An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Edoxaban (Lixiana) [Internet]. Ottawa (ON): Canadian Agency for Drugs and Technologies in Health; 2017 Apr.
The aim of this section is to review and critically appraise any indirect comparisons (IDCs) that compare edoxaban 60 mg (30 mg reduced for reduced kidney function) once daily with any appropriate comparison in the prevention of stroke and systemic embolic event (SEE) in patients with nonvalvular atrial fibrillation (NVAF).
Edoxaban has been compared with warfarin in the ENGAGE AF-TIMI 48 trial.9 However, no direct evidence exists that compares edoxaban with other new direct oral anticoagulants (DOACs). Therefore, IDCs that include edoxaban could provide information on the comparative effectiveness and safety of this drug and existing therapies and would be relevant to this CADTH Common Drug Review.
One IDC submitted by the manufacturer was reviewed and critically appraised. In addition, a comprehensive literature search was performed by an information specialist to identify published IDCs. Identified IDCs from the literature were summarized and contrasted with the manufacturer’s IDC. The details of the literature search are available in 0.
In addition to the submitted manufacturer’s IDC, the literature search identified seven published IDCs. A description of the research question from each study has been described in Table 21.
PICO Description of Identified Indirect Comparisons.
The objective of the manufacturer’s IDC was to analyze the comparative efficacy and safety of edoxaban 60 mg (30 mg reduced for reduced kidney function) once daily compared with other treatments for stroke and SEE prevention in NVAF patients. Specifically, the IDC aimed to answer the following research question: “What are the relative effectiveness and safety of edoxaban compared to other DOACs in patients diagnosed with nonvalvular atrial fibrillation in need of anticoagulation for stroke prevention?”8
The importance of understanding the relative efficacy and safety of DOACs for clinicians and policy-makers was used as a rationale for conducting this IDC. Although previous IDCs do exist and have been published, the manufacturer IDC claims that these IDCs reported in the literature do not adequately account for differences in trial methods — an issue that the manufacturer IDC attempted to address.
Inclusion criteria for the manufacturer’s IDC were patients diagnosed with NVAF, receiving vitamin K antagonists (VKAs) or a DOAC, in a randomized controlled trial (RCT) setting of 12 or more weeks and reported on stroke, SEE, and safety-related outcomes. Studies were not limited by language or date of publication. Specific exclusion criteria included studies with heparin, ximelagatran, and betrixaban as interventions or comparators.
The manufacturer conducted a systematic literature search of more than three bibliographical databases (PubMed, Embase, and the Cochrane Library). The search was last updated on January 13, 2016.
Retrieved citations were screened by two independent reviewers according to predefined eligibility criteria. Discrepancy between the reviewers was handled by a third independent reviewer. Citations went through two stages of screenings: first, at the title or abstract level and, second, if the citation was relevant, a full-text screen.
Two independent reviewers conducted data extraction. Any discrepancy was managed through a third independent reviewer. Trial characteristics, patient demographics, disease condition, intervention, and outcomes-related information were extracted.
All relevant comparators were included in the manufacturer IDC, including VKAs (warfarin, fluindione, phenprocoumon, and acenocoumarol) and DOACs (edoxaban, dabigatran, rivaroxaban, and apixaban). Aspirin was included as a comparator in an extended network presented in an appendix.
The following efficacy outcomes were included in the manufacturer’s IDC:
The producer of the manufacturer’s IDC used criteria based on guidance from the Institute for Quality and Efficiency in Health Care. The manufacturer’s IDC did not specify any specific action to be taken if an included study was of low quality (i.e., had a high risk of bias).
The manufacturer’s IDC provided two graphical representations of the evidence network: Figure 2: depicts the full-study population; Figure 3: depicts an evidence network with a restricted ≥ 2 CHADS2 (congestive heart failure, hypertension, age ≥ 75 years, diabetes mellitus, and prior stroke or transient ischemic attack [TIA] or thromboembolism) score.
Evidence Network of Full-Study Populations.
Evidence Network of CHADS2 ≥ 2 Score Population.
