Adding a New Anticoagulant or Antiplatelet Agent for Patient Receiving Aspirin after an Acute Coronary Syndrome?--Results from a Pairwise and Network Meta-Analysis of Randomized-Controlled Trials

Objectives: To synthesize the efficacy and safety outcomes from randomized-controlled trials regarding new oral anticoagulant, protease-activated receptor-1 (PAR-1) antagonist, and warfarin adjunctive to aspirin for patients after acute coronary syndrome (ACS) via pair-wise and network meta-analyses. (difference in incidence comparing to aspirin lone) showed new oral anticoagulants (-0.1974, [-0.284, -0.111]) and PAR-1 antagonists (-0.1239, [-0.215, -0.033]) to besuperior to warfarin (-0.1004, [-0.166, -0.035]) in the occurrences of MAE whereas PAR-1 antagonists (0.4292, [0.2123, 0.6476]) afforded better outcomes in major bleeding events against warfarin (0.5742, [0.3889, 0.7619]) and new oral anticoagulants (1.169, [0.8667, 1.485]). Conclusion: Based on the study results, we cannot recommend the routine administration of new oral anticoagulant as add-on treatment for patients after ACS. However, for ACS patients comorbid with atrial fibrillation, new oral anticoagulant might be superior to warfarin in both efficacy and safety outcomes.


INTRODUCTION
Cardiovascular disease is the number one cause of death globally (WHO). In particular, the Global Burden of Disease Study classified ischemic heart disease as the leading cause of global mortality, accounting for 1.4 million deaths in the developed world and 5.7 million deaths in the developing regions [1]. At the same time, antiplatelet therapy is the foundation therapy for the prevention and treatment for arterial thrombosis. Aspirin is known to reduce, by approximately 25%, the risk of any serious vascular accident, with greatest protection among patients with Acute Coronary Syndromes (ACS) [2]. Besides aspirin, clopidogrel [3][4][5][6], glycoprotein IIb/IIIa inhibitors [7,8] and new antiplatelet agents [such as prasugrel [9] and ticagrelor [10] are also effective in the management of patients with ACS. However, even with these powerful antithrombotic treatments (deactivation of P2Y12 ADP receptor and TxA2-related activation pathways at the same time), the recurrent ischemic events in ACS patients are still high) with an occurrence of greater than 11% [9,10]. This is in spite of the fact that P2Y12 receptor has a well-established role as antithrombotic agents in treating cardiovascular diseases [11], although the potent platelet inhibition by P2Y12 receptor might be associated with increased level of miR-223 [12]. Furthermore, a few studies also indicated that some polymorphisms are linked to the action of antiplatelet drugs that might increase the risk of cardiovascular and cerebrovascular events [13,14]; and a series of risk factors has been identified to be associated with the increased risk of acute coronary syndrome [15].
Meanwhile, it was reported that excess thrombin generation persists beyond the acute presentation after an ACS [16]. As a result, anticoagulants have been shown to be as effective as antiplatelet therapy in the long-term management of coronary artery disease [17]. Most importantly, the effectiveness of combination therapy (antiplatelet and anticoagulant) has been demonstrated for both initial and long-term therapy for ACS patients [18].
In recent years, the efficacy and safety of several new oral antiplatelet Thrombin-receptor antagonist, [TRA or protease-activated-receptor  antagonist [19][20][21][22]] and anticoagulant agents [23][24][25][26][27][28][29] have been assessed in a series of phase II and III clinical trials. While majority of the trials claimsa more desirable improvement in recurrent ischemic events after an ACS, the safety results, primarily the bleeding risks, are not consistent across studies. Furthermore, two meta-analyses of new oral anticoagulants suggested that the increased risk of major bleeding might offset the benefits in reduction of ischemic events [30,31]; while another metaanalysis concluded that new oral anticoagulant may be the optimal antithrombotic regime for patients with ACS [32]. Hence, the inconsistency in the safety outcome needs to be ascertained when adding new antithrombotic agent after an ACS.
Beyond the safety concern, clinicians would also need knowledge on how to choose among these competing antithrombotic agents. Nevertheless, the results from those clinical trials are heterogeneous in terms of different efficacy outcomes or they are under power (e.g. phase II trials) to detect the statistically significant difference. Furthermore, there is no head-tohead comparison between these newly invented antiplatelet and anticoagulant agents. Last but not least, the efficacy and safety of newer generation oral anticoagulants were unanimously investigated based on placebo-controlled trials rather than warfarin-controlled ones. In order to address all these, we performed a pairwise and indirect meta-analysis comparing a series of new antithrombotic agents to ascertain the therapeutic and safety outcomes and to aid clinicians to optimize the treatment for patients with ACS history.

