NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Wells G, Coyle D, Cameron C, et al. Safety, Effectiveness, and Cost-Effectiveness of New Oral Anticoagulants Compared with Warfarin in Preventing Stroke and Other Cardiovascular Events in Patients with Atrial Fibrillation [Internet]. Ottawa (ON): Canadian Agency for Drugs and Technologies in Health; 2012 Apr 9.

Cover of Safety, Effectiveness, and Cost-Effectiveness of New Oral Anticoagulants Compared with Warfarin in Preventing Stroke and Other Cardiovascular Events in Patients with Atrial Fibrillation

Safety, Effectiveness, and Cost-Effectiveness of New Oral Anticoagulants Compared with Warfarin in Preventing Stroke and Other Cardiovascular Events in Patients with Atrial Fibrillation [Internet].

Show details

3CLINICAL REVIEW

3.1. Primary Research Questions

In patients with non-valvular AF:

  • What is the clinical effectiveness and safety of new oral anticoagulants compared with warfarin?
  • What is the cost-effectiveness of new oral anticoagulants compared to warfarin?
  • How do the new oral anticoagulants compare to optimal warfarin therapy when considering the time spent in the time in therapeutic range (TTR)?
  • How do the new oral anticoagulants compare to warfarin therapy in specific groups of patients with older age, other medical conditions, or who are taking other drug therapies?
  • What are the costs associated with warfarin when patients are stratified according to TTR? How do these compare with estimates for the new oral anticoagulants?
  • What is the cost-effectiveness of new oral anticoagulants compared to warfarin when stratified by age and CHADS2 score (CHADS2: C= congestive heart failure, H = hypertension, A = older than age 75 years, D = diabetes mellitus, S2 = prior stroke or history of transient ischemic attack).?

3.2. Methods

The strategy for building and analyzing the evidence base for the NOAC consists of three fundamental steps based on a predefined systematic review protocol. First, a broad systematic review of the available randomized and non-randomized evidence in the published literature for the outcomes specified in the protocol was undertaken, following the methods and procedures outlined in the Cochrane Handbook for Systematic Reviews of Interventions.7 Second, a network meta-analysis was conducted involving three new oral anticoagulants in a network for each of the outcomes specified a priori. The methods and procedures followed are those developed by the Canadian Collaborative for Methods, Applications, and Capacity Development in Network Meta-Analysis for Drug Safety and Effectiveness, funded by the Drug Safety and Effectiveness Network (DSEN) of the Canadian Institutes of Health Research. Third, an economic evaluation was conducted based on the Canadian Agency for Drugs and Technologies in Health’s Guidelines for the Economic Evaluation of Health Technologies: Canada and using the results of the network meta-analysis.8 A summary of the protocol is provided in Appendix 7.2.

The systematic review followed a protocol written a priori and was conducted in line with the Cochrane Handbook for Systematic Reviews of Interventions.7 A systematic review was undertaken to build the evidence base for the three NOAC identified in the protocol. The objective was to present an unbiased summary of all relevant studies of adequate quality in order to evaluate the clinical safety and efficacy/effectiveness of dabigatran, rivaroxaban, and apixaban compared with adjusted-dose warfarin (and nicoumalone) in preventing morbidity and mortality in patients with non-valvular atrial fibrillation.

The patient/population, intervention, comparator and outcome (PICO) statement is:

  • The patient/population of interest consists of individuals with non-valvular atrial fibrillation requiring anticoagulation.
  • The interventions include dabigatran, rivaroxaban, and apixaban.
  • The comparators include warfarin and other oral coumadin derivatives (i.e., nicoumalone).
  • The outcomes include:
    • For the clinical assessment:
      • All-cause stroke or systemic embolism
      • Major bleeding (International Society of Thrombosis and Haemostasis [ISTH] definition)
      • All-cause mortality
      • Intracranial bleeding [including intracerebral hemorrhage(ICH)]
      • Cardiovascular mortality
      • Ischemic/uncertain stroke or systemic embolism
      • Life-threatening bleeds.
    • For the economic modelling, the following additional outcomes will be considered:
      • Primary:

        Stroke

        ICH

        Extracranial hemorrhage

        Minor bleeds.

      • Secondary:

        Myocardial infarction (MI)

        Pulmonary embolism (PE)

        TIA

        Non-cardiovascular mortality.

The evidence network will consist of published randomized controlled trials (RCTs) and non-randomized studies (NRS) if they are at least 12 weeks in duration.

3.2.1. Literature search methodology

Studies were included if the PICO criteria were met and the types of studies included published active and non-active controlled RCTs and NRS of at least 12 weeks in duration. A computerized search strategy based on this was created with the help of an information specialist in order to address clinical safety and efficacy/effectiveness simultaneously. The following electronic databases were searched through December, 2011, using an OVID interface: Database of Abstracts of Reviews of Effects (DARE), Cochrane Database of Systematic Reviews, The Cochrane Library, 2011; Embase, 1980 to 2011; and MEDLINE, 1947 to present and In-Process & Other Non-Indexed Citations. Searches in MEDLINE and Embase used validated filters for systematic reviews, RCTs, cohorts, and case-controlled studies. No language filters were applied. Where possible, retrieval was limited to the human population. The following keywords were used to search the Cochrane Database of Systematic Reviews, DARE, and grey literature: anticoagulants, atrial fibrillation, rivaroxaban, apixaban, edoxaban, and dabigatran. The literature search strategy is provided in Appendix 7.3.

Grey literature (literature that is not commercially published) was searched December 8, 2011 and again on January 8, 2012. Relevant grey literature was identified by searching the websites of health technology assessment and related agencies, guideline producers, and professional associations that maintain safety information on pharmaceutical products. In addition, bibliographies of included articles and relevant systematic and non-systematic reviews were searched for possible references not otherwise found. In addition, reviews for the advisory committee of the US Food and Drug Administration (FDA), Division of Cardiovascular and Renal Drug Products, on rivaroxaban and dabigatran were used as information sources. There are no FDA review documents for apixaban, as it is currently under review with an estimated date for decision listed as June 28, 2012.62 At the time of this review, Health Canada had not yet approved apixaban for use in Canada. NICE manufacturer submissions for dabigatran and rivaroxaban were also consulted, as both drugs are currently under review for AF.63,64

3.2.2. Selection of studies

Studies were included if the PICO criteria and type of study outlined in the protocol (Section 2.1) were appropriate. Selection eligibility criteria were applied to each title and abstract identified in the literature search by two independent review authors in a standardized manner. Any uncertainties were resolved by discussion and consensus with a third review author. Any RCT and NRS passing the selection criteria were obtained in full-text format. Only NRS with comparative control were eligible for inclusion. The eligibility criteria were then applied and a final decision was made for inclusion. The reviewers were not blinded to study authors or centre of publication prior to study selection, as this complicates the review process and there is only weak evidence to suggest this would improve results.65

3.2.3. Data extraction and management

All information was extracted using a standardized data abstraction form (Appendix 7.4), which was based on the Cochrane Consumers and Communication Review Groups’ data extraction template. Abstraction included:

  • characteristics of trial participants including, inclusion and exclusion criteria
  • type of intervention including dose, duration, and co-medication
  • results of the clinical safety and efficacy/effectiveness outcomes of the intervention.7

All extracted data were checked for accuracy by two independent review authors.

The original, primary publication for each unique study included was used for data extraction, except where multiple publications for a single RCT or NRS were found. Multiple publications for a unique RCT or NRS (e.g., supplemental online appendices, companion publications of specific outcomes, or populations from the original study) were handled by extracting the most recently adjudicated data for each outcome specified in the protocol.

3.2.4. Quality assessment

Two sets of quality assessment instruments were considered. For RCTs, the checklist proposed by the Scottish Intercollegiate Guidelines Network (SIGN 50) for RCTs and the Cochrane Collaboration’s tool for assessing risk of bias (ROB) were used.65,66 Similarly, for NRS, the SIGN 50 checklist for cohort studies66 and an adaption of the Cochrane Collaboration’s tool for assessing ROB applicable for cohort studies were considered.7,65

The SIGN 50 assessment form for RCT (Appendix 7.5) consists of 14 checklist questions assessing internal validity, a question identifying the source of funding for the trial, and an overall assessment of the study by rating the methodological quality based on answers to the these questions. Answers from the checklist are not weighted. The risk of bias is classified as either:

Low: All or most of the criteria from the assessment of internal validity are satisfied. Study conclusions would not likely be altered if methods were changed.