The manufacturer’s IDC reported using a Poisson regression model frequentist approach for the main analysis, with standard adjusted-dose VKA as a reference for analyses. In the Poisson model, treatments were considered as fixed effects and studies as random effects, and the total number of events and the number of person-years of exposure for each intervention group was modelled within each of the included studies. The overall approach has been conducted under the fixed-effects model assumptions, with a rationale that because of the limited number of included studies, a random-effects model would produce inaccurate and non-generalizable results. For efficacy analysis, the producers of the manufacturer’s IDC attempted to use the intention-to-treat population whenever possible and the modified intention-to-treat or safety population for analyses related to safety outcomes. The producer of the manufacturer’s IDC restricted the primary analysis to a patient population with a CHADS2 score ≥ 2. Subsequent analysis with the whole population set was conducted. In addition, based on possible heterogeneity or uncertainty, several subgroup analyses were conducted, but the producer was not able to restrict the analysis to a patient population with a CHADS2 score ≥ 2. These subgroups were for subpopulations of patients older than 75 years, renal function impairment and prior experience with VKA, time in therapeutic range (TTR) ≥ 60%, and a history of stroke or TIA at the baseline. A secondary network meta-analysis (NMA), which included Aspirin as a comparator, was reported in an appendix. The producer used a mixed log–binomial regression model for the secondary NMA as treatment follow-up was not available for the Aspirin trials. Relative risks with a corresponding 95% confidence interval (CI) were provided as an effect measure.
The producer of the manufacturer’s IDC adjusted the result of the outcomes of the RE-LY dabigatran versus warfarin trial for its open-label design. The producer of the manufacturer’s IDC utilized a published systematic review and meta-analysis that compared the pooled results of open-label and double-blind trials of DOACs versus VKA for the prevention of stroke and SEE in patients with atrial fibrillation.39 It is not clear how the producers of the manufacturer’s IDC employed the differences found in the results of the pooled double-blind versus open-label trials in their analysis. The producer of the manufacturer’s IDC could not perform any heterogeneity or inconsistency quantitative assessment as all comparisons (arms) were made of a single trial and no closed loop was available.
Four phase III trials of DOACs were included in the primary NMA. Five unique interventions were included in the network: a once-daily edoxaban 60 mg (30 mg reduced-dose) regimen, a once daily edoxaban 30 mg (15 mg reduced-dose) regimen, a standard adjusted-dose VKA therapy, a once daily rivaroxaban 20 mg (15 mg dose reduced-dose) regimen, a twice daily dabigatran 150 mg regimen, a twice daily dabigatran 110 mg regimen, and a twice daily apixaban 5 mg (2.5 mg reduced-dose) regimen. Results from the 30 mg edoxaban (15 mg reduced-dose) regimen were not reported.
Of the included four trials in the primary NMA, three employed a double-blind design, and one (RE-LY) was of open-label design, in which the producer of the manufacturer’s IDC used the Lega 2013 study to attempt an adjustment of the design issue.39 Mean patient age and the proportion of female patients were similar across the studies. The mean CHADS2 score was highest in the ROCKET-AF trial (3.5) and lowest in the RE-LY and ARISTOTLE trials (2.1 in both), reflecting that the inclusion criteria regarding CHADS2 score was not consistent across the included trials. Patients in the ROCKET-AF trial had an increased prevalence of diabetes, prior stroke or TIA, and heart failure than in the other studies. The differences in the proportion of patients with prior VKA exposure slightly varied across the trials. The manufacturer IDC reported that the assessment of the risk of bias for each trial showed a low risk of bias.
Clinical outcomes, Table 22, showed several statistically significant results when comparing edoxaban 60 mg (30 mg reduced dose) with standard adjusted-dose VKA therapy in favour of edoxaban. Specifically, the composite outcome stroke and SEE, all-cause mortality, and CV mortality were all in favour of edoxaban, but at the higher end if the 95% CIs in these outcomes were very close to 1. When comparing edoxaban with other DOACs, no statistically significant result was obtained. Subgroups analysis for the composite outcome stroke and SEE lost its statistical significance in the edoxaban-VKA comparison with the subpopulation of patients with TTR ≥ 60%, with the subpopulation of patients > 75 years of age, with the subpopulation of patients with renal insufficiency (creatinine clearance [CrCL] 30 mL/min to 50 mL/min), and with the subpopulation of patients with prior VKA exposure. However, the composite outcome stroke and SEE gained statistical significance in favour of edoxaban when compared with rivaroxaban in the subpopulation of patients with renal insufficiency (CrCL 30 mL/min to 50 mL/min). The rest of the subgroup analysis showed similar results to the primary analysis.