Data Sources
We conducted an electronic literature search for all prospective randomized controlled trials evaluating the efficacy and safety of a newly invented anticoagulant or antiplatelet agent for patients receiving aspirin after an ACS in Medline, Embase, Cochrane Library, Web of Knowledge, and Scopus from inception to 25 th June 2013. In addition, the reference list of identified articles was manually searched as well. The following key search terms were used: acute coronary syndrome or myocardial infarction, randomized controlled trial or (double-blind) controlled trial with one of the following terms: antithrombins, factor Xa inhibitor, oral anticoagulation, apixaban or edoxaban or darexaban or rivaroxaban or otamixaban, and dabigatran or argatroban or ximelagatran or warfarin or thrombin-receptor antagonist, protease-activated-receptor (PAR-1) antagonist, atopaxar, vorapaxar.

Inclusion Criteria
1． Studies should be reported in English and the full text could be retrieved. 2． All participants in the study should be explicitly diagnosed with ACS or at least included a subgroup of patients diagnosed with ACS. 3． Double-blind study should contain a placebo-controlled arm. 4． Patients were treated with aspirin or plus another thienopyridine. 5． Study should at least present the results regarding the Major Adverse Events (a composite of death, severe recurrent ischemia, myocardial infarction, and ischemic stroke) and incidences of major bleeding in each arm.

Data Extraction
Two reviewers independently extracted the data from each retrieved study. Discrepancy was resolved by thorough discussion and only agreed data were incorporated into the meta-analysis. The primary efficacy outcome was defined as the major adverse events (MAE), including a composite of all cause death, severe recurrent ischemia, myocardial infarction and ischemic stroke. The main safety endpoint was the major bleeding events (MB), according to the definition of each trial. The efficacy outcome was estimated based on Intention-to-Treat population, and the safety outcome assessed on safety population.

Direct meta-analysis
Both pairwise and indirect meta-analyses were performed to pool the efficacy and safety data. For the pairwise meta-analysis, given the different drugs and treatment effects, randomeffects model was adopted to assess the effect size. The reported event frequencies were used to calculate Odds Ratios (OR) with 95% Confidence Interval (95% CIs). Log odds ratios were pooled with inverse variance weighting. The degree of inconsistency across studies was quantified using the I 2 statistic [33]. The Cochrane Q heterogeneity test (Χ 2 test) was also applied. These data were reported as I 2 percentages, along with P values from the Χ 2 test. The pairwise meta-analysis was performed via Revman 5.1.

Network meta-analysis
In a network meta-analysis, treatment effects are calculated for all treatments using all available evidence in one simultaneous analysis. This method builds on the principles of indirect comparisons and preserves the randomized comparisons within each trial. Particularly, at first, the models were fitted to the data using the Bayesian Markov chain Monte Carlo simulations, utilizing the WinBUGS 1.4.3 (MRC Biostatistic Unit, Cambridge, UK) via both fixed-and random-effects models. WinBUGS code for network meta-analysis of dichotomous and standard Bayesian random-effects meta-analysis was adapted from code developed by the NICE Decision Support Unit [34].
Since all the clinical trials recruited patients receiving aspirin alone or aspirin plus another thienopyridine at the baseline, all the treatment comparisons could be viewed as anchored on aspirin treatment against another anticoagulant/antiplatelet agent and therefore formed the comparison network, subsequently enabling the indirect comparisons among those drugs. Comparisons were presented throughout using aspirin as reference treatment ( Supplementary Fig. 1).
The likelihood of publication bias was assessed visually by generating a funnel plot for the primary efficacy end point. A p value<0.05 was considered statistically significant.