Moderate: Some of the criteria from the assessment of internal validity are satisfied. Study conclusions would not likely be altered if methods were changed.

High: Few or none of the criteria from the assessment of internal validity are satisfied.

The Cochrane Collaboration’s ROB tool (Appendix 7.6) is a two-part tool addressing six specific domains (namely sequence generation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting and “other issues” which were identified as a source of funding). Each domain includes one or more specific entries in an ROB table, and a form was created in line with the Cochrane Collaboration’s ROB template: the first part involves describing what was reported to have happened in the study; and the second part involves assigning a judgment relating to the risk of bias for that entry by answering a pre-specified question about the adequacy of the study in relation to the entry, including a judgment of “LOW” risk of bias, “HIGH”risk of bias, and “UNCLEAR” or unknown risk of bias. Two entries were considered for the domain “blinding” for assessments of outcomes that were subjective or objective.

For each unique RCT or NRS, information for the quality appraisal was first obtained from the original publication, but additional relevant study literature was also used to conduct the assessment, including where available: design and rationale documents, companion study publications, protocols, published comments on the study, and contact with investigators.

3.2.5. Assessment of heterogeneity

We qualitatively assessed clinical and methodological heterogeneity across studies. To adjust for potential sources of heterogeneity, subgroup analyses were performed at the study-level where data was available. Additional details on assessment of heterogeneity are presented in Section 3.3.16.4.

3.2.6. Data synthesis

All outcomes of interest reported in each study were dichotomous and analyzed. For each outcome, the odds ratio (OR) and corresponding 95% confidence interval were calculated for the overall treatment effect. The absolute risk difference per 1,000 patients treated each year for each outcome was also calculated for each outcome using study-level data. Confidence intervals for absolute risk reductions were calculated by multiplying the confidence intervals of the relative risk estimates (see Appendix 7.7) by the annual baseline event rate. Although hazard ratios were reported for each study (see Appendix 7.7), ORs were used because hazard ratios were often not available for certain patient subgroups. Moreover, use of ORs facilitated comparison of study-level results with those from Bayesian MTC meta-analyses where WinBUGS code for dichotomous data is readily available and widely used.67 Nevertheless, there were negligible differences between study-level ORs, and hazard ratios and relative risks (see Appendix 7.7).

3.2.7. Assessment of publication bias

Reporting bias was assessed by constructing funnel plots for each outcome. An asymmetrical plot would imply publication bias, as in the absence of bias the plot should resemble an inverted funnel.

3.2.8. Subgroup analysis

Subgroups of interest included: TTR, CHADS2 (or CHADS2-VASC) score, age (stratified as < 65 years, 65 years to 74 years, and ≥ 75 years), weight, impaired renal function (including mild, moderate, and severe), prior history of gastrointestinal (GI) bleed, concurrent use of antiplatelet agents, concurrent use of nonsteroidal antiinflammatory drus (NSAIDs).

Rational for subgroup inclusion:

  1. Weight – Although no dosing modifications are indicated for low or high body weight in the product monographs, low weight is known to increase drug exposure in apixaban.52 Older age is known to be associated with weight loss. Also, overweight patients are becoming increasingly common in clinical practice. There is limited evidence on novel treatments relating to weight-related safety and whether dosing should be adjusted based on higher or lower renal or hepatic clearance, or variations in other pharmacokinetic areas.
  2. Age – Prevalence of AF increases with age and the majority of AF patients are elderly; however, this group is often under-represented in clinical trials. Both risk of bleeding and risk of stroke increases with age. Prior studies with different antithrombotic drugs showed significant interactions with age.68 In addition, older patients have more comorbidities and are on more concomitant drugs, which could influence the efficacy and safety of old and new treatments.
  3. Renal Impairment – The renal clearance of each new oral anticoagulant varies. According to the individual drug monographs, the renal clearance of dabigatran is 80%, meaning that the drug is mostly eliminated through the kidneys. Rivaroxaban (33%) and apixaban (27%) are less dependent on renal elimination.49,52,53 The potential for bioaccumulation in patients with renal failure could necessitate dosage adjustments or prevent this cohort of individuals from using novel anticoagulant drugs.69 No dosage adjustments are required for warfarin use in renally impaired patients.
  4. CHADS2 – CHADS2 score is a validated risk score to predict the risk of stroke in patients with AF. In addition, CHADS2 score is also related to the risk of bleeding. Subgroup analyses according to CHADS2 allow further exploration into the efficacy and safety of novel treatments in low- and high-risk patients.
  5. CHA2DS2VASc – CHA2DS2VASc expands on the CHADS2 score by including additional risk categories for age 64 to 75, vascular disease, and gender.
  6. Prior Use of Vitamin K Agonist – VKA-naïve participants may be more at risk for complications in warfarin arms until the dose has been stabilized, and, therefore, warfarin arms could be subject to higher discontinuation rates. This subgroup could have implications in first- or second-line therapy choice.
  7. History of Gastrointestinal Bleed – All bleeding complications are of interest for the novel anticoagulants and warfarin, as they can increase the risk of GI bleeding. Although this population of patients is potentially excluded in clinical trials, understanding the role of anticoagulation therapy following GI bleeding is important in the clinical setting.
  8. Concurrent use of NSAID medication – This subgroup was prioritized through consultation with clinical experts across Canada. There are drug interactions to consider with the novel anticoagulants, and bleeding risk could increase with concurrent NSAID use.
  9. Concurrent use of antiplatelet medication – This subgroup was prioritized through consultation with clinical experts across Canada. Bleeding risk and other adverse events associated with combining therapy with antiplatelet medications has not been quantified extensively in the novel anticoagulants.
  10. TTR – TTR is one of the most important determinants of therapeutic effectiveness in VKA therapy, and good INR control is essential to minimizing the risk of hemorrhagic events associated with over-coagulation and stroke associated with under-coagulation. TTR can also be influenced by many factors external to anticoagulation.

3.2.9. Sensitivity Analysis

If relevant heterogeneity was present, sensitivity analysis were conducted based on aspects of the PICO statement and study methodology.

3.2.10. Network Meta-Analysis

3.2.10.1. Methods for Bayesian Mixed Treatment Comparison Meta-analysis

Bayesian mixed treatment comparison (MTC) meta-analyses were conducted for the following outcomes: all-cause stroke or systemic embolism, all-cause stroke, all-cause mortality, cardiovascular mortality, major bleeding, intracranial hemorrhage (including ICH), GI bleeding, and myocardial infarction (MI). WinBUGS software (MRC Biostatistics Unit, Cambridge, UK) was used to conduct Bayesian MTC meta-analysis using a binomial likelihood model which allows for the use of multi-arm trials.67,70 Both fixed and random-effects network meta-analyses were conducted; assessment of model fit and choice of model was based on assessment of the deviance information criterion (DIC) and comparison of residual deviance to number of unconstrained data points.67,71 Trials with zero cells in both arms were excluded from evidence networks because they do not contribute information.67

Point estimates and 95% credible intervals were modelled for OR using Markov chain Monte Carlo (MCMC) methods. We also assessed the probability that each drug was the most efficacious regimen, the second best, the third best, and so on, by calculating the OR for each drug compared with a warfarin control group, and counting the proportion of iterations of the Markov chain in which each drug had the highest OR, the second highest, and so on. Vague or flat priors, such as N(0, 1002) were assigned for basic parameters throughout.67 To ensure convergence was reached, trace plots and the Brooks-Gelman-Rubin statistic were assessed.72 Three chains were fit in WinBUGS for each analysis, with at least 20,000 iterations, and a burn-in of at least 20,000 iterations.

