Article Text
PDF
Abstract
Background Antithrombotic therapy reduces stroke, embolism and mortality in patients with atrial fibrillation (AF); however, meta-analyses have focused on pairwise comparisons of treatments.
Objective To synthesise the evidence from trials using a multiple treatment comparison methods thereby permitting a broader comparison across multiple therapies.
Design, setting, patients Randomised controlled trials in patients with AF of antithrombotics were identified from MEDLINE, Embase, and Cochrane Central Register of Controlled Trials through May 2012. We performed a random-effects model within a Bayesian framework using Markov Chain Monte Carlo simulation to calculate pooled OR and 95% credibility intervals (CrI). We also ranked therapies by their likelihood of leading to the best results for the outcomes.
Main outcome measure Multiple endpoints including stroke, embolism, death and bleeding.
Results We identified 20 studies with 79 808 patients allocated to 8 treatments: ASA, ASA plus clopidogrel, vitamin K antagonists (VKAs), dabigatran 110 mg, dabigatran 150 mg, rivaroxaban, apixaban or placebo/control. Compared with placebo/control, dabigatran 150 mg was associated with the lowest risk of stroke (OR=0.25, 0.15–0.43), the composite of ischaemic stroke or systemic embolism (OR=0.26, 0.12–0.54) and mortality (OR=0.53, 0.28–0.88). ASA plus clopidogrel was associated with the highest risk of major bleeding (OR=3.65, 1.22–13.56). In simulated comparisons, the novel oral anticoagulants ranked better than VKA or antiplatelet therapies for prevention of stroke, ischaemic stroke or systemic embolism and mortality.
Conclusions In this network meta-analysis, novel oral anticoagulants were the most promising treatments to reduce stroke, stroke or systemic embolism, and all-cause mortality in patients with AF.
Statistics from Altmetric.com
Introduction
In patients with atrial fibrillation (AF), Vitamin K antagonists (VKAs), such as warfarin, reduce the risk of stroke by 65% compared with placebo, but increase the risk of haemorrhage.1 Although VKAs are recommended for patients who have AF and at least one risk factor,2 these drugs have several well-known limitations and the need for frequent laboratory monitoring. As a result, adherence is low,3 ,4 and many patients who take VKAs are inadequately anticoagulated.5 ,6 Antiplatelet agents also appear to significantly reduce the incidence of stroke,1 ,7 but are less effective than oral anticoagulants. A meta-analysis of trials comparing warfarin versus ASA showed better stroke reduction with warfarin, but a higher rate of intracranial bleeding, with no difference in total mortality.1 In a recent randomised trial,8 anticoagulant therapy was superior to the combination of clopidogrel plus ASA in reducing stroke, with no difference in bleeding.
Several novel oral anticoagulants have been developed in an attempt to overcome the many challenges faced by clinicians and patients with the use of VKAs.9–11 Dabigatran,12 rivaroxaban13 and apixaban14 were each ‘at least as good as’ warfarin in reducing the risk of stroke or embolism in large phase III randomised clinical trials designed to test whether the new agent was non-inferior to warfarin. Not only did these new agents meet criteria for non-inferiority, but additionally, dabigatran 150 mg twice daily and apixaban 5 mg twice daily were superior to warfarin in reducing the composite of stroke or systemic embolism. Finally, in a fourth trial of a novel anticoagulant, apixaban reduced the risk of stroke or systemic embolism compared with aspirin monotherapy without significantly increasing the risk of major bleeding or intracranial haemorrhage.15
In a recent meta-analysis of large phase III trials of novel oral anticoagulants compared with warfarin in patients with AF, statistically significant reductions with novel oral anticoagulants were observed in the primary outcome of stroke or systemic embolism (18% relative reduction), as well as all-cause mortality (9%) and haemorrhagic stroke (49%).16 However, standard meta-analyses are unable to integrate all available randomised evidence in one analysis that also includes indirect comparisons between therapies. By contrast, network meta-analysis allows a unified, coherent analysis of all randomised controlled trials that compare antithrombotic drugs head to head or with placebo/control while fully respecting randomisation.17 In the context of a systematic review, a network meta-analysis can compare three or more treatments using both direct comparisons of interventions within randomised controlled trials and indirect comparisons across trials based on a common comparator. The idea underlying multitreatment meta-analysis methodology is that, for any given comparison between two treatments A and B, direct evidence (coming from studies comparing A with B) and indirect evidence (coming from combining studies through an intermediate comparator, eg, A vs C and B vs C studies) can be synthesised into a single effect size. When direct and indirect evidence are available, the two sources of information can be combined as a weighted average.