Clinical Outcomes of Edoxaban Versus Other Anticoagulation Therapies in the Manufacturer’s IDC.
Safety outcomes (Table 23) showed edoxaban to be statistically significantly better than VKA in terms of major bleed, intracranial bleed, fatal bleed, and clinically relevant non-major bleed. However, the safety outcomes also showed edoxaban to be statistically significantly more harmful than VKA therapy in terms of major gastrointestinal bleed. In addition, edoxaban was statistically significantly better than dabigatran 150 mg, but not 110 mg, in terms of major bleed. When compared with rivaroxaban, edoxaban showed a statistically significantly improvement in terms of major bleed, major gastrointestinal bleed, and clinically relevant non-major bleed. Subgroups analysis showed the outcome of major bleed losing its statistical significance in the edoxaban-VKA comparison in the subpopulation of patients with TTR ≥ 60%, the subpopulation of patients with renal insufficiency (CrCL 50 mL/min to 80 mL/min), and the subpopulation of patients with prior stroke. Comparison with rivaroxaban also lost its statistical significance status in major bleed in the subpopulation of patients with prior stroke and the subpopulation of patients with no previous VKA exposure. Along the same lines, the major bleed comparison with dabigatran 150 mg lost its statistical significance in the subpopulation of patients with TTR ≥ 60%, the subpopulation of patients with renal insufficiency (CrCL 50 mL/min to 80 mL/min), the subpopulation of patients with prior stroke, and the subpopulation of patients with no previous VKA exposure. In contrast, the comparison of edoxaban with dabigatran 110 mg in terms of major bleed gained significance when the population was restricted to CHADS2 patients. Also, the major bleed comparison of edoxaban with apixaban gained statistical significance in favour of apixaban in the subpopulation of patients with TTR ≥ 60% and the subpopulation of patients with prior VKA exposure.
Safety Outcomes of Edoxaban Versus Other Anticoagulation Therapies in the Manufacturer’s IDC.
The manufacturer’s IDC provided research questions that incorporated clear population, intervention, comparisons, and outcomes. The population and the comparisons reported in the manufacturer’s IDC are relevant to the Canadian setting. Also, the outcomes reported in the manufacturer’s IDC were relevant to this review and to the assessment of the comparative clinical efficacy and safety of edoxaban. The literature search strategy employed in the manufacturer’s IDC was comprehensive and covered several bibliographical databases. Screening and data extraction were carried out in duplicate, providing confidence in the accuracy of the extracted data. The included trials were thoroughly assessed for quality and have proven to be of sufficiently good quality. In addition, the manufacturer’s IDC transparently reported the characteristics of the included trials. The assessment of the characteristics of the included studies shows similar values across the baseline characteristics. An exception was the ROCKET-AF trial having the highest mean CHADS2 score and a higher proportion of patients with diabetes, prior stroke or TIA, and heart failure than in the other studies. A further exception was the variation in the proportion of patients with prior VKA exposure across trials and the open-label study design of the RE-LY trial, which the producer of the manufacturer’s IDC attempted to adjust for.
The choice of the overall statistical method used in the analysis (frequentist Poisson regression model) is valid and would provide results that are similar to the Bayesian NMA approach. In the data analysis, the producer of the manufacturer’s IDC chose to use the fixed-effects model to build the interpretations and conclusion. The fixed-effects model offers less generalizability than the random-effects model and is based on an assumption that all trials share a common true treatment effect, regardless of any differences in the study or patient characteristics. The manufacturer’s IDC argued that the choice to use the fixed-effects model was based on constraints imposed by the nature of the evidence network, which only provided one trial in the direct assessment of any two connected comparisons. The manufacturer’s IDC argued that a random-effects model provides no viable results and that when a random-effects model was attempted, all the comparisons that were statistically significant under a fixed-effects model lost their statistical significance. Although the argument for the choice of the fixed-effects model is valid, the limitations imposed on the results will lead to a possible inflated type I error and uncertainty in the accuracy of the results.