Both MTC and traditional meta-analysis require studies to be sufficiently similar in order to pool their results. Consequently, heterogeneity across trials in terms of patient characteristics, trial methodologies, and treatment protocols across trials was carefully assessed.70,71

3.2.10.2. Methods for the Frequentist General Linear Mixed Models

A general linear mixed model (GLMM) was also used to conduct the network meta-analysis for the prespecified outcomes following careful assessment of heterogeneity across trials regarding subject characteristics, trial methodologies, and treatment protocols.

As the outcomes followed a binomial distribution, a mixed log-binomial model was implemented with the logit link function to generate the OR estimates. Point estimates and 95% confidence intervals were provided to summarize findings.

A random effects GLMM model was conducted. The random effects trial accounted for the response variables of patients within a given trial being correlated. The random effects trial X treatment was considered to account for the correlation of responses between any two patients from the same treatment arm within a given study. However, the random effects trial X treatment had to be excluded from the model because of the small number of trials. In cases where the number of observations is lower than the number of model parameters to be estimated, then the model cannot sustain the inclusion of the trial*treatment random effect.

The GLIMMIX procedure in SAS 9.2 (SAS Institute Inc., Cary, NC, USA) was used for all analyses.

3.3. Systematic Review Results

In this section, the results of the literature review, critical appraisal of the studies identified, and comparability of the studies are provided.

3.3.1. Literature search results

The initial literature search returned 4,677 database abstracts and 30 grey literature documents. Of the 66 full-text articles assessed for inclusion after duplicates were removed, 51 were excluded for a variety of reasons detailed in the PRISMA flow diagram detailing the literature search screening and selection process (Figure 1).73

Figure 1. PRISMA Flow Diagram of Study Selection.

Figure 1

PRISMA Flow Diagram of Study Selection.

Five unique RCTs were identified, with a total of 51,302 study participants.913 Ten companion studies, conference abstracts, conference posters, and online supplemental data appendices were also located.7483 Two trials had full data results posted online at www.clinicaltrials.gov, as located during the grey literature search. The grey literature search also located supplemental data for dabigatran and rivaroxaban in two reports published by the FDA from its drug approval process. The FDA data were only used where gaps in the published literature data existed; this has been noted where relevant in this report. No FDA report was available for apixaban at the time of this therapeutic review. A complete list of included and excluded studies is available in Appendix 7.8.

3.3.2. Summary of included studies

Study characteristics of the five included trials are summarized in Table 1. The review included five unique RCTs (reported in 13 literature sources and two FDA reports) evaluating the effects of NOAC in patients with non-valvular AF.913,7483 No NRS were located.

Table 1. Trial Characteristics of Included Studies.

Table 1

Trial Characteristics of Included Studies.

3.3.3. Trial characteristics

Table 2 provides summaries of the characteristics of the included RCTs.913,7483 Three of the included studies are large, multinational, multicentre studies that account for 50,578 of the 51,302 randomized patients in this review.9,11,13 The final number of randomized patients included in this review is 51,066 due to the exclusion of the 50 and 300 mg twice-daily treatment arms from PETRO.10 One smaller RCT was conducted in four international locations (n = 502, 53 sites), and one small trial (n = 222, 23 sites) was conducted in a single country (Japan).10,12 All trials published results in 2007 or later, and three of the trials published results in 2011.

Table 2. Summary of Trial Characteristics.

Table 2

Summary of Trial Characteristics.

3.3.4. Treatments evaluated

Evidence was available for the following interventions: apixaban (two unique RCTs),11,12 dabigatran (two unique RCTs),9,10 rivaroxaban (one RCT).13 A total of five distinct treatment strategies were tested in the included studies (Table 3).

Table 3. Summary of Treatments Evaluated.

Table 3

Summary of Treatments Evaluated.

In all studies, dose-adjusted warfarin was the comparator. All warfarin arms in the five RCTs adjusted the warfarin dosage based on a target INR range of 2 to 3. One RCT used an alternative INR range of 2 to 2.6 in patients aged 70 or higher (ARISTOTLE-J). The PETRO trial evaluated dose escalation of dabigatran: 50 mg, 150 mg, and 300 mg twice daily. The 50 and 300 mg twice-daily treatment arms from this RCT have been excluded from this therapeutic review. Only the 110 mg and 150 mg twice-daily doses have been approved in Canada and Europe. It is also important to note that PETRO was aimed at evaluating primary end points related to safety and not efficacy.

The ARISTOTLE trial reduced the dose of apixaban (2.5 mg twice daily) for participants with two or more of the following characteristics:

  • ≥ 80 years
  • body weight ≤ 60 kg
  • a serum creatinine level ≥ 1.5mg/dL (N = 428 or 4.9%).

ROCKET-AF allowed for a reduced dose of rivaroxaban (15 mg/day) in patients with reduced renal function, described as creatinine clearance of 30 to 49 mL/min.

ROCKET-AF administered study medication once daily. All other trials administered study drugs twice daily.

3.3.5. Study design features

All of the studies included were RCTs with active comparators (warfarin). Placebo studies were excluded from this review. Four of five of the trials were parallel group design, and one trial (PETRO) employed a 3×3 factorial design. In the PETRO trial, randomization was stratified in the ratio 6:9:9:4 (for dabigatran 50 mg, 150 mg, and 300 mg, and warfarin, respectively). The 3×3 factorial design incorporated three different levels of concomitant aspirin (none, high, low dose) with the dabigatran; however, the aspirin dosing was not stratified in the randomization. PETRO was a dose-finding study designed to evaluate safety, and was not designed to look at efficacy.

Warfarin comparison arms were blinded in two RCTs (ROCKET-AF and ARISTOTLE) and remained open-label in the other three trials (PETRO, RE-LY and ARISTOTLE-J). Three trials (ROCKET-AF, ARISTOTLE, and PETRO) employed a placebo pills and/or sham INR adjustment to maintain the blinding of the intervention arms. The RE-LY trial blinded the dose of dabigatran, but otherwise remained open-label. All trials blinded outcome assessment.

3.3.6. Follow-up duration

The five included RCTs ranged from three to 48 months in duration. Both smaller studies (ARISTOTLE-J, PETRO) (n = 724, n = 458 included in this review) followed patients for 84 days (12 weeks), whereas the three large studies (ROCKET-AF, RE-LY, ARISTOTLE) (n = 50,578) had a mean follow-up time of 708.7 days (SD ± 27.5).

3.3.7. Funding

All the studies included were sponsored by the pharmaceutical industry.

3.3.8. Populations

Three large RCTs (ROCKET-AF, ARISTOTLE, and RE-LY) account for 50,578 of the 51,302 randomized participants. The individual trial inclusion and exclusion criteria can be found in Appendix 7.9. The five included studies randomized adults of both genders. Gender balance was relatively similar across the large trials; however, patients in all arms of the two smaller studies were predominantly male (> 81%). Mean age differed slightly among trials. The ROCKET-AF study had a higher median age (73 years) when compared with the other included studies, and differed in a number of other baseline characteristics. Patients enrolled in the ROCKET-AF trial had notably higher patient percentages of congestive heart failure, diabetes mellitus, hypertension, and prior stroke/TIA or MI (although prior MI was not reported for all trials). Specifically, the history of stroke or TIA was 20% or less in the ARISTOTLE and RE-LY studies, but was almost 55% in the ROCKET-AF study. ROCKET-AF was the only trial that allowed newly diagnosed patients with AF to enter the trial. (See Table 4.)

Table 4. Baseline Population Characteristics.

Table 4

Baseline Population Characteristics.

These baseline differences were reflected in the higher mean CHADS2 score for the ROCKET-AF study (mean 3.58 versus 2.1 for both RE-LY and ARISTOTLE) (Table 5). The RE-LY study had approximately 32% of participants, with a low-risk CHADS2 score of 0 or 1; and ARISTOTLE had 34%, with a CHADS2 score of 1. The ROCKET-AF trial excluded these patients. The ROCKET-AF trial had more than 85% of patients with a CHADS2 score of 3 or higher, which means they were at a higher risk for stroke at baseline. Guidelines recommend that patients with a CHADS2 score of 0 are not indicated for anticoagulation, as their risk of stroke is smallest.32

Table 5. Baseline CHADS2 Score.