Model fit
Since the mean residual deviances provided an estimate of how well the values predicted by the model fit the observed dataset, for an adequate model fit, the sum of the residual deviances should be approximately equal to the total number of study arms in the observed the dataset. In addition, deviance information criterion (DIC) was recorded by the WinBUGS to appraise the model as well. The model with the lowest value of DIC would best predict a replicate dataset of the same structure as currently observed.

Baseline treatment effect
To assess the absolute effect of each treatment, a baseline model that represents the absolute natural history under a standard treatment in the comparator setting was developed separately. Thus, both the posterior and predictive distributions of baseline treatment effect could be obtained to model the baseline response. It has been reported that predictive distribution for a new baseline incorporates the uncertainty about the value a new observation might take, as well as the observed variation in the data [34]. It is however important to ensure that the uncertainty conveyed by the predictive distribution reflects genuine uncertainty in the baseline. Therefore, the baseline effect was drawn from both posterior and predictive distribution, whereas the results from the predictive distribution were adopted in the subsequent computation.

Effect of covariates
As various follow-up times and ages of subjects in the included RCTs might have influenced on the treatment, these two factors were modelled as covariates in the network meta-analysis. The study-level data were taken into account by the following three constructed models: (I) follow-up times and baseline age of subjects: the model encompassed two study level continuous variables to adjust for the effects they may have. X years was a covariate centred at mean followup across the studies, thus the coefficient βyears estimated the incremental difference (above or below) in (log) treatment effect for each year from the average follow-up across studies. Similarly, the Xage was covariate centred at mean age, such that the coefficient βage estimated the incremental difference (above or below) in the (log) treatment effect for each year from the average age across studies [35]. (II)Follow-up times: the model only included this variable to adjust for the time point at which the response was measured (in years). (III) Baseline age of subjects: this covariate model included this continuous variable to adjust for differences in patient age across studies. Again, the fixedand random-effects models were both utilized to examine the difference in results.

Indirect comparison between warfarin, new oral anticoagulants and antiplatelet agents (PAR-1 antagonist)
The effect of each treatment category (grouped into warfarin, new oral anticoagulants and PAR-1 antagonists) was first synthesized individually via pairwise meta-analysis. Then the probability that each treatment strategy has the most preferable incidences in terms of MAE or major bleeding was estimated via network meta-analysis using aspirin as the reference treatment. Likewise, the baseline treatment effect was computed based on the afore-mentioned method. Thus, the absolute effect of specific treatment strategy could be calculated subsequently.

Description of the Included Studies
Publication bias was assessed via the funnel plot, except for PAR-1, the other two antithrombotic agents did not show significant publication biases based on the results of major adverse events (Appendix-Figs. 1 to 3).
For the studies assessing new oral anticoagulants, number of subjects varied from 1279 [28] to 15526 [26], and average age from 57 [25] to 68 years [29]. All the included studies were with at least 6 months of follow-up and recruited more male than female patients. In defining bleeding events, four out of seven studies defined the major bleeding according to the International Society on Thrombosis and Haemostasis (ISTH) [23,[27][28][29] whereas the other three used the definition based on the Thrombosis in Myocardial Infarction (TIMI) Trial [24][25][26].
Four RCTs investigated the efficacy and safety effect of PAR-1 antagonist (atopaxar and vorapaxar) with sample sized varying from 117 [20] to 12944 [22] and follow-up times ranging from 8 weeks [20] to 502 days [22]. All four studies utilized the TIMI criteria to define the key safety endpoint.
Eleven RCTs assessed warfarin against aspirin alone for ACS patients. Specifically, the number of subjects varied from 57 [47] to 8803 [36] with mean age from 57 [47] to 67 years [45]. The length of follow-up was between 2.5 months [47] and 5 years [43]. Each study defined the major and minor bleeding based on their own criteria (other than TIMI and ISTH).