Thus, we performed a network meta-analysis using available phase II and III trial data to answer the following research questions: What are the relative effects of different classes of antithrombotic drugs in reducing the incidence of stroke, the composite of ischaemic stroke or systemic embolism, death from any cause, and major bleeding in patients with non-valvular AF?
Methods
Search strategy
We searched Medline, Embase and the Cochrane database of systematic reviews through May 2013 with no language restriction, using the following Medical Subject Heading and keywords: anticoagulant, antiplatelet, aspirin, clopidogrel, warfarin, vitamin K antagonists, dabigatran, apixaban, rivaroxaban, atrial fibrillation, atrial arrhythmias, coupled with outcome searched using the terms stroke, cerebrovascular accident, transient ischemic attack, and TIA. We also reviewed the reference lists of published meta-analyses of anticoagulant and antiplatelet therapies to prevent stroke and embolic events in patients with AF. Overall, 20 manuscripts were included in our analyses (figure 1).
Flow diagram of literature search.
Eligibility ofstudies
Inclusion criteria for retrieved studies were: (a) randomised controlled phase II or III trials of VKAs, ASA, clopidogrel and novel oral anticoagulants in patients with non-valvular AF; (b) randomised treatment allocation, (c) intention-to-treat analysis and (d) follow-up more than 1 year. In order to reflect current practice patterns, we excluded studies or study arms where VKAs were administered at non-standard doses (eg, low fixed doses) or antiplatelet agents other than ASA or clopidogrel were tested.
Data abstraction and quality assessment
Two independent reviewers (AD, EP) performed data abstraction. We used consensus to resolve discrepancies. The endpoints of primary interest were: stroke, composite of ischaemic stroke or systemic embolism, death from any cause, and major bleeding. Definitions of endpoints were the same across all trials, with the exception of major bleeding where we used the trial-specific definition. We used data from the intention-to-treat populations, unless otherwise specified.
Statistical analysis
This meta-analysis was prospectively registered online in metcardio.org (Protocol #2-2013). First, pair-wise meta-analyses with a random-effects model18 were carried out. We used visual inspection of the forest plots and the I2 statistic to investigate the possibility of statistical heterogeneity using the Higgins–Thompson method19 (low heterogeneity 25%, moderate 50%, high 75%). All data were analysed using a Stata V.9. Second, a random-effects model multitreatment meta-analysis within a Bayesian framework (using Markov Chain Monte Carlo simulation)20 was performed.
Multiple treatment meta-analysis was performed using GeMTC R package.21 The statistical analysis is based on binomial likelihoods with a logit link function. GeMTC automatically specified vague prior distributions for the trial baseline effects, the relative effects (normal with mean 0 and SD 37.5), and the random effects SD (uniform in the interval 0–2.5). We used a technique known as ‘node-splitting’22 to evaluate for inconsistency in the findings of the network meta-analysis coming from direct versus indirect evidence. Node-splitting assesses whether direct and indirect evidence on a specific node (the split node) are in agreement. The basic idea is that, for the node of interest, two posterior distributions are generated from independent sources: trials that directly compare two interventions and trials that do not. Measures of conflict between these sources are identified from the measures of compatibility between these two posterior distributions.