Table 5

Baseline CHADS2 Score.

ARISTOTLE-J had a very low percentage of patients with CHD (< 1.5%), which could be attributed to the small number of participants (n = 222). It also included patients with a higher baseline level of prior stroke/TIA (21.6 to 35.1%) when compared to all other studies except for ROCKET-AF.

None of the trials reported a history of GI bleed in their baseline characteristics; however, this was an excluded population in many of the trials.

Previous use of VKA therapy also differed significantly across trials, and certain trials aimed to randomize specific proportions of VKA-naïve participants. ARISTOTLE aimed for 40% VKA-naïve, RE-LY required a balanced proportion of VKA-naïve and experienced participants, and ROCKET-AF preferentially sought VKA-naïve participants. ARISTOTLE also had stratified randomization based on participant’s VKA experience. This could impact trial results, as VKA-naïve participants may be more at risk for complications in warfarin arms until the dose has been stabilized and, therefore, warfarin arms could be subject to higher discontinuation rates. ROCKET-AF rationalized that VKA-naïve patients may reflect a more realistic experience with anticoagulants as compared with those already successfully taking oral anticoagulant therapy. The PETRO trial had 100% of participants on long-term VKA therapy. Only 12% to 16% of participants in the ARISTOTLE-J trial were VKA-naïve. ROCKET-AF reported 37.5% of participants were VKA-naïve, despite having preferential enrollment for these patients. Approximately 50% of the participants in RE-LY were first-time users of VKA therapy. Numbers for ARISTOTLE fell in between those of the ROCKET-AF and RE-LY trials, with 42% being VKA-naïve at randomization.

The type of AF was reported in all three large trials, but different classification criteria were used. PETRO and ARISTOTLE-J did not report this data (see Table 6). ARISTOTLE and ROCKET-AF did not report permanent and persistent separately. Inclusion of patients with new onset atrial fibrillation was only allowed in ROCKET-AF, but the number of included patients was small (1.4%). Importantly, almost one-third of patients in RE-LY had paroxysmal A, whereas ROCKET-AF and ARISTOTLE included only 17% and 15%, respectively. The AF criteria used for patient enrollment is reported in Table 7.

Table 6. Patient Characteristics by Type of Atrial Fibrillation.

Table 6

Patient Characteristics by Type of Atrial Fibrillation.

Table 7. Atrial Fibrillation Entry Criteria.

Table 7

Atrial Fibrillation Entry Criteria.

Mean time in TTR (target INR of 2 to 3) ranged from 55% in ROCKET-AF to 64% in RE-LY. In ROCKET-AF, TTR included time on and off the drug (on treatment safety population), whereas in ARISTOTLE and RE-LY, only time on the study drug was included in this statistic. TTR in PETRO and ARISTOTLE-J may be challenging to compare with the three larger trials given their short trial duration (12 weeks) and small sample sizes.

3.3.9. Concomitant medications

Trials varied regarding permitted concomitant medications. Table 8 summarizes concomitant medications allowed in the included studies. PETRO allowed high-dose acetylsalicylic acid or aspirin, defining high dose as ≥ 325 mg. Three trials (ARISTOTLE-J, RE-LY, and ROCKET-AF) allowed low-dose aspirin, up to 100 mg/day, whereas ARISTOTLE permitted up to 165 mg/day and PETRO 81 mg/day.

Table 8. Concomitant Medications Permitted.

Table 8

Concomitant Medications Permitted.

RE-LY permitted dual antiplatelet therapy, whereas ROCKET-AF permitted dual antiplatelet therapy only in patients undergoing appropriate vascular interventions, and at the discretion of trial investigators. ARISTOTLE-J did not permit any thienopyridine use. ARISTOTLE reported the percentage of patients taking clopidogrel at baseline, but was unclear as to whether patients could continue this medication or if it was discontinued when study medications were allotted.

Nonsteroidal antiinflammatory drugs (NSAID) were not permitted in ROCKET-AF, and “discouraged” in the RE-LY trial (no usage statistics were provided). ARISTOTLE reported NSAID use at baseline, but, again, it was unclear if participants were allowed to continue use during the trial.

3.3.10. Frequency of INR monitoring and outcome assessment

INR monitoring across included studies varied at study outset (Table 9). Participants were generally monitored for INR every 4 weeks in all five trials. ARISTOTLE, ARISTOTLE-J, and PETRO stipulated shorter interval INR assessment in the first month of the trial, and ARISTOTLE had a different initial monitoring frequency based on the participants’ VKA-naïve or experienced status. All three large trials permitted individual INR monitoring to be adjusted as clinically indicated to maintain an INR between 2 and 3. PETRO and ARISTOTLE-J did not report if more frequent monitoring was permitted to maintain INR, or whether any increased monitoring occurred. These trials were much shorter in duration and, therefore, less time was available to assess and adjust INR during the trial.

Table 9. Frequency of INR Monitoring.

Table 9

Frequency of INR Monitoring.

Frequency of outcome assessment during the trial and at the end of study varied (Table 10). ARISTOTLE was an event-driven trial, and the only follow-up was sham or real INR monitoring, with no other scheduled visits until 448 primary events occurred, or four years, whichever came first. PETRO and ARISTOTLE-J did not provide detailed outcome or assessment details; however, given the short duration of both trials, it is assumed that follow-up and outcome assessment occurred simultaneously, ending at 12 weeks. RE-LY and ROCKET-AF followed patients up for a maximum of three and four years, respectively. After year one, RE-LY monitored patients every three months, and then every four months during year 2 and 3. ROCKET-AF followed up with patients every month until the end of the study.

Table 10. Frequency of Outcome Assessment.

Table 10

Frequency of Outcome Assessment.

3.3.11. Patient disposition

3.3.11.1. Early withdrawals

Study-level detail regarding the proportion of patients who withdrew from each trial is presented in Table 11. There is a potential source of bias due to withdrawals and early discontinuation of study medications in some of the included RCTs. This potential issue with participant retention could be attributed to either the proportion of withdrawals across study arms, or for the trial as a whole, or the reasons cited for discontinuing the study medications (see Table 11).

Table 11. Patent Disposition.

Table 11

Patent Disposition.

Reporting of withdrawals and discontinuation was difficult to reconcile across trials. Both PETRO and ARISTOTLE-J reported study medication discontinuation, but zero loss to follow-up. Details on discontinuation and loss to follow-up were well reported for the three large trials; however, each differed in presentation and how final numbers and analysis populations were described.

The trials report low percentages of participants who were lost to follow-up, but have very high numbers of patients who discontinued from the study medications (20% to 35%), with varying rates of follow-up completion. ROCKET-AF and RE-LY report both discontinuations from the study and of study medications (FDA reports). In those who discontinued study medications, many continued follow-up. Tables reporting this data state that subjects may be double-counted; however, it is not possible to differentiate exactly where double-counting occurred.

RE-LY and ARISTOTLE report both loss-to-follow-up, and subjects where they have participants with final vital status unknown at the end of study. Further detail is not provided.

Analysis populations

Analysis populations were assessed for all studies included in this review. PETRO and ARISTOTLE-J reported no formal statistical hypothesis testing. PETRO provided no definitions for study populations used to present outcome data. ARISTOTLE-J reported tghree different populations used to assess bleeding, efficacy, and safety, but no P-values or confidence intervals are presented due to the lack of a formal statistical test. ARISTOTLE and RE-LY analyzed primary outcomes using ITT populations with ARISTOTLE, also reporting a modified ITT analysis for safety. ROCKET-AF presented a minimum of four study populations, reporting primary non-inferiority analysis on per-protocol populations, primary superiority analysis on the “safety/on treatment” population, and post-hoc analysis in ITT “up to site notification” or ITT “up to data cut-off.” Definitions, where available, are listed as follows:

ROCKET-AF:

  1. Per-Protocol = Received one dose of study drug, no major protocol violations, and were followed for events while receiving a study drug, or within two days after discontinuation.
  2. Safety On Treatment = Received at least one dose, were followed for events, regardless of adherence to protocol, while receiving assigned study drug or within two days of discontinuation.