The producer of the manufacturer’s IDC attempted to adjust for the open-label study design in the RE-LY trial using a published systematic review and meta-analysis of double-blind trials compared with open-label trials of anticoagulants for stroke prophylaxis.39 Although we agree with the overall assumption that an open-label design will bias the results in favour of the intervention, the source of the adjustment factor, Lega 2013, is not necessary reliable. In Lega 2013, the two meta-analyzed groups (open-label and double-blind trials) that were compared with one another had significant differences in the baseline characteristics beside the blinding design, thus precluding the possible conclusion that an observed difference is attributed solely to the open-label rather than the double-blind design.39
The structure of the network did not allow the producer of the manufacturer’s IDC to provide statistical testing for possible heterogeneity and/or inconsistency. Thus, two essential assumptions for the validity of the IDC remain untested. The producer of the manufacturer’s IDC attempted to conduct several subgroup analyses to try and gain insight into possible heterogeneity in the studies. These results indicated that heterogeneity existed in the included trials.
To summarize, although the conduct of the study was sound, the following major limitations add uncertainty to the results:
Bajaj et al. aimed to compare the effectiveness and safety of FDA-approved stroke prophylaxis treatment strategies for patients diagnosed with NVAF. The authors conducted a systematic review and NMA. Their analysis approach used a frequentist multivariate meta-regression model. The authors do not specify how they handled between-study heterogeneity (fixed- or random-effects model). Although the authors mention that their regression model used random-effects multivariate regression, this statement is not informative about the method of dealing with between-study heterogeneity. The authors did not report on the results of testing for statistical heterogeneity or inconsistency, even though they report in their methods that such tests were to be conducted. The authors reported, on an odds ratio (OR) scale, ischemic stroke outcome, major bleed outcome, and a “primary safety” outcome defined as major bleeding and clinically relevant non-major bleeding.
The Bajaj et al. IDC included six RCTs in the analysis; four were the same as those included in the manufacturer’s IDC, and the two additional studies compared VKA to Watchman, a left atrial appendage closure device. Considering the outcome of ischemic stroke, edoxaban showed no statistically significant differences when compared with VKA, apixaban, dabigatran, rivaroxaban, or the Watchman device. In terms of the major bleed outcome, edoxaban showed a favourable statistically significant result when compared with VKA (OR edoxaban versus VKA = 0.78; 95% CI, 0.69 to 0.90) and when compared with rivaroxaban (OR rivaroxaban versus edoxaban = 1.31; 95% CI, 1.08 to 1.59). No statistically significant differences were observed when edoxaban was compared with apixaban, dabigatran, or the Watchman device. For the third outcome, a composite of major bleeding and clinically relevant non-major bleeding, edoxaban showed a favourable statistically significant result when compared with VKA (OR edoxaban versus VKA = 0.83; 95% CI, 0.77 to 0.90) and when compared with rivaroxaban (OR rivaroxaban versus edoxaban = 1.23; 95% CI, 1.10 to 1.38). However, it also showed a statistically significantly unfavourable result when compared with apixaban (OR edoxaban versus apixaban = 1.23; 95% CI, 1.08 to 1.41).
The Bajaj et al. IDC carries several limitations that are mainly related to a lack of reporting on the results of testing for inconsistency and statistical heterogeneity, especially considering that the authors recognize the existence of heterogeneity in study and patient characteristics. In addition, the authors do not specify many aspects of their methods approach and leave unclear whether they used a fixed-effects or random-effects model to account for between-study differences in treatment effects.
Compared with the manufacturer’s IDC, Bajaj et al. included one more intervention in their analysis, used a slightly different approach in synthesizing the data, and did not report on the composite outcome of stroke and SEE. Bajaj et al. showed similar results in the outcome of major bleeding, except in the comparison of edoxaban and dabigatran, for which there is no longer any statistical significance.