Direct and indirect estimates of effect and the corresponding Bayesian ‘p values’ for inconsistency were calculated. We expressed the comparative effectiveness of the treatments as the OR of an outcome, with 95% credibility intervals (CrIs). The credibility interval is the Bayesian analogue to CIs used in traditional frequentist statistical approaches. A result was considered ‘significant’ if the CrI did not include 1.0. No statistical adjustment was made for multiple comparisons. Because the studies had different lengths of follow-up and to account for censored data, we obtained the rates of outcomes for all trials, and calculated the log HRs (assuming a constant rate of hazards for individual trials) from the event rate reported and mean duration of follow-up.
We also ranked the different interventions in terms of their likelihood of leading to the best results for each outcome. In each Markov chain Monte Carlo cycle, each treatment, j, is ranked according to the estimated effect size. Then, the proportion of the cycles in which a given treatment ranks first out of the total gives the probability P(j=1) that treatment j ranks first, that is, ‘is the best’ among the a available treatment options. Similar probabilities are calculated for being the second best, the third best, and so on -P(j=b), b=1,…, a. These probabilities sum to 1 for each treatment and each rank.
Sensitivity analysis
We performed four sensitivity analyses including and excluding specific studies that utilised substantial different methods to report or analyse their endpoint data, or had important differences in study entry criteria: (1) Since major bleeding data in the Randomized Evaluation of Long-Term Anticoagulation Therapy (RE-LY) trial was presented and analysed as ‘on treatment’, we summarised the data from the studies as published, and then excluded the data from RE-LY. (2) Because the SPAF-I,23 European Atrial Fibrillation Trial (EAFT),24 Primary prevention of arterial thromboembolism in non-rheumatic atrial fibrillation in primary care (PATAF),25 ACTIVE (A)7 and Apixaban Versus Acetylsalicylic Acid [ASA] to Prevent Stroke in Atrial Fibrillation Patients Who Have Failed or Are Unsuitable for Vitamin K Antagonist Treatment (AVERROES)15 studies each included treatment arms of patients in whom VKAs therapy was unsuitable, we performed sensitivity analyses excluding these treatments. (3) Sensitivity analyses that eliminated two phase II trials26 ,27 and two smaller phase III28 ,29 were performed. (4) Since the ROCKET-AF13 trial by design included only patients at high risk of stroke, we performed a sensitivity analysis excluding the data from this one study.
Finally, we performed a network meta-analysis computing HRs with a Poisson regression model and random effects method, as such analyses explicitly exploit differences in follow-up between studies, thus maximising precision and validity.
Results
Of the 312 studies identified, 292 were excluded due to non-randomised design, registry data, case series, subgroup analysis, posthoc analysis, low-dose warfarin, use of antiplatelet agents other than ASA or clopidogrel, or follow-up <1 year (figure 1). Twenty studies enrolling 79 808 patients with non-valvular AF met our criteria for analysis, including eight different strategies: ASA, ASA plus clopidogrel, dabigatran 150 mg twice daily, dabigatran 110 mg twice daily, adjusted dose of VKAs, rivaroxaban, apixaban and placebo/control. The main study characteristics are listed in table 1. Baseline characteristics of patients are shown in table 2.
Main study characteristics
Baseline characteristics of the patients
The eight interventions (including placebo/control) resulted in a total of 28 theoretical comparisons for each of the four primary outcomes of interest. The following comparisons were direct (figure 2): ASA versus ASA plus clopidogrel, ASA versus control, ASA plus clopidogrel versus VKA, VKA versus control, VKA versus apixaban, ASA versus apixaban, VKA versus apixaban, VKA versus dabigatran 150 mg, VKA versus dabigatran 110 mg, dabigatran 150 mg versus dabigatran 110 mg and rivaroxaban versus VKA. The results of the multiple treatment meta-analysis for the primary endpoints of interest are shown in table 3 and for secondary endpoints in online supplementary table 1.
Clinical results
Network of eligible comparisons for the multiple-treatment meta-analysis for efficacy. The numbers represent the number of trials or arms comparing each pair of treatment. In brackets, the number of studies that were performed with that treatment. ASA, aspirin; Clopi, clopidogrel; Dabig, dabigatran; VKAs, vitamin K antagonist.