ARISTOTLE:

  1. ITT.
  2. Modified ITT sensitivity analysis to analyze bleeding that occurred in patients who received at least one study dose.

RE-LY: All analyses ITT.

ARISTOTLE-J: No formal statistical hypothesis tested. No P-values or confidence intervals presented.

  1. “Treatment period” analysis for bleeding events = Starting on the day of first dosing and continuing until two days after discontinuation.
  2. Efficacy: based on “intended treatment period” = Starting on the day of randomization and ending either two days after the last dose of study drug or at the 12-week visit, whichever came last.
  3. Safety = All patients who received at least one dose of study medications.

PETRO: Dose-finding study. No formal statistical hypothesis tested.

3.3.12. Outcomes

The relative safety and efficacy of the new oral anticoagulant drugs was assessed for the outcomes listed in section 6. An overview of the RCT evidence available for each outcome of interest is presented in Table 12.

Table 12. Summary of RCT Evidence by Outcome.

Table 12

Summary of RCT Evidence by Outcome.

3.3.12.1. All-cause stroke or systemic embolism

This composite outcome had similar definitions across trials in which the outcome definition was reported. PETRO did not report this outcome, and ARISTOTLE-J did not provide a definition (Table 13).

Table 13. Composite Definition for All-cause Stroke or SE.

Table 13

Composite Definition for All-cause Stroke or SE.

3.3.12.2. Major bleeding (ISTH definition)

The Subcommittee on Control of Anticoagulation, of the Scientific and Standardization Committee of the ISTH, endorses the following criteria for major bleeding in non-surgical patients:

  • fatal bleeding, and/or
  • symptomatic bleeding in a critical area or organ, such as intracranial, instraspinal, intraocular, retroperitoneal, intraarticular, or pericardial, or intramuscular with compartment syndrome, and/or
  • bleeding causing a fall in hemoglobin level of 20g L (−1) (1.24 mmol/L[−1]) or more, or leading to a transfusion of two or more units of whole blood or red cells.

Major bleeding definitions are comparable across trials, and all trials followed the ISTH definition, even where it was not specifically stipulated that this definition was used. ROCKET-AF added permanent disability to its definition of major bleeding, the only departure from the ISTH classification (Table 14).

Table 14. Major Bleeding Definitions.

Table 14

Major Bleeding Definitions.

3.3.12.3. All-cause mortality

All-cause mortality was reported in similar fashion in the included studies.

3.3.12.4. Cardiovascular mortality

Cardiovascular (CV) mortality was sometimes reported in studies within vascular mortality.

3.3.12.5. Ischemic/uncertain stroke or systemic embolism

This composite outcome was not reported in any trials and, therefore, no definitions were reported in any of the included studies.

3.3.12.6. Life-threatening bleeds

Regarding RE-LY, life-threatening bleeding was a subset of major bleeding that included fatal or symptomatic intracranial bleeding, bleeding associated with a hemoglobin decrease of 5.0 g/dL, or requiring transfusion of four units of blood or inotropic agents, or bleeding necessitating surgery.

3.3.12.7. Minor bleeding

There is currently no established definition or criteria for minor bleeding similar to the ISTH definition for major bleeding. Definitions for minor bleeding varied significantly in all five included studies, and made across-trial comparisons difficult. ARISTOTLE and ROCKET-AF reported clinically relevant non-major (CRNM) bleeding, whereas RE-LY and PETRO report only minor bleeding. PETRO and ARISTOTLE-J reported both CRNM bleeding and minor bleeding separately, or as subgroups within a broader definition. The primary safety outcome for ROCKET-AF was the composite outcome of CRNM bleeding and major bleeding, and ARISTOTLE also reported this composite outcome (Table 15).

Table 15. Minor Bleeding Definitions.

Table 15

Minor Bleeding Definitions.

PETRO is specific in its definition of minor bleeding; as this was a primary outcome assessed in the trial; however, the definition also includes a clause where any other relevant bleeding deemed important by investigators could be added. RE-LY reported a truncated definition of minor bleeding that included all other bleeding that was not major or ICH-related. ARISTOTLE-J likewise gave a somewhat open minor bleeding definition that includes all acute, clinically overt bleeding events that do not fit into the CRNM or major bleeding categories. The definitions of CRNM bleeding also vary in those trials that reported this as an outcome. Definitions for CRNM bleeding in ARISTOTLE and ARISTOTLE-J are equivalent; however, the CRNM bleeding definition used by ROCKET-AF differs. For example, the definition used for the ROCKET-AF trial bleeding requiring any medical intervention or unscheduled contact, not only hospitalization as specified in the comparable definitions. It also includes the impairment of any daily activities, with no further instruction on level of impairment or functional limitations. ISTH notes that characteristics for hospitalization on medical contact have been intentionally left out of the major bleeding definition, as they can be influenced by a number of external factors, not limited to community medical access, availability of beds, and presence of comorbidities.

3.3.12.8. Gastrointestinal bleeding

GI bleeding was not explicitly defined across all studies included in this therapeutic review. ROCKET-AF included GI bleeding within a category defined as major mucosal bleeding, which included gingival, epistaxis, and upper GI tract bleeding, and further categorized data into life-threatening mucosal bleeding. RE-LY reported data for both major GI bleed and any GI bleed, whereas ARISTOTLE-J reported GI bleeds within the ISTH major bleeding outcome. Life-threatening versus non–life-threatening bleeding was not defined or reported in RE-LY or ARISTOTLE-J. No GI bleeding outcome data has been reported for ARISTOTLE.

3.3.13. Subgroups

The following sub-groups of interest were specified a priori:

  • Weight
  • Age
  • Renal impairment
  • CHADS2
  • CHADS2VASC
  • Prior use of vitamin K Agonist
  • History of GI
  • Concurrent use of NSAID medication
  • Concurrent use of antiplatelet medication
  • TTR.

Subgroup analysis was limited by availability of data. Appendix 7.10 contains a comparison table of data available for each subgroup across all included studies. Event numbers were too small to include study-level data from ARISTOTLE-J or PETRO in the subgroup analyses. Availability of data and categorization differed across trials, and it was not possible to conduct analysis for the majority of the pre-specified subgroups.

Where published data were available, subgroup categories were collapsed into comparable sets for ARISTOTLE, RE-LY, and ROCKET-AF. Subgroup analyses were conducted for age (< 75 or ≥ 75), TTR (< 66% or ≥ 66%), and CHADS2 score (< 2 or ≥ 2) for the outcomes of all-cause stroke/SE and major bleeding. The reported data and study subgroups in ARISTOTLE, RE-LY, and ROCKET-AF were not informative or consistent enough to identify across these three studies a common subgroup based on weight, renal impairment, CHADS2VASC, prior use of VKA, history of GI bleed, concurrent use of NSAID medication. and concurrent use of antiplatelet medication (Appendix 7.10).

3.3.13.1. Subgroup credibility assessment

The outcomes of all-cause stroke/SE and major bleeding were assessed with respect to the criteria for the credibility of subgroup analyses (Oxman 199284, updated by Sun 201085) (Table 16). Data were analyzed for the subgroups of age, TTR, and CHADS2 for the primary outcomes of all-cause stroke or SE and major bleeding. No other subgroup data was analyzed in this review.

Table 16. Criteria to Assess the Credibility of Subgroup Analyses.

Table 16

Criteria to Assess the Credibility of Subgroup Analyses.

The intention of applying the criteria suggested by Sun et al.85 is to aid the decision-making process by providing clarity and perspective on whether subgroup analyses can be considered spurious or reliable. Evaluating the difference of effect between study subgroups usually involves some measure of uncertainty or confidence. Often, theoretical subgroup effects cannot be explicitly accepted or rejected but, rather, placed on a spectrum of likelihood based on whether we believe the subgroup effect to be true (highly plausible) or false (unlikely). In looking at the criteria in Table 16, the greater the number of criteria that are satisfied for each subgroup and outcome, the more plausible is the hypothesized subgroup effect.84,85

3.3.14. Critical appraisal

The SIGN 50 quality assessment instrument and the Cochrane Collaboration’s ROB tool were used to critically appraise the included studies. The details of these assessments are provided in Appendices 7.5.1 and 7.6.1. Summaries of these assessments are given in Table 17.