Cameron et al. aimed to compare the efficacy and safety of antithrombotic therapies (apixaban, dabigatran, edoxaban, rivaroxaban, and VKA) and ASA with or without clopidogrel in patients with NVAF. The authors conducted a systematic review and NMA. Their analysis approach used a noninformative prior Bayesian analysis with a binomial likelihood model fitted using Markov chain Monte Carlo simulations. The authors conducted the analysis under both random-effects and fixed-effects models but only reported the fixed-effects results because of data constraints. The consistency assumption was tested by comparing the deviance and the deviance information criterion of a fitted consistency and inconsistency model. No quantified statistical heterogeneity was provided; the authors conducted several subgroup analyses to test for possible heterogeneity. The authors reported the OR and the 95% credible intervals (CrIs) for the composite outcome of stroke and SEE and for the safety outcome of major bleeding.
Cameron et al. included 16 RCTs in their analysis: four were the same as those included in the manufacturer’s IDC; the additional 12 RCTs compared VKA with ASA or with ASA and clopidogrel. Considering the outcome of ischemic stroke and SEE, edoxaban 60 mg showed no statistically significant differences when compared with VKA. In terms of the major bleeding outcome, edoxaban 60 mg showed a favourable statistically significant result when compared with VKA (OR edoxaban 60 mg versus VKA = 0.79; 95% CrI, 0.69 to 0.90).
The Cameron et al. IDC carries several limitations that are reported by the authors: notable heterogeneity, insufficient data for subgroup analysis, and the use of the fixed-effects model for reporting values. These limitations are similar to the ones in the manufacturer’s IDC.
Further, compared with the manufacturer’s IDC, the main differences are in the wider inclusion of intervention and comparators in the Cameron et al. study, the use of OR in Cameron et al. instead of the risk ratio (RR) in the manufacturer’s IDC, and the approach to the data synthesis, in which a Bayesian NMA was used in Cameron et al. as opposed to a Poisson frequentist regression model in the manufacturer’s IDC. Despite these differences, both show very similar results when considering the stroke and SEE outcome and the major bleeding outcome of edoxaban versus VKA.
Lip et al. aimed to compare the effectiveness and safety of apixaban with dabigatran, rivaroxaban, and edoxaban for stroke prevention in NVAF patients with a CHADS2 score ≥ 2 and in NVAF patients with a history of stroke or TIA. The authors conducted an update on a systematic review and NMA. Their analysis approach used a noninformative prior Bayesian analysis under a fixed-effects model fitted using Markov chain Monte Carlo simulations. Statistical testing for the consistency and heterogeneity was not presented. The authors reported the hazard ratio (HR) and the 95% CrI for the composite outcome of stroke and SEE and for the safety outcome of major bleeding.
The Lip et al. IDC included four RCTs in the analysis, the same as were included in the manufacturer’s IDC. Considering the outcome of ischemic stroke and SEE in the subpopulation of patients with a CHADS2 score ≥ 2 and in the subpopulation of patients with a previous stroke or TIA, edoxaban 60 mg showed no statistically significant differences when compared with VKA, apixaban, rivaroxaban, or dabigatran. In terms of the major bleeding outcome, edoxaban 60 mg showed a favourable statistically significant result when compared with VKA in the CHADS2 subpopulation (HR edoxaban 60 mg versus VKA = 0.80; 95% CrI, 0.70 to 0.91), but not in the subpopulation of patients with a history of stroke or TIA. Edoxaban 60 mg also showed a favourable statistically significant result for the major bleeding outcome when compared with rivaroxaban in the CHADS2 subpopulation (HR edoxaban 60 mg versus rivaroxaban = 0.76; 95% CrI, 0.63 to 0.91), but not in the subpopulation of patients with a history of stroke or TIA; it also showed a favourable statistically significant result when compared with dabigatran 150 mg in the CHADS2 subpopulation (HR edoxaban 60 mg versus dabigatran 150 mg = 0.80; 95% CrI, 0.65 to 0.97), but not in the subpopulation of patients with a history of stroke or TIA. None of the rest of the comparisons for the major bleeding outcome in either patient population were statistically significant.
The Lip et al. IDC carries several limitations related to the use of the fixed-effects model, a lack of reporting on detailed methods, and a lack of testing for inconsistency and statistical heterogeneity, particularly given that the authors recognize the existence of heterogeneity in study and patient characteristics.