Stroke reduction
The rate of stroke was lowest with dabigatran 150 mg twice daily followed by rivaroxaban, apixaban, dabigatran 110 mg twice daily and VKA. Each of the oral anticoagulants reduced stroke, compared with placebo/control, ASA and ASA plus clopidogrel (except for dabigatran 110 mg compared with dual antiplatelet therapy). Dabigatran, apixaban and rivaroxaban were not significantly more effective than VKA in reducing stroke. Antiplatelet regimens achieved a borderline statistically significant reduction in stroke over placebo/control.
Ischaemic stroke or systemic embolism
The composite rate of ischaemic stroke or systemic embolism was lower with oral anticoagulants when compared with either placebo/control, ASA alone, or ASA plus clopidogrel, with the exception of dabigatran 110 mg, which was superior to placebo/control and ASA (but not ASA plus clopidogrel). ASA alone and dual antiplatelet therapy were not superior to placebo/control.
Total mortality
Oral anticoagulants were superior to placebo/control for death from any cause, however, only VKA lowered total mortality compared with ASA.
Major bleeding
ASA plus clopidogrel and VKAs increased major bleeding compared with placebo/control. VKAs were associated with more major bleeding than ASA.
Ranking of treatments
Dabigatran 150 mg was ranked highest in prevention of stroke (70% chance of being the best), the composite of ischaemic stroke or systemic embolism (60%), and mortality (31%). Rivaroxaban (30%) and ASA plus clopidogrel (29%) had the highest likelihood for causing the most major bleeding in the simulations (table 4).
Rankings based on simulations
Sensitivity analyses
In sensitivity analyses restricted to the 16 large phase III trials, the results comparing the effects of the eight treatments were qualitatively similar (results not shown) to the results obtained with 20 studies. After excluding data from ACTIVE (A),7 AVERROES,15 and patients ineligible for anticoagulation in the SPAF-I,23 EAFT24 and PATAF25 trials, there were no substantial changes in the overall results with regard to the four outcomes (data not shown). Similarly, exclusion of the data from the ROCKET-AF trial13 did not result in any qualitative changes in the results (data not shown). Additionally, when we incorporate time of follow-up, and the ORs were converted in HRs in a Poisson analysis, we found only two qualitative differences: for major bleeding control versus ASA shifted from non-significant to significant, and VKA versus ASA shifted from significant to non-significant.
Consistency of network-model
No major inconsistencies or qualitative differences (eg, change in directionality of the estimate) were observed when we compared the effect estimates based on direct versus indirect evidence from the comparisons (table 5), supporting the robustness of the model. In one instance out of 28 comparisons (table 5), the inclusion of indirect evidence (compared with analyses using only direct evidence) shifted the statistical significance of the effect estimate. Specifically, in the analyses comparing major bleeding with ASA versus control/placebo the effect estimate shifted from significant (OR=3.03, 1.06–8.85) to non-significant (OR=2.17, 0.96–5.27).
Effect estimates from multiple treatment meta-analysis compared with direct and indirect estimates, based on node-splitting,22 and pair-wise meta-analyses
Heterogeneity
Most of the comparisons showed little or no heterogeneity. In only one of 28 comparisons (ASA vs ASA + clopidogrel for the endpoint of stroke or systemic embolism) did the I2 value exceeded 50% (I2=55.8%), indicating the presence of moderate heterogeneity.
Discussion
In this meta-analysis of the 20 trials of antithrombotics in patients with non-valvular AF, we observed statistically significant reductions in stroke, ischaemic stroke and systemic embolism, and all-cause mortality with novel oral anticoagulants compared with no treatment. Unlike frequentist statistics that are limited to direct comparisons between therapies, multiple treatments meta-analysis allows the integration all available treatments studied. This methodology also can rank order the tested therapeutic interventions from most likely to least likely of being the best. In this latter assessment, the novel oral anticoagulants were most likely to be the best therapy with respect to each of the three major efficacy outcomes of stroke, ischaemic stroke or systemic embolism, and all-cause mortality.