Table 17. Critical Appraisal Summary.

Table 17

Critical Appraisal Summary.

The overall SIGN 50 assessment of the methodological quality of the study is based on criteria related to the design and conduct of the study using the following coding system:

++

All or most of the criteria have been fulfilled. Where they have not been fulfilled, the conclusions of the study or review are thought very unlikely to alter.

+

Some of the criteria have been fulfilled. Those criteria that have not been fulfilled or not adequately described are thought unlikely to alter the conclusions.

Few or no criteria have been fulfilled. The conclusions of the study are thought likely or very likely to alter.

All studies had reasonable quality, with the quality of the ARISTOTLE, ROCKET-AK, and PETRO identified as “very good,” and RE-LY and ARISTOTLE-J considered “good.”

For the ROB, low, unclear, and high risk of bias are identified for different sources of bias in the design of the study. For eight sources of bias, all studies did well with ≥ 5 of the 8 criteria scored low or unclear bias.

3.3.15. Sources of Bias

3.3.15.1. Handling of missing data

No bias could be identified for handling missing data. A small number of patients (0% to 0.66%) were lost to follow-up (Table 11).

3.3.15.2. Methodological limitations

The small number of trials limited the applicability of random effects models because vague and weak informative prior distributions of the between-study variance have been shown to exert an unintentionally large degree of influence on any inference. Similarly, the random effects trial treatment had to be excluded from the GLMM model due to the small number of trials. Without including random effects terms, the full variability in the data cannot be included with the results.

Further, not all outcomes of interest were reported, in particular for various predefined subgroups. In particular, life-threatening bleeding and the composite ischemic/uncertain stroke or SE were often not reported in studies and CV-mortality was reported within vascular mortality (Table 12). All-cause mortality, intracranial bleeding, MI, and major GI bleeding were not reported by subgroups of interest TTR, age, and CHADS2 score.

3.3.15.3. External validity

As all major studies were performed as multinational, multicentre trials, generalizability to the Canadian health care system may be limited. Treatment of comorbidities and management of patients who are candidates for warfarin may differ between various countries. Further, TTR of warfarin treatment showed substantial differences between the trials and was also affected by geography. In RE-LY and ARISTOTLE, INR-control rates were substantially better than in ROCKET-AF. Skillful warfarin use as a predictor of treatment success might play a role to transfer trial results to the Canadian setting. In addition, the usual generalizability issues associated with randomized controlled trials need to be considered. In particular, the inclusion and exclusion criteria identifying the patients’ eligibility for the study may be different than in clinical practice, where prescribing practices may provide these drugs to patients with contraindications (such as low body weight or impaired renal function) leading to increased bleeding risk.

3.3.15.4. Clinical and methodological heterogeneity

The validity of indirect comparisons is determined by the extent of clinical and methodological trial similarity, so that differences in study populations, interventions, and outcomes definitions are potential sources of biases or errors.

ROCKET-AF included higher-risk patients with a CHADS2 score of at least 2, which limits the comparability of treatment effects to lower risk patients and to results from RE-LY and ARISTOTLE (which included more patients at low risk). In addition, ROCKET-AF also aimed to include a substantially higher number of patients with prior TIA, stroke, or SE (55%) compared to RE-LY (20%, not including prior SE) and ARISTOTLE (20%).

While ROCKET-AF and ARISTOTLE were designed as double-blind trials, warfarin treatment was not blinded in RE-LY, which might be a potential source of bias such as performance bias (i.e., systematic differences between groups in the care provided or exposure to factors other than the interventions of interest).

In ROCKET-AF, non-inferiority of rivaroxaban was achieved for the primary outcome stroke/SE in the safety of treatment population, mostly due to a rather high number of events in the transition phase after stopping the study drug. Thus, the magnitude of treatment effect was smaller in the ITT population. The implication of this observation for drug use in clinical routine is unclear, but does indicate differences in methodology between the trials.

While outcome definitions for efficacy end points are similar throughout the included trials, definitions of bleeding events differed substantially, in particular for minor bleeding (RE-LY) or non-major clinically relevant bleeding (ROCKET-AF and ARISTOTLE). Thus, these bleeding rates were markedly higher in RE-LY and ROCKET-AF compared to ARISTOTLE, limiting the comparability of results in network meta-analysis.

A number of areas were identified where there was clinical and methodological heterogeneity across trials, particularly for the ROCKET-AF trial (Table 18). The differences in the trials are partially reflected in the variation of the event rates in the warfarin control group seen across studies, particularly for the ROCKET-AF study. The small number of data points in our evidence network, however, restricted the ability to assess the impact of heterogeneity using standard approaches such as meta-regression. Heterogeneity was therefore assessed by conducting network meta-analysis using subgroup data reported in the individual RCTs. The methodological limitations with this approach are acknowledged (e.g., lack of information on similarity of patients across subgroups). In most instances, subgroup data was only available for the primary efficacy and safety outcomes in the trials — all-cause stroke/SE and major bleeding outcomes. As a consequence, the ability to explore the impact of heterogeneity between studies regarding patient population and study design for other outcomes considered in the network meta-analyses was limited (Table 18).

Table 18. Approach to Addressing Key Areas of Clinical and Methodological Heterogeneity.

Table 18

Approach to Addressing Key Areas of Clinical and Methodological Heterogeneity.

3.4. Individual Study Results

Four RCTs (N = 50,498) reported data on all-cause stroke/SE, major bleeding, all-cause mortality, intracranial bleeding, MI, and major GI bleeding. However, results for Section 3.4 and 3.5 are limited to data reported in ARISTOTLE, RE-LY, and ROCKET. The ARISTOTLE-J study was much smaller (N = 222) than ARISTOTLE (N = 18,201), RE-LY (N = 18,113), and ROCKET-AF (N = 14,264) and zero events were reported in both arms of ARISTOTLE-J.

A summary of the individual study result ORs for all-cause stroke/SE, major bleeding, all-cause mortality, intracranial bleeding, major GI bleeding, and MI are shown in Table 19 and Figure 2. Results using hazard ratios were similar and are reported in Appendix 7.12. A summary of the number needed to treat for all outcomes of interest are shown in Table 20 for:

Table 19. Summary of Individual Study Results — Odds Ratio (95% CI) for Each Outcome.

Table 19

Summary of Individual Study Results — Odds Ratio (95% CI) for Each Outcome.

Figure 2. Forest Plot of the Individual Study Results — Odds Ratio for Each Outcome.

Figure 2

Forest Plot of the Individual Study Results — Odds Ratio for Each Outcome. b.i.d. = twice daily; CrI = credibility interval; GI = gastrointestinal; SE = systemic embolism.

Table 20. Summary of Individual Study Results — Absolute Risk Reduction per 1,000 patients Treated Each Year.

Table 20

Summary of Individual Study Results — Absolute Risk Reduction per 1,000 patients Treated Each Year.