Compared with the manufacturer’s IDC, the main differences lie in the choice of reporting on only subpopulations in the Lip et al. IDC; thus, the results of Lip et al. cannot be used as a contrast to the main results of the manufacturer’s IDC. In addition, Lip et al. used HRs to report on the outcomes, as opposed to RRs used in the manufacturer’s IDC. Also, for the approach to the data synthesis, a Bayesian NMA was used in Lip et al. as opposed to a Poisson frequentist regression model in the manufacturer’s IDC.
Morimoto et al. aimed to compare the efficacy and safety of apixaban, betrixaban, dabigatran, edoxaban, and rivaroxaban as DOAC with ximelagatran as an oral anticoagulant and warfarin, idraparinux, and Aspirin. The authors conducted a systematic review and NMA. Their analysis approach is unclear, as it has not been reported in the published article or appendices beyond a reference to a Bayesian NMA methods paper. The authors explain how they adjusted for the open-label study design but do not provide any details related to the conduct and validity of the NMA approach, nor do they mention if the results were based on a fixed- or random-effects model. The authors reported the OR and the 95% CrI for the composite outcome of stroke and SEE and for the safety outcome of major bleeding.
Morimoto et al. included nine RCTs in their analysis: four were the same as included in the manufacturer’s IDC; four compared VKA with idraparinux, a Japan-specific dose of rivaroxaban, ximelagatran; and one compared Aspirin with apixaban. For the outcome of ischemic stroke and SEE, edoxaban 60 mg showed no statistically significant differences when compared with VKA or other DOACs. In terms of the major bleeding outcome, edoxaban 60 mg showed a favourable statistically significant result when compared with VKA (OR = 0.78; 95% CrI, not reported numerically), dabigatran 150 mg (OR = 0.72; 95% CrI, not reported numerically), and rivaroxaban (OR = 0.76; 95% CrI, not reported numerically).
The Morimoto et al. IDC has severely underreported the methods used in the NMA. Because of this underreporting, we cannot pass any informed judgment about the certainty of the results. As such, the most conservative approach is to assume a large degree of uncertainty and exert extreme caution when interpreting the results.
Compared with the manufacturer’s IDC, the main differences lie in the choice of wider inclusion criteria for intervention or comparator, the use of the OR instead of the RR used in the manufacturer’s IDC, and the approach to the data synthesis, in which a Bayesian NMA was used as opposed to a Poisson frequentist regression model in the manufacturer’s IDC. In the results, the point estimates of Morimoto et al. (CrI was not reported numerically) were similar in the edoxaban versus VKA comparison and in all comparisons for the outcome of major bleeding, but differences in the point estimate were noticeable when edoxaban was compared with other DOACs for the outcome of stroke and SEE.
Skjoth et al. aimed to compare the efficacy and safety of edoxaban with apixaban, dabigatran, and rivaroxaban in patients with NVAF. The authors included four phase III RCTs that compared the aforementioned drugs with warfarin and performed an IDC meta-analysis. Their analysis approach used the Bucher method. The authors reported the HR and the 95% CrI for the composite outcome of stroke and SEE for the safety outcome of major bleeding and for several other efficacy and safety outcomes.
Skjoth et al. included the same four trials as the manufacturer’s IDC. For the outcome of ischemic stroke and SEE, edoxaban 60 mg showed no statistically significant differences when compared with apixaban, dabigatran 110 mg, and rivaroxaban but showed an unfavourable statistically significance difference when compared with dabigatran 150 mg (HR dabigatran 150 mg versus edoxaban 60 mg = 0.75; 95% CI, 0.56 to 0.99). In terms of the major bleeding outcome, edoxaban 60 mg showed a favourable statistically significant result when compared with rivaroxaban (HR rivaroxaban versus edoxaban 60 mg HR = 1.30; 95% CI, 1.08 to 1.57). Most of the other comparisons with edoxaban 60 mg showed no statistically significant differences, with the following efficacy exceptions, which showed more favourable outcomes with dabigatran compared with edoxaban: HR of stroke in dabigatran 150 mg versus edoxaban 60 mg (0.73; 95% CI, 0.55 to 0.96), HR of hemorrhagic stroke in dabigatran 150 mg versus edoxaban 60 mg (0.48; 95% CI, 0.23 to 0.99).