New oral anticoagulant vs Placebo
For the primary efficacy endpoint, the new oral anticoagulant produced moderately better outcomes than placebo, with OR of 0.85 (95% confidence interval: 0.78, 0.93). However, the major bleeding incidence was considerably higher than the placebo, with OR of 3.04 (95%CI: 2. 21, 4.19). No heterogeneity was detected in the efficacy and safety outcomes, with I 2 of 0% respectively ( Table 2).
Except for overall synthesis, we also estimated the OR for each kind of the oral anticoagulant in order to compare with the result from the indirect comparisons where applicable. As a result, the ORs (95% CI)

Warfarin vs Placebo
The pairwise meta-analysis for warfarin showed that the OR for MAE was similar to that for new anticoagulants of 0.87 (95% CI: 0.74, 1.02), while generating a moderate increase in MB with of 1.77 (95%CI: 1.46, 2.14). Again, for the efficacy outcome, medium heterogeneity was detected (I 2 =62%) ( Table 2).

Major adverse events
In terms of the absolute effect, both the random and fixed-effects models showed similar results. More specifically, ximelagatran (ORs: 0.1243 and 0.1227 respectively) afforded the best profile in the incidences of MAE, whereas darexaban (0.1918 and 0.1903 respectively) had the highest occurrences in this outcome. In addition, dabigatran, rivaroxban and vorapa generated lower MAE incidence compared to the other antithrombotic agents from those two models. The comparative effects were consistent with the absolute effects as well. For instance, ximelagatran (-0.3044 and -0.3039), dabigatran (-0.2144 and -0.2173) and rivaroxban ( and -0.2161) were superior to darexaban (0.2777

Flow diagram of study selection
generating a moderate increase in MB with OR Again, for the efficacy outcome, medium heterogeneity was In terms of the absolute effect, both the randomeffects models showed similar results. cifically, ximelagatran (ORs: 0.1243 and 0.1227 respectively) afforded the best profile in the incidences of MAE, whereas darexaban (0.1918 and 0.1903 respectively) had the highest occurrences in this outcome. In addition, dabigatran, rivaroxban and vorapaxar also generated lower MAE incidence compared to the other antithrombotic agents from those two models. The comparative effects were consistent with the absolute effects as well. For instance, 0.3039), dabigatran 0.2173) and rivaroxban (-0.2179 0.2161) were superior to darexaban (0.2777 and 0.2788) in lowering the MAE incidences. From the random-effects model, ximelagatran (32.3%) followed by dabigatran (24.0%) might have higher probability than the other dru reducing the incidences of MAE.
The total residual deviance and DIC values both favoured random-effects model for the indirect comparisons, with lower values in DIC (365.89) and total residual deviance (57.8) than the fixed effects model (374.11 and 73.09 respectively) ( Table 3).

MB
Both the random and fixed effects models showed darexaban generated the highest risk of major bleeding, whereas vorapaxar followed by warfarin, ximelagatran and apixaban produced improvement for the same outcome adjusted for baseline treatment effect, the absolute effect also produced identical results, with vorapaxar having the lowest risk of major bleeding (other than aspirin alone). However, in , 2015; Article no. BJMMR.2015.194 and 0.2788) in lowering the MAE incidences. effects model, ximelagatran (32.3%) followed by dabigatran (24.0%) might have higher probability than the other drugs in The total residual deviance and DIC values both effects model for the indirect comparisons, with lower values in DIC (365.89) and total residual deviance (57.8) than the fixednd 73.09 respectively) Both the random and fixed effects models darexaban generated the highest risk of major bleeding, whereas vorapaxar followed by warfarin, ximelagatran and apixaban produced improvement for the same outcome. After adjusted for baseline treatment effect, the absolute effect also produced identical results, with vorapaxar having the lowest risk of major bleeding (other than aspirin alone). However, in terms of the total residual deviance and DIC, these two values in the random effects model (mean 44.11 and 262.76) were substantially lower than those from the fixed effects model (51.5 and 265.40), indicating the results from the former model should be adopted. Additionally, except for the aspirin treatment alone, dabigatran (17.0%) and atopaxar (16.8%) might be more likely to have the best profile of major bleeding risk (Table 4).

Major adverse events
Three models were constructed to investigate the impact of covariate on the MAE incidences. However, the results did not change substantially when adjusted for these factors together or separately. Again, according to the DIC and total residual deviance, the results from the randomeffect models should be preferred (Appendix).