On the other hand, aspirin appears to be the second worst (or ranked 7th out of 8, with only control therapy ranking lower) with a 92% likelihood of being the second worst therapy for the prevention of stroke, 79% chance of being the second worst for the composite of ischaemic stroke or systemic embolism, and 70% chance of being the 2nd worst for the endpoint of all-cause mortality.
However, comparisons between individual agents studied in only 1–2 trials should be cautiously interpreted. Table 3 clearly shows the benefit of oral anticoagulants, particularly new anticoagulants as a class, but in general, we believe that none of the individual novel agents should be considered superior to another in the absence of a direct comparison. Additionally, it is important not to overinterpret differences in the rankings based on simulations since small differences in results can result in large differences in rank.
Other network meta-analyses with indirect comparisons recently have been published, although we believe none were as robust as ours. Roskell et al30 performed a network meta-analysis to indirectly compare dabigatran with antiplatelets and placebo. Compared with placebo, dabigatran 150 mg twice daily significantly reduced the risk of any stroke, systemic embolism and mortality. Dabigatran 150 mg twice daily also significantly reduced the risk of any stroke compared with aspirin monotherapy, and also when compared with aspirin plus clopidogrel. Harenberg et al,31 performed a network meta-analysis of the three new oral anticoagulants from the three large phase III trials comparing dabigatran, rivaroxaban and apixaban with warfarin in patients with AF. Dabigatran 150 mg twice daily was superior in preventing ischaemic stroke plus systemic embolism compared with dabigatran 110 mg twice daily, and also compared with rivaroxaban, but was similar to apixaban. Apixaban had similar efficacy as compared with rivaroxaban, and also when compared with dabigatran (either dose), but less major bleeding than dabigatran (150 mg twice daily) or rivaroxaban. Intracranial haemorrhage occurred with similar frequency for all agents except for rivaroxaban (higher risk than dabigatran 110 mg twice daily). Baker and Phung32 conducted an adjusted indirect comparison meta-analysis between new oral agents in AF and concluded that significant differences in efficacy and safety parameters may exist between oral anticoagulant agents in patients with AF. In their analyses, dabigatran lowered the composite of stroke or systemic emboli, and reduced ischaemic stroke versus rivaroxaban. Meanwhile, apixaban lowered the risk of major and gastrointestinal bleeding versus dabigatran and versus rivaroxaban. Biondi-Zoccai et al.33 published a network meta-analysis using a fixed-effects model, including only new oral anticoagulants (including pilot studies) versus warfarin. They found that apixaban reduced: (1) stroke/systemic embolism compared with warfarin and both doses of dabigatran; and (2) major bleeding compared with warfarin, dabigatran 150 mg and rivaroxaban.
Other meta-analyses also have been published,34–37 however, these studies were less comprehensive than ours as they did not include data from the 17 trials with an antiplatelet or placebo as the comparator. Furthermore, few of the prior studies used a network meta-analysis approach that would permit indirect comparisons. By contrast, we included 20 trials summarising the totality of the evidence to date with antithrombotic therapy in AF, used a more appropriate and conservative random effects model, and performed a network meta-analysis that permitted indirect comparisons between agents.
Our analyses show that the novel oral anticoagulants and VKA are superior to placebo/control and antiplatelet drugs (as monotherapy or dual antiplatelet therapy) in reducing stroke and in lowering the risk of ischaemic stroke or embolism. However, only the VKAs were superior to ASA in reducing mortality. Since no large randomised studies have directly compared dabigatran or rivaroxaban versus ASA in patients with AF, the precision of the results in a network meta-analysis is reduced, and thus we believe such comparison must be cautiously interpreted. Nevertheless, our analysis was the only one to include all available phase III studies of anticoagulants and/or antiplatelet therapies in AF, which represents an important distinction from the aforementioned network meta-analyses.