The Skjoth et al. IDC carries several limitations, some of which are reported by the authors: notable heterogeneity, the use of the Bucher method restricting the comparisons to only two interventions with a common comparator, and the statement by the authors that the results are to be considered exploratory for hypothesis generation.
Compared with the manufacturer’s IDC, the main difference lies in the choice of the approach to data synthesis: Skjoth et al. used the Bucher method as opposed to the Poisson frequentist regression model used in the manufacturer’s IDC. Subsequently, the Bucher method restricts the number of trials that can inform on each IDC; this restriction does not apply in the Poisson frequentist regression model. Despite this difference, the results from Skjoth et al. were similar to the manufacturer’s IDC.
Tawfik et al. aimed to compare the efficacy of all antithrombotic therapies for patients with AF. The authors conducted a systematic review and NMA. Their analysis approach used a noninformative prior Bayesian analysis with Poisson likelihood model fitted using the Markov chain Monte Carlo process. The authors conducted the analysis under both random-effects and fixed-effects models but only reported the fixed-effects results because of large variance in the results of the random-effects model. The consistency assumption was tested by inspecting heterogeneity plots. No quantified statistical heterogeneity was available; the authors conducted several sensitivity analyses to test for possible heterogeneity. The authors reported the rate ratio and the 95% CrI for the outcomes of stroke or major bleeding and for several other safety and efficacy outcomes.
Tawfik et al. included 16 RCTs in their analysis: four were the same as were included in the manufacturer’s IDC; 12 compared VKA with ASA or with ASA and clopidogrel. For the outcome of stroke, edoxaban 60 mg showed no statistically significant differences when compared with VKA, apixaban, dabigatran 110 mg, and rivaroxaban. However, edoxaban 60 mg showed an unfavourable statistically significant difference in the stroke outcome when compared with dabigatran 150 mg (edoxaban 60 mg versus dabigatran 150 mg RR = 1.37; 95% CrI, 1.04 to 1.82). In terms of major bleeding, edoxaban 60 mg showed a favourable statistically significant result when compared with VKA (RR edoxaban 60 mg versus VKA = 0.80; 95% CrI, 0.71 to 0.91) and a favourable statistically significant outcome when compared with rivaroxaban (RR edoxaban 60 mg versus VKA = 0.78; 95% CrI, 0.64 to 0.94). All other outcomes showed no statistically significant differences between edoxaban 60 mg and VKA or other DOACs, except for the intracranial bleed outcome when edoxaban 60 mg was compared with VKA (RR = 0.46; 95% CrI, 0.34 to 0.62).
The Tawfik et al. IDC carries several limitations, some of which have been reported by the authors: notable heterogeneity, especially with the inclusion of patients with valvular atrial fibrillation and NVAF, and the use of fixed-effects model for reporting values.
Compared with the manufacturer’s IDC, the main differences are the choice of wider inclusion criteria for intervention or comparator, Tawfik et al. not reporting on the composite outcome of stroke and SEE, and the approach to data synthesis (a Bayesian NMA was used as opposed to the use of a Poisson frequentist regression model in the manufacturer’s IDC). Tawfik et al. showed similar results for the outcome of major bleeding, except in the comparison between edoxaban and dabigatran, for which the Tawfik et al. result did not show the statistical significance exhibited in the manufacturer’s IDC.
Tereshchenko et al. aimed to compare the efficacy and safety of DOACs (apixaban, dabigatran, edoxaban, and rivaroxaban), VKA, Aspirin, and the Watchman device in patients with NVAF. The authors did not explain if this study followed a systematic review approach to capturing all relevant trials. The authors employed a frequentist NMA with an overall random-effects model assumption, which turned into a fixed-effects model in comparisons with fewer than two informing trials. The consistency assumption was tested by comparing direct and indirect results in closed loops. Statistical heterogeneity was assessed in closed loops using an empirical Bayes approach. The authors reported the OR and the 95% CIs for the composite outcome of stroke and SEE and for the safety outcome of major bleeding.
