An indirect comparison of dabigatran, rivaroxaban and apixaban for atrial fibrillation

Transcription

1 476 Schattauer 2012 Review Article An indirect comparison of dabigatran, rivaroxaban and apixaban for atrial fibrillation Simon Mantha 1 ; Jack Ansell 2 1 Department of Medicine, Lahey Clinic, Burlington, Massachusetts, USA; 2 Department of Medicine, Lenox Hill Hospital, New York, New York, USA Summary New oral anticoagulant drugs are emerging as alternatives to warfarin for the prevention of stroke in patients with non-valvular atrial fibrillation. Two agents are direct factor Xa inhibitors (rivaroxaban and apixaban), and the third is a direct thrombin inhibitor (dabigatran). They have been separately compared to warfarin in large randomised trials. Our objective was to indirectly compare the three agents to each other for major efficacy and safety outcomes. Studies were assessed for comparability and the odds ratios of selected outcomes for each anticoagulant versus one another were estimated indirectly. The three cohorts differed significantly in terms of CHADS 2 score and the number of individuals with a past history of stroke, transient ischemic attack or systemic embolism. The estimated odds ratio of stroke or systemic embolism was 1.35 for rivaroxaban vs dabigatran 150 mg (p=0.04), 0.97 for rivaroxaban versus dabigatran 110 mg (p=0.81), 1.22 for apixaban versus dabigatran 150 mg (p=0.18), 0.88 for apixaban versus dabigatran 110 mg (p=0.34) and 0.90 for apixaban versus rivaroxaban (p=0.43). The esti- mated odds ratio of major bleeding was 1.10 for rivaroxaban versus dabigatran 150 mg (p=0.36), 1.28 for rivaroxaban versus dabigatran 110 mg (p=0.02), 0.74 for apixaban versus dabigatran 150 mg (p=0.004), 0.87 for apixaban versus dabigatran 110 mg (p=0.17) and 0.68 for apixaban versus rivaroxaban (p<0.001). In conclusion, the available data indicate no significant difference in efficacy between dabigatran 150 mg and apixaban for the prevention of stroke or systemic embolism in patients with non-valvular atrial fibrillation. It appears however that apixaban is associated with less major bleeding than dabigatran 150 mg or rivaroxaban and that rivaroxaban is less effective than dabigatran 150 mg in preventing stroke or systemic embolism. Such an indirect comparison should be used only to generate hypotheses which need to be tested in a dedicated randomised trial comparing the three drugs directly. Keywords Atrial fibrillation, anticoagulants, stroke Correspondence to: Simon Mantha, MD, MPH Department of Hematology/Oncology Lahey Clinic Medical Center 41 Mall Road Burlington, MA 01805, USA Tel.: , Fax: simon.mantha@lahey.org Received: February 16, 2012 Accepted after major revision: May 7, 2012 Prepublished online: June 28, 2012 doi: /th Thromb Haemost 2012; 108: Introduction Atrial fibrillation affects about 9% of individuals aged 80 years or older (1) and its presence multiplies one s baseline risk of ischaemic stroke by about five (2). Oral anticoagulation with a vitamin K antagonist (VKA) reduces that risk by about 64% (3). However, chronic treatment confers a significant risk of bleeding, the most feared event being intracranial haemorrhage (ICH), which results in death or severe incapacitation in the majority of cases (4). In order to maximise efficacy and keep the risk of bleeding complications to a minimum, patients taking a VKA must be monitored closely with frequent International Normalised Ratio (INR) testing to assess drug effect. The existence of many food and drug interactions leads to the need for frequent dose adjustments in some patients (5). This complicates management of patients with atrial fibrillation in the community, and there is abundant evidence that warfarin is underused in this patient population (6). In the last few years, new orally administered anticoagulant drugs have emerged as potential alternatives to VKAs for the prevention of ischaemic stroke or systemic embolism in patients with chronic