Major bleeding
Similarly, the results from the three models did not contradict each other, but unanimously indicated vorapaxar as having the best major bleeding risk profile and darexaban having the highest major bleeding risk. Again, the results favour the random-effects model (Appendix).
Indirect comparison among aspirin, new oral anticoagulants, warfarin and new antiplatelet agents Results from both absolute and comparative effects favoured new oral anticoagulants and PAR-1 antagonists over warfarin or aspirin alone in terms of MAE. However, the major bleeding incidences showed the PAR-1 antagonist as superior to either new oral anticoagulants or warfarin, with the highest bleeding risk borne by new oral anticoagulants. Though with slightly different results, the DIC and total residual deviance did not vary substantially between random-and fixed-effects models for both MAE and major bleeding results. Both models unanimously showed new oral anticoagulants having the highest probability of being the best (85.4% vs 32.8%) in terms of efficacy end point, and treatment with aspirin alone being the best as to the safety outcome (100% vs 33%) (Tables  5 and 6).

DISCUSSION
Overall, both pair-wise and network metaanalyses showed add-on newer antithrombotic agent might be more preferable to aspirin treatment alone but with higher risk of major bleeding. The main findings from the present study were, first, when considering MAE and MB together, vorapaxar and rivaroxban were associated with lower incidences of these two events than the other antithrombotic agents included in the study. Although ximelagatran also showed improved occurrences, the potential liver toxicity refuted its further use in the market [29]. Second, from the indirect comparisons between all the new oral anticoagulants, warfarin and PAR-1 antagonists, the new oral anticoagulants were more likely to decrease the MAE incidences but with highest risk of MB. Alternatively, PAR-1 antagonists were able to lower the occurrences of MAE with lowest risk of MB. Nonetheless, it was worth noting that there was a slight difference in the baseline antiplatelet treatment. For majority of the RCTs, dual antiplatelet treatment was allowed whereas all the studies for warfarin and one RCT for ximelagatran [29] only enrolled subjects on single antiplatelet therapy (aspirin). This would introduce a confounding factor for the final result as it is recommended that all the post-ACS patients should receive dual antiplatelet treatment up to 12 months [49].
Since the absolute effect of each treatment/category in terms of MAE and MB was estimated in the network meta-analysis simultaneously, the risk-benefit analysis could be conducted subsequently. In consistent with the results from the meta-analyses, except for ximelagatran and vorapxar, all the other antithrombotic agents did not achieve a positive net benefit per 10000 ACS patients treated. This revealed the substantial increase in bleeding risk compared to the decrease in MAE when adding the antithrombotic agent to aspirin. Particularly, compared to aspirin alone, each treatment strategy would generate more major bleeding events than the avoided major adverse events (Net bleeding events: 144 for all the new oral anticoagulant, 42 for PAR-1 antagonist and 130 for warfarin per 10000 ACS patients treated). In terms of the MB specifically, the increase in MB incidences varied from 37.6% for vorapaxar to 119% for deraxaban when using aspirin as comparator, which was fairly discouraging. In contrast, the relative increase in annual rate of TIMI non-CABG bleeding events was approximately 25% with the introduction of new P2Y12 ADP-receptor antagonist [9,10]. Since the prognostic significance of bleeding complications is as serious as the occurrence of ischemic events after percutaneous coronary intervention, these results would partly discourage its future extensive use in the clinical setting.
Covariates exerted little impact on the final results according to the covariate analysis. It is important that network meta-analysis has the underlying assumption that trails and outcomes are sufficiently similar to allow the data to be pooled, and the consistency assumption relies on there being no imbalance in modifiers of relative treatment effects across studies. In our network meta-analysis, the similarity assumption was supported by the inclusion criteria for study selection and also the adjustment of the results by way of covariate analyses for the potential effect modifiers, e.g. length of follow-up, baseline age of subjects. Specifically, the covariate analysis aimed to reduce the impact of any bias due to similarity and/or consistency violations [35]. The results of our covariate analyses showed the assumptions for network metaanalysis were not violated even with different follow-up times and baseline age of subjects across studies.
