the same outcomes but involving different interventions [23,24]. NMA combines direct and indirect evidence across a network of studies into a single effect size. Forest plots were used to represent the quantitative results. For pairwise comparison, statistical heterogeneity between studies was assessed by the Cochran's Q test (p<0.05 for significativity) and the I 2 statistic (<25%: low heterogeneity, 25-50%: moderate, 50-75%: high and >75%: very high). All analyses were performed using R (netmeta package version 0.9-5 for treatment comparison, meta package version 4.8-2, R Language and Environment for Statistical Computing, Vienna, Austria) [25,26]. A subgroup analysis was performed, including RCTs only and cohort studies only.

Literature search results and studies characteristics
After duplicates removal, 10 454 articles were found using PubMed/Medline, Science direct, Web of science, Cochrane library databases and ACR and EULAR conference abstracts. Of the 10 454 identified records, 10 319 were excluded at title/abstract level, leaving 135 full-text examined. Of these, we selected 29 studies, 27 original articles and two conference abstracts [18,27] for qualitative analysis (Fig 1). 106 articles did not meet the inclusion criteria for the following reasons: one was a case-control study, 55 were not controlled studies, 44 did not have clinical outcomes (biological or imaging), five because the studied population was already included, and one because the full-test could not be accessed.
We identify two major concerns due to cofounding in two cohort studies [40,49], because of the absence of multivariate analysis. The other potential bias was selective reporting due to the absence of online recorded protocol for 8 studies. Detailed bias assessment is reported in Table 2.

Stroke risk with TCZ vs other bDMARD
Five studies were included for stroke event (two RCT and three cohorts . We lacked data to compare TCZ and other bDMARD (Fig 2). Similar results were found with the analysis that included both designs (Fig  4), and with observational studies only (Fig 3).

Comparison of CHF
There was not enough data available to perform a network meta-analysis.

Discussion
We performed the first network meta-analysis comparing the cardiovascular outcomes of TCZ and other biologics. However, we found a lower risk of MACE with RTX in the RCT NMA, which is not significant when we included only cohorts or both designs. This study showed that TCZ has similar cardiovascular outcomes comparing with other bDMARD and csDMARD. TCZ significantly increases lipids levels more than TNFi [12], ABA and RTX [50]. This is one of the reasons why TCZ was initially used with a particular attention in patient with a high   We also found less myocardial infarction with TCZ than ABA and TNFi in observational studies. This beneficial effect on myocardial infarction could be explained by the fact that TCZ decreases more lipoprotein A levels, and SAA-HDL than adalimumab in the ADACTA study [12]. Indeed, lipoprotein-A has been associated with MI [51] and the HDL linkage by SSA protein could be responsible for HDL-C loss anti atherogenic function [12,50,52,53]. Furthermore, inflammatory cytokines, particularly IL-1 and IL-6 seem to play a determinant role in the atherosclerotic process. Recently, the CANTOS trial performed in general population shown a significantly lower rate of recurrent cardiovascular events with canakinumab than placebo [54]. A cohort study found an improvement of the endothelial function in high risk patients with RA with TCZ but not with TNFi [55]. IL-6 seems strongly involved in the coronary heart disease (CHD), especially via the trans-signaling pathway [56][57][58]. Mendelian randomization studies showed that IL-6 up-regulation increases CHD risk [59,60]. Conversely, the variant Asp358 causes an increase in the soluble IL-6 receptor (IL-6Rs) and is associated with a reduced risk of coronary artery disease [61,62]. Indeed, in a placebo-RCT in general population with MI, a single dose of TCZ decreases more troponin levels than placebo at day 3 [63]. In an observational study Kobayashi & al. found that TCZ treatment in RA patients significantly increased left ventricular ejection fraction and decreased left ventricular mass index comparing to healthy controls [64]. Finally, in highrisk population of CHD, high IL-6 levels were associated with MI risk, cardiovascular death and all causes death [65]. This systematic review and meta-analysis have several strengths. Indeed, we performed an exhaustive review, through four international databases, and two international meetings databases. The included studies showed high quality. RCT were at low risk of bias except for which who were open-label. However, one of these had MACE as primary outcome with an openlabel design. Furthermore this study is not published and was found in the 2016 ACR meetings abstract [16]. Most of the cohort studies were at "moderate risk of bias" only because of their non-randomized design. We realized a network meta-analysis which allowed us to make indirect comparisons of multiple treatments and thus to analyze a large amount of studies including placebo RCT. Due to multiple comparators, heterogeneity had to be measured which was possible with this technique. Furthermore, we used random effects model even if the heterogeneity test was no statistically significant because it considers fluctuations of sampling and also treatment effect fluctuations due to multiple covariates. Finally, although we restricted our research to English and French languages, no article was excluded because of the language.
However, our studies showed some limits. The heterogeneity test was significant for MACE outcome which was controlled by using a random effects model. The transivity hypothesis has not been tested, which is a limitation of the study. Unfortunately, the baseline CV risk was not specified in the included studies, we were unable to carry out an analysis to determine whether this risk was an effect modifier.
Our network meta-analysis illustrates the indication bias in observational studies. Indeed, the beneficial effect of TCZ for MI could be explained by a lesser use of TCZ in high CV risk patients because of the cholesterol increase under treatment. The fact that TCZ is under used in high-risk CV patients is not supported by the literature, the indication bias due to the prudent use of TCZ in high-risk patients is therefore purely hypothetical, not supported by the indirect), TCZ-tocilizumab; csDMARDs-conventional synthetic disease-modifying antirheumatic drugs; TNFitumor necrosis factor inhibitor; ABA-abatacept; RTX-rituximab.
https://doi.org/10.1371/journal.pone.0220178.g003 literature, but remains likely. Another explanation is that observational studies were designed to study cardiovascular events, with high sample sizes, unlike RCTs.

Conclusion
Despite these limits, our study showed a good cardiovascular safety profile of TCZ compared to other bDMARD, with potential benefit on MI compared to other bDMARD. Further studies are needed to corroborate these data.  . http://acrabstracts.org/abstract/comparison-ofchanges-in-cardiovascular-risk-associated-biomarkers-in-ra-patients-treated-with-anti-tnf-or-otherbiological-agents-a-metabolic-study-from-a-randomized-trial/