atrial fibrillation. Dabigatran etexilate, a direct thrombin inhibitor, has the longest track record with a landmark study published in 2009 that showed superior efficacy over warfarin at the highest dose assessed, with no difference in major bleeding but a decreased rate of ICH (7, 8). It appears to be cost-effective when compared to the current warfarin standard (9, 10). The results of two similar, large randomised trials assessing orally administered direct activated coagulation factor X inhibitors were recently reported (11, 12). Rivaroxaban was found to be non-inferior to warfarin in terms of its ability to prevent stroke or systemic embolism, with no difference in major bleeding but a decrease in the rate of ICH. Apixaban also compared favourably to warfarin, with superior efficacy and safety and with a survival benefit as well. The three novel oral anticoagulant drugs featured in the clinical trials mentioned above have predictable pharmacokinetics and do not require routine monitoring like VKAs (13 15). Additionally, they have a low potential for drug interactions and their effect is not altered significantly by diet (16 18). These agents are interest- Thrombosis and Haemostasis 108.3/2012

2 Mantha, Ansell: Indirect comparison of dabigatran, rivaroxaban and apixaban 477 ing alternatives to the current standard. They have not yet been compared side by side in a large randomised trial and due to cost issues, this kind of study may never be done. Management of patients on any of the new agents is distinctly different from that of individuals on warfarin (19, 20). Also, no randomised trial so far has evaluated if any of these drugs performs better in a subgroup of patients versus another, even though modelling data suggests all of the new anticoagulants are probably superior to warfarin in individuals at high risk of stroke or bleeding (21, 22). An indirect comparison can be helpful to try to assess the performance of one agent versus the other. We performed such a comparison, using data from the RE-LY, ROCKET AF and ARISTOTLE trials. Data sources A literature search was conducted in PubMed, using the keywords atrial fibrillation, dabigatran, rivaroxaban and apixaban. The only pertinent phase III trials identified were RE-LY, ROCKET AF and ARISTOTLE (7, 11, 12). Of note, the updated results for RE-LY were retrieved as well (8). The journal articles detailing the methodology of respective studies or pertinent sub-group analyses were assessed (23 26); appendices to the original papers were downloaded from the publisher s website. The data used for this work consisted only of publicly available information. Methods We used phase III trials comparing a novel anticoagulant to warfarin. We excluded pilot studies and phase II trials due to their greater heterogeneity in design and small size. Given that VKAs are the accepted standard of care for the prevention of stroke in patients with non-valvular atrial fibrillation, a warfarin comparator was all but ensured in most major studies evaluating a new agent. A common comparator was essential in order to perform a comparison of all novel anticoagulants of interest. Quality assessment Individual studies were compared in regards to the following aspects of their methodology: inclusion and exclusion criteria, dosing regimen for the experimental drug, definition of the outcomes reported, use of randomisation and blinding, population analysed (intent to treat, safety population or per protocol). The trials were evaluated for comparability in terms of the following characteristics of the participants: number of individuals lost to follow-up, CHADS 2 score, age, gender, prior treatment with a VKA, past history of stroke or systemic embolism, time in the therapeutic range (TTR) for the warfarin arm, aspirin use and creatinine clearance. Table 1: Study populations and measured endpoints. Characteristic RE-LY ROCKET AF ARISTOTLE Populations ITT All randomised patients (n=18,113) Endpoints