  • All-cause stroke/SE: With the exception of dabigatran 110 mg and rivaroxaban, all treatments achieved statistically significant reductions in all-cause stroke/SE relative to adjusted-dose warfarin. The use of dabigatran 150 mg produced the largest effect, with a reduction in odds of all-cause stroke/SE (OR [95% CI]: 0.65 [0.52 to 0.81]. The absolute risk reduction of all-cause stroke/SE ranged from 2 to 6 fewer events per 1,000 patients treated per year.
  • Major bleeding: Apixaban and dabigatran 110 mg achieved statistically significant reductions in major bleed relative to adjusted-dose warfarin. The use of apixaban produced the largest effects, with a reduction in odds of major bleed (OR [95% CI]: 0.69 [0.60 to 0.80]). The absolute risk reduction of major bleeding ranged from one more to eight fewer events per 1,000 patients treated per year.
  • All-cause mortality: Apixaban was associated with a statistically significant reduction in all-cause mortality relative to adjusted-dose warfarin (OR [95% CI]: 0.89 (0.79 to 0.997]). Relative to adjusted-dose warfarin, the reduction in odds for dabigatran 150 mg, dabigatran 110 mg, and rivaroxaban 20 mg ranged from 0.88 to 0.92, and all were not statistically significant. The absolute risk reduction of all-cause mortality ranged from three to four fewer events per 1,000 patients treated per year.
  • Intracranial bleeding: All treatments were associated with a statistically significant reduction in intracranial bleeding, with ORs ranging from 0.30 to 0.65. The use of dabigatran 110 mg produced the largest effect, with a reduction in odds of intracranial bleeding (OR [95% CI]: 0.30 [0.19 to 0.46]). The absolute risk reduction of intracranial bleeding ranged from three to five fewer events per 1,000 patients treated per year.
  • Major GI bleeding: No treatments were associated with a statistically significant reduction in major GIbleeding relative to adjusted-dose warfarin. However, dabigatran 150 mg (OR [95% CI]: 1.45 [1.13 to 1.85]) and rivaroxaban (OR [95% CI]: 1.60 [1.29 to 1.98]) were associated with a statistically significant increase in major GI bleeding. The absolute risk reduction of major GI bleeding ranged from eight more to one fewer events per 1,000 patients treated per year.
  • Myocardial infarction: No treatments were associated with a statistically significant reduction in MI relative to adjusted-dose warfarin. Apixaban was associated with the most favourable results (OR [95% CI]: 0.88 [0.66 to 1.17]). The ORs for the other treatments ranged from 0.918 to 1.31. The absolute risk reduction of all-cause mortality ranged from two more to two fewer events per 1,000 patients treated per year.

A summary of the absolute risk reduction per 1,000 patients treated per year by age for stroke/SE and major bleeding by centre TTR are shown in Table 21. More detailed results are provided in Appendix 7.7 for:

Table 21. Summary of Individual Study Results by TTR — Absolute Risk Reduction Per 1,000 Patients Treated Each Year.

Table 21

Summary of Individual Study Results by TTR — Absolute Risk Reduction Per 1,000 Patients Treated Each Year.

  • TTR < 66% — the absolute risk reduction of stroke/SE ranged from 2 to 9 per 1,000 patients treated in a year, whereas the absolute risk reduction for major bleed ranged from 2 to 11.
  • TTR ≥ 66% — the absolute risk reduction of stroke/SE ranged from 1 to 5 per 1,000 patients treated in a year, whereas the absolute risk reduction for major bleed ranged from 11 more to 6 fewer.

A summary of the absolute risk reduction per 1,000 patients treated per year by age for stroke/SE and major bleeding are shown in Table 22. More detailed results are provided in Appendix 7.12 for:

Table 22. Summary of Individual Study Results by Age — Absolute Risk Reduction Per 1,000 Patients Treated Each Year.

Table 22

Summary of Individual Study Results by Age — Absolute Risk Reduction Per 1,000 Patients Treated Each Year.

  • Age < 75 years — the absolute risk reduction of stroke/SE ranged from 1 to 5 per 1,000 patients treated in a year, whereas the absolute risk reduction for major bleed ranged from 2 to 11.
  • Age ≥ 75 years — the absolute risk reduction of stroke/SE ranged from 2 to 7 per 1,000 patients treated in a year, whereas the absolute risk reduction for major bleed ranged from 8 more to 15 fewer.

A summary of the absolute risk reduction per 1,000 patients treated per year by CHADS2 for stroke/SE and major bleeding are shown in Table 23. More detailed results are provided in Appendix 7.12 for:

Table 23. Summary of Individual Study Results by CHADS2 Score — Absolute Risk Reduction Per 1,000 Patients Treated Each Year.

Table 23

Summary of Individual Study Results by CHADS2 Score — Absolute Risk Reduction Per 1,000 Patients Treated Each Year.

  • CHADS2 < 2 — the absolute risk reduction of stroke/SE ranged from 0 to 4 per 1,000 patients treated over a year, whereas the absolute risk reduction for major bleed ranged from 7 to 9.
  • CHADS2 ≥2 — the absolute risk reduction of stroke/SE ranged from 2 to 6 per 1,000 patients treated over a year, whereas the absolute risk reduction for major bleed ranged from 1 more to 8 fewer.

3.5. Network Meta-Analysis Results

The evidence networks were restricted to three studies — ARISTOTLE, RE-LY, and ROCKET-AF (Figure 3) — as no events were reported in both arms for the many of outcomes in the other studies and trials with zero cells in both arms do not contribute information. Adjusted-dose warfarin was chosen as the reference group.

Figure 3. Schematic of Evidence Network Used for the Network Meta-analysis.

Figure 3

Schematic of Evidence Network Used for the Network Meta-analysis. b.i.d. = twice daily; od = .once daily

A summary of the ORs of each oral anticoagulant compared with warfarin, based on the Bayesian fixed-effects MTC network meta-analysis, are provided in Table 24 and Figure 4 for stroke/SE, major bleeding, all-cause mortality, intracranial bleeding, major GI bleeding, and MI. For:

Table 24. Summary of Results from the MTC Network Meta-analyses — Odds Ratio (95%CrI) for Each Outcome.

Table 24

Summary of Results from the MTC Network Meta-analyses — Odds Ratio (95%CrI) for Each Outcome.

Figure 4. Forest Plot of the Results from the MTC Network Meta-analysis for Each Outcome — Comparison with Warfarin.

Figure 4

Forest Plot of the Results from the MTC Network Meta-analysis for Each Outcome — Comparison with Warfarin. b.i.d. = twice daily; q.d. = once daily; SE = systemic embolism.

  • Stroke/SE — With the exception of dabigatran 110 mg and rivaroxaban, all treatments achieved statistically significant reductions in the odds of all-cause stroke/SE (range 0.65 to 0.80) relative to adjusted-dose warfarin. The use of dabigatran 150 mg produced the largest effects, with a reduction in odds of stroke/ SE (OR [95% confidence interval {CrI}]: 0.65 [0.52 to 0.81]) relative to adjusted-dose warfarin. There were no statistically significant differences between agents, the exception being dabigatran 150 mg versus 110 mg (OR [95% CrI]: 0.72 [0.58 to 0.91]).
  • Major bleedingApixaban and dabigatran 110 mg achieved statistically significant reductions in the odds of major bleed relative to adjusted-dose warfarin. The use of apixaban produced the largest effects, with a reduction in odds of major bleeding (OR [95% CrI]: 0.70 [0.61, 0.81]) relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the reduction in odds for other treatments ranged from 0.81 to 1.03.
  • All-cause mortality: Except for apixaban, none of the NOACs significantly reduced all-cause mortality (OR [95% CrI]: 0.90 [0.80 to 0.998]).
  • Intracranial bleeding: All treatments were associated with a statistically significant difference in intracranial bleeding compared with adjusted-dose warfarin with ORs, ranking from 0.30 to 0.66. The use of dabigatran 110 mg produced the largest effects in intracranial bleeding, with a reduction in odds (OR [95% CrI]: 0.29 [0.19 to 0.45]) relative to adjusted-dose warfarin.
  • Major GI bleeding: No treatments were associated with a statistically significant reduction in major GI bleeding relative to adjusted-dose warfarin. However, dabigatran 150 mg (OR [95% CrI]: 0.1.45 [1.14 to 1.86]) and rivaroxaban (OR [95% CrI]: 1.61 [1.30 to 1.99]) were associated with a statistically significant increase in major GI bleeding.
  • Myocardial infarction: No treatments were associated with a statistically significant reduction in MI relative to adjusted-dose warfarin. The ORs for other treatments ranged from 0.88 to 1.32.

The estimates of effects derived from fixed-effects Bayesian MTC analyses aligned closely with individual study results and frequentist network meta-analysis results in both direction and magnitude (Appendix 7.13). The point estimates for the Bayesian random-effects MTC analysis were similar to those reported in the Bayesian fixed-effects MTC, although the credible intervals were wider (Appendix 7.11).