At present, the use of oral anticoagulants in patients with ischemic heart disease is commonly restricted to individuals at high risk of thromboticembolic complications, those with atrial fibrillation, valve prostheses, intracardiac thrombi, recurrent thrombo-embolism, or antiphospholipid syndrome [7,50,51]. As demonstrated in a couple of RCTs, the use of apixaban or dabigatran was associated with a decreased risk for thromboembolic events compared with warfarin or low-molecular-weight heparins [52,53]. Nevertheless, the use of new oral anticoagulant on the top of antiplatelet treatment for patients without aforementioned comorbidities after ACS is not recommended yet [54]. From our present study, the excess bleeding risk of new oral anticoagulant refutes its unrestricted use in combination with other antiplatelet therapy (although ximelagatran yielded a positive net benefit, its liver toxicity has rendered discontinuation of further investigation).
Except for warfarin, it should be noted that the large majority of RCTs were phase II trials (8 out of 11). However, the primary results from the phase II trials were in agreement with the phase III trials for the particular agent. The bleeding incidences in these three large trials were similar to the phase II experiments corresponding to each drug. Furthermore, the bleeding outcomes were also consistent between two phase III trials for apixaban (2.55 [1.48, 4.41]) and rivaroxaban (3.92 [2.43, 6.33]). Even with nominally larger relative increase in bleeding in rivaroxaban trial [26] than the prematurely discontinued apixaban trial [24], these were substantially greater than the results derived from another phase III trial pertaining to vorapaxar (1.55 [1.24, 1.92]) [22]. Considering the statistical power and longer follow-up of phase III trials, this could be viewed as evidence to support the results from our network meta-analysis.
Since the occurrences of recurrent ischemic events are still strikingly high for patients after ACS even with standard dual antiplatelet therapy, it is imperative to consider more potent antiplatelet/anticoagulant agent to reduce these events. With the availability of newly invented oral anticoagulant and antiplatelet, clinicians would need the evidence to assist them to choose among a variety of options. As efficacy and safety endpoints are of equal importance for this cohort, decision should be made to optimize these two outcomes. From the evidences of our present study, adding an antiplatelet agent (e.g. vorapaxar) would be much better than adding an anticoagulant to improve both the incidences of recurrent ischemic and major bleeding events for patient with ACS history. This is supported by the consistency in results between pairwise and network meta-analyses. For patients with a combination of ACS and atrial fibrillation, where a combination of anticoagulant and dual antiplatelet treatment is currently recommended, the information from our present study might be very helpful. First of all, from our network metaanalysis, it was indicated that ximelagatran, dabigatran and rivaroxaban were superior to warfarin in improving MAE. Additionally, the profile of major bleeding was comparable among ximelagatran, apixaban, rivoroxaban and warfarin. But the new oral anticoagulants are able to provide more reliable effect without the need for laboratory monitoring, and might be an ideal replacement of warfarin. Hence, for ACS patient comorbid with atrial fibrillation, new oral anticoagulant like dabigatran, apixaban and rivoroxban might be more appropriate to be prescribed.
However, several study limitations needed to be addressed. Firstly, large numbers of studies were excluded due to shorter follow-up times and small sample size. Second, lack of head-to-head comparisons between the included agents also necessitates caution in interpretation of our results. Third, the definitions of MAE and major bleeding events were heterogeneous across studies, which may explain the heterogeneity of efficacy endpoint in the pair-wise meta-analysis.

CONCLUSION
From the results of the network meta-analysis, ximelagatran, dabigatran, rivoroxaban and vorapaxar were identified to be superior to the other antithrombotic agents in MAE incidences for patients with ACS histories. The combined comparisons showed new oral anticoagulants and PAR-1 antagonists to be superior to warfarin in the occurrences of MAE for the same cohort whereas PAR-1 antagonists afforded optimum outcomes in the events of major bleeding against warfarin and new oral anticoagulants. Therefore, the routine administration of new oral anticoagulant as add-on treatment for patients after ACS might not be recommendable due to its increased bleeding risk. Nonetheless, for ACS patient comorbid with atrial fibrillation, new oral anticoagulant might be superior to warfarin in both efficacy and safety outcomes. Future headto-head RCT comparing new oral anticoagulant with new antiplatelet is needed to testify the results from our network meta-analysis.

CONSENT
Not applicable.

ETHICAL APPROVAL
As the current systematic review and metaanalysis was based on published data sources with original approval from individual ethical committees, no ethical issue was involved.