analysed All randomised patients, except 93 from a GCP violating site (n=14,171) All randomised patients (n=18,201) mitt NA NA Individuals who received at least one dose of a study drug, including all events from randomization to January 30, 2011* Safety NA Individuals who received at least one dose of a study drug and were followed for events, up to two days after discontinuation (n=14,143) Per-protocol NA Same as safety, but only including subjects without a major protocol violation (n=13962) Stroke or systemic embolism ITT ITT, safety and per-protocol ITT Individuals who received at least one dose of a study drug and were followed for events, up to two days after discontinuation (n=18,140) Mortality ITT ITT, safety ITT Major ITT Safety Safety, mitt bleeding Intracranial haemorrhage ITT Safety Safety GCP, Good Clinical Practice; ITT, intention-to-treat; mitt, modified intention-to-treat; NA, not applicable. All the data for this table come from the original journal articles and their appendices as found on the publisher s website (7, 11, 12). *Size of the population could not be determined from published literature. NA Schattauer 2012 Thrombosis and Haemostasis 108.3/2012

3 478 Mantha, Ansell: Indirect comparison of dabigatran, rivaroxaban and apixaban Endpoints The primary efficacy endpoint for this indirect comparison was the composite of systemic embolism and stroke. The primary safety endpoint was major bleeding. Secondary endpoints were ICH, all-cause mortality, ischaemic or uncertain stroke, haemorrhagic stroke, myocardial infarction, gastrointestinal bleeding and premature discontinuation of the study drug. The odds ratio (OR) of any given outcome was taken from the intent to treat (ITT) population, because our aim was to assess superiority of one agent versus another. The three trials featured aimed to demonstrate non-inferiority to warfarin, and there is no consensus as far as which population is optimal for such studies, so methods were not uniform in this area. In the absence of ITT data, the results for the safety population were considered adequate. The details of the patient populations used in relation to the endpoints assessed are given in Table 1. For any given pair of novel anticoagulants, the estimate of the OR for a specific event was computed only if, for all practical purposes, the endpoint was defined comparably in the two original trials. Since the trials under study were contemporary, we expected this condition to be met for most endpoints. Statistical analysis Statistical techniques for indirect comparison were described elsewhere (27, 28). In brief, the OR of an event for patients receiving treatment A versus B (OR AB ) is estimated by dividing the OR for treatment A versus C by the OR for treatment B versus C. The standard error (SE) on the logarithmic scale for the indirect comparison AB can be easily computed (SE AB = [SE AC2 +SE BC2 ]); from this result the confidence interval (CI) for OR AB can be calculated. The SE can be calculated for any given direct OR, using the 2x2 table as reported in the original journal article (29). This method assumes that the results of the studies are independent. In order for the comparison to be valid, the trials need to have been performed under similar conditions; if they differ markedly for a given aspect, one should be able to assume that those differences do not modulate the measure of effect for treatment A versus C or B versus C, otherwise the analysis cannot be performed. The OR of selected events was calculated for rivaroxaban versus each dosage of dabigatran, for apixaban versus each dosage of dabigatran and for apixaban versus rivaroxaban. A two-sided test of hypothesis was applied to each of the five comparisons for primary efficacy and safety endpoints. The alpha level was Table 2: Patient characteristics. Characteristic Total number of subjects enrolled Individuals lost to follow-up (n) CHADS 2 scores (n) 0 or to 6 Age (years) Female (n) Previously treated with a VKA (n) Previous embolic episode (n) RE-LY ROCKET AF ARISTOTLE P-value 18,113 14,264 18, <0.001 < , ,183 6,455 1,859 6,516 5,882 12,402 5, (mean) 73 (median) 70 (median) -* 6,599 5,663 6,416 < ,989 8,904 10,401 < ,623 (stroke or TIA only) 7,845 (stroke, TIA or systemic embolism) 3,458 (stroke, TIA or systemic embolism) Mean time in the therapeutic range for ** the warfarin group (%) Aspirin use (n) 7,198 5,205 5,632 <0.001 Creatinine clearance < ml/min (n) >30 to 50 ml/min (n) 3,505 2,949 2,747 >50 ml/min (n) 14,592 11,205 15,105 VKA, vitamin K antagonist; TIA, transient ischaemic attack. The chi-square test was used for proportions where applicable. All the data for this table come from the original journal articles and their appendices as found on the publisher s website (7, 11, 12). *A formal test of hypothesis could not be performed because the mean age was not available for all studies. **A formal test of hypothesis could not be performed because the complete data on time in the therapeutic range was not available. Creatinine clearance was reported differently across the three trials. In RE-LY, the categories were <50 ml/min, ml/min and 80 ml/min. In ROCKET AF, the subgroups were <50 ml/min, ml/min and >80 ml/min. In ARISTOTLE, they were: not reported (n=79), 30 ml/min, >30 to 50 ml/min, >50 to 80 ml/ min and >80 ml/min. Patients with a clearance <30 ml/min were excluded from participation in RE-LY and ROCKET AF. Because the boundaries for the subgroups vary slightly across studies, the values for the categories listed in this table are approximations. Thrombosis and Haemostasis 108.3/2012 Schattauer 2012 <0.001

4 Mantha, Ansell: Indirect comparison of dabigatran, rivaroxaban and apixaban 479 set at Statistical analysis was performed with the R for Windows platform (R Foundation for Statistical Computing). The chisquare test was used to compare proportions. Results RE-LY, ROCKET AF and ARISTOTLE accrued about 50,000 individuals in aggregate. Table 2 highlights patient characteristics. Patients were randomised to dabigatran, rivaroxaban and apixaban, respectively, versus warfarin, the latter administered to achieve a goal international normalised ratio (INR) of 2.0 to 3.0. The updated results were used for the RE-LY study (8). The incidence of selected outcomes and the corresponding ORs used for the indirect comparisons are listed in Tables 3 and 4, respectively. All the data used to calculate the ORs come from peer-reviewed journals. A graphical summary of the different comparisons featured in this analysis is presented in Figure 1; only five of the six possible indirect assessments were computed, because the data for dabigatran 150 mg versus 110 mg twice a day is already available from the randomised trial. Assessment for comparability Inclusion criteria were based on clinical predictors of ischaemic stroke risk. ROCKET AF was notable in that it did not accrue individuals with a CHADS 2 score less than 2. The acceptable daily dose of aspirin was no more than 165 mg in all cases, and patients with a creatinine clearance less than 30 ml/minute (min) were almost always excluded. The primary efficacy outcome was defined in the same manner across the three trials, consisting of a composite of ischaemic, haemorrhagic or indeterminate stroke. The International Society on Thrombosis and Haemostasis (ISTH) definition of major bleeding was retained in all protocols (decrease in haemoglobin by at least 2 g/dl, transfusion of at least two units of red cells, bleeding at a critical site or resulting in death). RE-LY was distinct in that it randomised patients to two doses of the experimental drug or warfarin, while ROCKET AF and ARISTOTLE used one dose. Dabigatran and apixaban were administered twice a day, and rivaroxaban was given daily. RE-LY was the only trial not using a double-blind, doubledummy setup with sham INRs. Finally, while RE-LY reported efficacy and safety endpoints for the ITT