While these novel oral anticoagulants offer several practical and clinical advantages over VKAs, they present new and different challenges for the clinician. First, there are no laboratory tests widely available to monitor the anticoagulant effects of these new drugs. Thus, there is no simple, widely available method to determine whether a patient is taking a novel anticoagulant, or to determine whether patients have subtherapeutic or supratherapeutic levels of anticoagulation. Second, the half-lives are prolonged in patients with renal insufficiency (particularly with dabigatran), and drug concentration also may be increased by several concomitant therapies that either inhibit hepatic metabolism or interfere with intestinal p-glycoprotein efflux.38 ,39 In the literature, reports of excess bleeding with dabigatran in elderly patients with renal dysfunction40 raise doubts regarding whether the superior safety profile observed with dabigatran in clinical trials will be generalisable to clinical practice. Finally, no proven antidote to dabigatran, rivaroxaban or apixaban has been tested clinically. Therefore, patients on one of these novel agents may be at a relative disadvantage compared with patients on VKAs in the case of trauma,41 need for emergency surgery, or in the case of life-threatening bleeding.
Nevertheless, concerns about bleeding need to be counterbalanced by the finding of substantial benefits seen in this meta-analysis with dabigatran 150 mg, rivaroxaban and apixaban when compared with warfarin or antiplatelet therapy. In the meantime, standard laboratory and point-of-care assays permitting rapid, easy, widely available assessment of anticoagulation with thrombin inhibitors and factor Xa inhibitors are urgently needed. Similarly, ongoing work to develop antidotes42 or therapies that could reverse the anticoagulant effects43 of these novel agents would be of great clinical interest. When interpreting the results of these analyses and considering how an individual patient should be treated, it is important for the clinician to consider and weigh the relative balance of thromboembolic events and risk of bleeding for the individual patient.
We recognise that our multiple treatment meta-analysis has several limitations. The trials were heterogeneous with respect to the characteristics of the patients, the definitions and assessment of some of the outcomes (particularly bleeding), the concomitant interventions allowed by the protocols, the length of, and differential loss to, follow-up, and the time frame during which the studies were conducted. The findings based on clinical trial populations may not be generalisable to clinical practice where patients tend to be older and have more comorbidities. There were limited numbers of some events (e.g, ischaemic strokes) and patients who received certain therapies across the 20 studies, which reduces the power to find differences and may affect the precision of our results. We included in our analyses older studies that used no antithrombotic or aspirin monotherapy, which may have diluted the observed benefits of NOACs compared with warfarin; these studies conducted 10–20 years ago were of lower quality, with less rigour in the design, definitions, time in therapeutic range, monitoring and blinded assessment of clearly defined endpoints, however, to analyse the wide variety of treatments (including no antithrombotic therapy and aspirin monotherapy). Lastly, we did not adjust for multiple comparisons, and thus, difference of marginal statistical significance should be interpreted cautiously.
Conclusion
In this meta-analysis of direct and indirect comparisons of antithrombotic therapies in patients with AF using a multiple treatment meta-analysis, we found that novel anticoagulants lowered the rates of stroke and the composite of ischaemic stroke or systemic embolism and mortality compared with antiplatelet regimens. Rankings from model simulations demonstrated that anticoagulants were superior to antiplatelet-based regimens, and that the novel anticoagulants were at least as good as, if not more effective than, VKAs.
Acknowledgments
We would like to recognise the many contributions in the field of cardiovascular medicine made by Dr Ernesto Paolasso prior to his untimely death during the final stages of preparation of this manuscript.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online table
Footnotes
-
Contributors The manuscript has been reviewed and approved by all authors for submission. All authors have contributed substantively to the work, including participation in conception/design or analysis/interpretation, drafting and critical revision, and final approval. Please note that EPo had closely participated in all aspects of the manuscript preparation until his untimely death earlier this year, as described in the acknowledgements.
-
Competing interests Robert Giugliano: Dr Giugliano is a member of the TIMI Study Group, which has received research grant support from Johnson & Johnson and from Daiichi-Sankyo related to clinical trials of anticoagulants. Dr Giugliano has received honoraria for consultation/lectures from Bristol-Myers Squibb, Daiichi-Sankyo and Johnson & Johnson.
-
Ethics approval Ethics Committees at each centre for each trial.
-
Provenance and peer review Not commissioned; externally peer reviewed.