Tereshchenko et al. included 21 RCTs in their analysis: four were the same as those included in the manufacturer’s IDC; the rest of the RCTs compared VKA with ASA, with ASA and clopidogrel, with control, or with the Watchman device or compared ASA with DOACs or control. For the outcome of ischemic stroke and SEE, edoxaban 60 mg showed no statistically significant differences when compared with VKA or other DOACs. In terms of the major bleeding outcome, edoxaban showed a favourable statistically significant result when compared with VKA (OR edoxaban versus VKA = 0.64; 95% CI, 0.46 to 0.90). No statistically significant differences were observed in the major bleeding outcome when edoxaban was compared with other DOACs.
The Tereshchenko et al. IDC carries several limitations, some of which were reported by the authors: notable heterogeneity, pooling of different doses of edoxaban, inconsistency and statistical heterogeneity testing not being feasible for any comparison that included a DOAC drug, and the use of a fixed-effects model for reporting values as most comparisons were informed by one or two trials. Compared with the manufacturer’s IDC, the main differences lie in the wider inclusion of intervention or comparator in Tereshchenko et al., the slightly different approach to data synthesis, and the reporting of OR instead of RR. For the outcome of stroke and SEE, the Tereshchenko et al. results are similar to the manufacturer’s IDC in the edoxaban comparison with dabigatran and apixaban but are different in the comparison with VKA (in which edoxaban is no longer statistically significant) and with rivaroxaban (in which the point estimate has switched sides around the 1 but remains statistically not significant). The Tereshchenko et al. results for the major bleeding outcome no longer show a statistically significant finding in the comparison with other DOACs, only in the comparison with VKA.
Our search strategy identified seven IDCs in addition to the one submitted by the manufacturer. All IDCs had similar PICO research questions, and they all included the same four trials that the producer of the manufacturer’s IDC included in its analysis. The approaches to conducting the IDC differed among the identified IDCs: some used a frequentist NMA approach, some used a Bayesian NMA approach, and one used the Bucher method. The reported outcomes were similar in definition. However, the treatment effect measure did differ among the IDCs, with some reporting HR, some OR, and some relative risk. Table 24 presents the stroke and SEE outcome and the major bleeding outcome from different IDCs that reported these two outcomes on the overall population.
Outcome of Stroke and SEE Outcome and Major Bleeding Outcome in All IDCs Reviewed.
Overall, the results from all the reviewed IDCs were similar to the manufacturer’s IDC results. These results would mostly tend to indicate that edoxaban is superior to VKA in terms of safety and is similar in terms of efficacy to VKA and all other DOACs. However, since all the reported IDCs share the same evidence base, they also share the same limitations arising from the architecture of the evidence network. Specifically, all comparisons between different DOACs are informed indirectly through a single trial that compares a DOAC with VKA. This marginally informative connection precluded an analysis using a random-effects model; thus, a fixed-effects model was used in all the reviewed IDCs. The fixed-effects model makes unrealistic assumptions about the true treatment effect and assumes that all trials share the same common effect and that any differences between trials are due to sampling error. In other words, the fixed-effects model assumes that all the differences in study and patient characteristics between studies have no effect on the true treatment effect. Such assumptions are not justifiable in the presence of the observed clinical heterogeneity in the evidence network (i.e., different inclusion and exclusion criteria and differences in study design). In addition, the lack of any head-to-head DOAC comparative trial also means that we cannot assess the consistency assumption in the IDC.
The use of fixed-effects model analysis and the lack of assessment of inconsistency drastically reduce the external validity of the results. As such, the safest approach would be to consider all the results of the IDCs as exploratory in nature and requiring further hypothesis testing.
In addition to the manufacturer’s submitted IDC, seven IDCs were identified in the literature. The eight IDCs were consistent in showing that edoxaban is statistically significantly superior to VKA in terms of major bleeding. The efficacy results from the IDCs tend to support the hypothesis that edoxaban is similar in efficacy to VKA and other DOACs. All the reviewed IDCs were limited by the small number of trials to support a robust analysis, the existence of clinical heterogeneity, and the use of fixed-effects models. These limitations considerably reduce the external validity of the results and can arguably limit their value to hypothesis generation. However, considering the current available evidence, no better-quality IDC could have been produced.
Except where otherwise noted, this work is distributed under the terms of a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International licence (CC BY-NC-ND), a copy of which is available at http://creativecommons.org/licenses/by-nc-nd/4.0/
Your browsing activity is empty.
Activity recording is turned off.
See more...