population, the other studies used the safety population for bleeding events, except for ARIS- TOTLE which reported haemorrhagic strokes for the ITT population. The safety population was defined as all patients who had received at least one dose of medication on study and this cohort was very close in number to the ITT population. The pertinent characteristics of each cohort are listed in Table 2. The proportion of patients lost to follow-up was less than 1% for all studies. The mean CHADS 2 score was significantly higher in the ROCKET AF trial at 3.5 compared to 2.1 for RE-LY and ARISTOTLE (p<0.001). This was associated with more patients having a past history of stroke, transient ischaemic attack (TIA) or systemic embolism in ROCKET AF, with 55% of patient affected in comparison to 19% for ARISTOTLE and 20% (stroke or TIA only) for RE-LY (p<0.001). The patients were of similar ages between the three trials, with the middle of the distribution situated close to 71 years. No test of hypothesis could be performed for this parameter because the mean age was not available for all cohorts. Between 50% and 62% of patients had been previously treated with a VKA; the specific values varied significantly from study to study (p<0.001), but with such great numbers of patients even small differences will reach statistical significance. Aspirin use varied from 31% in ARISTOTLE to 36% in ROCKET AF and 40% in RE-LY (p<0.001). Mean TTR for the warfarin arm ranged from 55% in ROCKET AF to 62% in ARISTOTLE Table 3: Incidence of selected outcomes for the novel oral anticoagulants in comparison to warfarin. RE-LY ROCKET AF ARISTOTLE Dabigatran 110 mg (n=6,015) Dabigatran 150 mg (n=6,076) Warfarin (n=6,022) Rivaroxaban* Warfarin** Apixaban Warfarin Stroke or systemic embolism (n) Major bleeding (n) Intracranial haemorrhage (n) All-cause mortality (n) Ischaemic or uncertain stroke (n) Haemorrhagic stroke (n) Myocardial infarction (n) Gastrointestinal bleeding (n) ITT, intention-to-treat. All values for RE-LY are from the ITT population. All the data for this table come from the original journal articles and their appendices as found on the publisher s website (8, 11, 12). *n=7,081 for ITT population and 7,111 for safety population. **n=7,090 for ITT population and 7,125 for safety population. n=9,120 for ITT population and 9,088 for safety population. n=9,081 for ITT population and 9,052 for safety population. Using the ITT population. Using the safety population. Schattauer 2012 Thrombosis and Haemostasis 108.3/2012

5 480 Mantha, Ansell: Indirect comparison of dabigatran, rivaroxaban and apixaban Figure 1: Comparison of anticoagulants for the prevention of stroke and systemic embolism in atrial fibrillation. Graphical presentation derived from the work of Fadda et al. (36). and 64% in RE-LY; no formal test of hypothesis could be performed for this measure due to the limited amount of data available. The authors could not find information about the proportion of time spent below an INR of 2.0 and above an INR of 3.0 in the peer-reviewed literature for any of the trials. Finally, the proportion of participants with a creatinine clearance inferior or equal to 50 ml/min varied from 17% to 21%; very few individuals had a clearance inferior or equal to 30 ml/min (p<0.001). Stroke and systemic embolism CI= ). The derived indirect ORs are 1.35 for rivaroxaban versus dabigatran 150 mg (95% CI= , p=0.04), 0.97 for rivaroxaban versus dabigatran 110 mg (95% CI= , p=0.81), 1.22 for apixaban versus dabigatran 150 mg (95% CI= , p=0.18), 0.88 for apixaban versus dabigatran 110 mg (95% CI= , p=0.34) and 0.90 for apixaban versus rivaroxaban (95% CI= , p=0.43) ( Table 5, Fig. 2). Using an alpha of 0.05, rivaroxaban appears to be associated with more ischaemic events than dabigatran 150 mg. The HR of stroke or systemic embolism was 0.72 for dabigatran 150 mg versus 110 mg (95% CI= , p=0.004), as published in the original journal article. When compared to warfarin, the