A summary of the ORs by TTR, derived from a Bayesian fixed-effects MTC analysis, for stroke/SE and major bleeding are shown in Table 25. For:

Table 25. Summary of Results from MTC Analysis by TTR — Odds Ratio (95%CrI) for Each Outcome.

Table 25

Summary of Results from MTC Analysis by TTR — Odds Ratio (95%CrI) for Each Outcome.

  • Stroke/SE
    • For TTR < 66%, dabigatran 150 mg had a strong trend in reducing the odds of all-cause stroke/SE relative to adjusted-dose warfarin (OR [95% CrI]: 0.54 [0.40 to 0.74]). Relative to adjusted-dose warfarin, the reduction in odds for the other treatments ranged from 0.80 to 0.91.
    • For TTR ≥ 65%, no treatments had a strong trend in reducing the odds of all-cause stroke/SE relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the reduction in odds for all treatments ranged from 0.80 to 0.91
  • Major Bleeding
    • For TTR < 66%, apixaban, dabigatran 110 mg, and dabigatran 150 mg were all associated with a strong trend in reducing the odds of major bleeding relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, rivaroxaban was associated with a reduction in odds of 0.92.
    • For TTR ≥ 66%, apixaban was associated with a strong trend in reducing the odds of major bleeding relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, rivaroxaban had a trend of an increase in the odds of major bleeding (OR [95% CrI]: 1.30 [1.01, 1.69]), and dabigatran 110 mg and 150 mg had an odds of 0.86 and 1.15, respectively.

A summary of the ORs by age, derived from a Bayesian fixed-effects MTC analysis, for stroke/SE and major bleeding are shown in Table 26. For:

Table 26. Summary of Results from MTC Analysis by Age — Odds Ratio (95%CrI) for Each Outcome.

Table 26

Summary of Results from MTC Analysis by Age — Odds Ratio (95%CrI) for Each Outcome.

  • Stroke/SE
    • For age < 75 years, dabigatran 150 mg had a strong trend of reducing the odds of all-cause stroke/SE relative to adjusted-dose warfarin (OR [95% CrI]: 0.64 [0.46, 0.87]). Relative to adjusted-dose warfarin, the reduction in odds for other treatments ranged from 0.85 to 0.94.
    • For age ≥ 75 years, apixaban, dabigatran 150 mg and rivaroxaban all had strong trends in reducing the odds of all-cause stroke/SE relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the reduction in odds for dabigatran 110 mg was 0.89.
  • Major bleeding:
    • For age < 75 years, apixaban, dabigatran 110 mg and dabigatran 150 mg all were associated with a strong trend in reducing the odds of major bleeding relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the reduction in odds for rivaroxaban was 0.93.
    • For age ≥ 75 years, apixaban was associated with a strong trend in reducing the odds of major bleeding relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the other treatments were associated with an odds ratio ranging from 1.03 to 1.20.
    • With the exception of apixaban, the benefits diminished for age ≥ 75 years compared to age < 75 years.

A summary of the ORs by CHADS2 score, derived from a Bayesian fixed-effects MTC analysis, for stroke/SE and major bleeding are shown in Table 27. For:

Table 27. Summary of Results from MTC Analysis by CHADS2 Score — Odds Ratio (95%CI) for Each Outcome.

Table 27

Summary of Results from MTC Analysis by CHADS2 Score — Odds Ratio (95%CI) for Each Outcome.

  • Stroke/SE
    • For CHADS2 < 2, dabigatran 150 mg was associated with a strong trend to reducing the odds of all-cause stroke/SE relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the reduction in odds for apixaban and dabigatran 110 mg was 0.86 and 1.00, respectively. For rivaroxaban, results were not available, as patients with a CHADS2 < 2 were not recruited into the study.
    • For CHADS2 ≥ 2, apixaban 150 mg and dabigatran 150 mg were associated with strong trends in reducing the odds of all-cause stroke/SE relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the reduction in odds for dabigatran 110 mg and rivaroxaban was 0.89 and 0.88, respectively. For rivaroxaban, the reduction in the odds stroke/SE using the published as treated per protocol results was statistically significant, but not so for the results based on the intention-to-treat analysis.
  • Major Bleeding
    • For CHADS2 < 2, apixaban 110 mg and dabigatran 110 mg had a strong trend in reducing the odds of major bleeding relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the reduction in odds for dabigatran 150 mg was 0.77. Results were not available for rivaroxaban, as patients with CHADS2 < 2 were not included in the study.
    • For CHADS2 ≥ 2, apixaban was associated with a strong trend in reducing the odds of major bleeding relative to adjusted-dose warfarin. Relative to adjusted-dose warfarin, the reduction in odds for all treatments ranged from 0.74 to 1.03.

For the five treatments — apixaban, dabigatran 110 mg, dabigatran 150 mg, rivaroxaban, and adjusted-dose warfarinFigure 5 shows the distribution of the probabilities of each of these treatments being ranked first, second, third, fourth, or fifth (the possible five ranking positions) for each of the outcomes of stroke/SE, major bleeding, all-cause mortality, intracranial bleeding, major GI bleeding, and MI. These results — which should be considered from a descriptive and not inferential, statistical perspective — are based on the Bayesian fixed-effects MTC network meta-analysis. Apixaban had a low probability of being ranked fifth for all outcomes considered. Dabigatran had a high probability of being best for some outcomes (e.g., all-cause stroke/SE) and a high probability of being ranked last for others (e.g., MI). In particular:

Figure 5. Rankograms for Bayesian MTC Network Meta-analysis — for Each Outcome*.

Figure 5

Rankograms for Bayesian MTC Network Meta-analysis — for Each Outcome*. CV =cardiovascular; GI = gastrointestinal; SE = systemic embolism. * Intention to treat (ITT) population for all treatments for efficacy outcomes (i.e., stroke/SE, all-cause (more...)

  • Warfarin has a high probability of being one of the top treatment options for MI and GI bleeding, but a high probability of being ranked last for intracranial hemorrhage, all-cause mortality and all cause stroke/SE.
  • Apixaban: Apixaban has a low probability of being ranked last for all outcomes and a high probability of being best for major bleeding and GI bleeding.
  • Dabigatran 110 mg has a high probability of being one of the top treatment options for intracranial hemorrhage and major bleeding and a high probability of being one of the worst options for MI and all cause stroke/SE.
  • Dabigatran 150 mg has a high probability of being one of the top treatment options for all cause stroke/ SE and intracranial hemorrhage and a high probability of being one of the worst options for MI and major GI bleeding.
  • Rivaroxaban has a high probability of being one of the top treatment options for MI and a high probability of being one of the worst options for intracranial hemorrhage and major GI bleeding.

The rankograms in Figure 5 are not adjusted for heterogeneity across trials. In Figure 6, the rankograms from five subgroup analyses are shown: TTR < 66%, TTR ≥ 66%, age < 75 years, age ≥ 75 years, and CHADS2 ≥ 2. For all-cause stroke or SE, the probability of each rank is similar for the various subgroup analyses for warfarin 110 mg, apixaban 110 mg, and dabigatran 110 mg. However, for dabigatran 150 mg and rivaroxaban 110 mg, there are variations in the probability of each rank, depending on subgroup analysis. For example, for age ≥ 75 years and TTR ≥ 66%, the probability that dabigatran 150 mg is best reduces relative to the reference case analysis. There is less consistency across rankograms for major bleeding, particularly for warfarin and dabigatran. For example, the probability that warfarin is second best increases for the age ≥ 75 subgroup analysis relative to the reference case, whereas the probability that dabigatran is ranked fifth increases.

Figure 6. Rankograms for Bayesian MTC Network Meta-analysis by Subgroups — for Stroke/SE and Major Bleeding.

Figure 6

Rankograms for Bayesian MTC Network Meta-analysis by Subgroups — for Stroke/SE and Major Bleeding.

Copyright © 2012 Canadian Collaborative for Drug Safety, Effectiveness and Network Meta-Analysis.

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/

Bookshelf ID: NBK169813

Views

  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (4.9M)
  • Disable Glossary Links

Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...