hazard ratio (HR) of an event was 0.90 for dabigatran 110 mg twice a day (95% confidence interval [CI]= ), 0.65 for dabigatran 150 mg twice a day (95% CI= ), 0.88 for rivaroxaban 20 mg daily (95% CI= ) and 0.79 for apixaban 5 mg twice a day (95% Bleeding The HR of major bleeding in comparison to warfarin for dabigatran 110 mg twice a day was 0.80 (95% CI= ),while it was Table 4: Odds ratio of selected outcomes for the novel oral anticoagulants in comparison to warfarin. Dabigatran 110 mg Dabigatran 150 mg Rivaroxaban Apixaban OR OR OR OR Stroke or systemic embolism 0.90 ( ) 0.65 ( ) 0.88 ( ) 0.79 ( ) Major bleeding 0.80 ( ) 0.94 ( ) 1.03 ( ) 0.69 ( ) Intracranial haemorrhage 0.30 ( ) 0.41 ( ) 0.65 ( ) 0.42 ( ) All-cause mortality 0.91 ( ) 0.88 ( ) 0.92 ( ) 0.89 ( ) Ischaemic or uncertain stroke 1.12 ( ) 0.77 ( ) 0.91 ( ) 0.92 ( ) Haemorrhagic stroke 0.31 ( ) 0.26 ( ) 0.58 ( ) 0.51 ( ) Myocardial infarction 1.31 ( ) 1.29 ( ) 0.80 ( ) 0.88 ( ) Gastrointestinal bleeding 1.09 ( ) 1.49 ( ) 1.47 ( ) 0.88 ( ) CI, confidence interval; OR, odds ratio. These results were computed only from information contained in the original journal articles and their appendices as found on the publisher s website (8, 11, 12). Thrombosis and Haemostasis 108.3/2012 Schattauer 2012

6 Mantha, Ansell: Indirect comparison of dabigatran, rivaroxaban and apixaban for dabigatran 150 mg twice a day (95% CI= ), 1.04 for rivaroxaban (95% CI= ) and 0.69 for apixaban (95% CI=0.60 to 0.80). In this case the confidence intervals did not overlap for rivaroxaban and apixaban, suggesting a lower rate of bleeding for the later. This was supported by the indirect comparison. The OR of major bleeding was estimated at 1.10 for rivaroxaban versus dabigatran 150 mg (95% CI= , p=0.36), 1.28 for rivaroxaban versus dabigatran 110 mg (95% CI= , p=0.02), 0.74 for apixaban versus dabigatran 150 mg (95% CI= , p=0.004), 0.87 for apixaban versus dabigatran 110 mg (95% CI= , p=0.17) and 0.68 for apixaban versus rivaroxaban (95% CI= , p<0.001) ( Table 5, Fig. 2). Using an alpha of 0.05, this indirect comparison suggests that apixaban is associated with less major bleeding episodes when compared to dabigatran 150 mg twice a day and rivaroxaban. Also, dabigatran 110 mg appears to be safer than rivaroxaban. The HR of major bleeding was 1.16 for dabigatran 150 mg versus 110 mg (95% CI= , p=0.04), as published in the original journal article. All three investigational agents were associated with lower rates of ICH when compared with warfarin. The HR of this complication was 0.30 for dabigatran 110 mg versus warfarin (95% CI= ), 0.41 for dabigatran 150 mg versus warfarin (95% CI= ), 0.67 for rivaroxaban versus warfarin (95% CI= ) and 0.42 for apixaban versus warfarin (95% CI= ). The CIs for those values largely overlapped, except for dabigatran 110 mg and rivaroxaban. The indirect comparison seemed to suggest that rivaroxaban was associated with more events than dabigatran 110 mg, with an OR of 2.20 (95% CI= , p=0.005). The OR of ICH was estimated at 1.58 for rivaroxaban versus dabigatran 150 mg (95% CI= , p=0.08), 1.02 for apixaban versus dabigatran 150 mg ( , p=0.95), 1.42 for apixaban versus dabigatran 110 mg (95% CI= , p=0.21) and 0.64 for apixaban versus rivaroxaban (95% CI= , p=0.07). The HR of ICH was 1.39 for dabigatran 150 mg versus 110 mg (95% CI= , p=0.19), as published in the original journal article. Mortality and other endpoints Only apixaban was shown to have a clear survival benefit when compared to warfarin, with a HR of all-cause mortality of 0.89 (95% CI= ). The result for dabigatran 150 mg versus warfarin was of borderline statistical significance, at 0.88 (95% CI= ). Rivaroxaban was deemed non-inferior to warfarin in regards to this endpoint, with a HR of 0.92 (95% CI= ). The indirect comparison for all-cause mortality did not suggest superiority of any new agent versus another with the estimated OR of that event being 1.04 for rivaroxaban vs dabigatran 150 mg ( , p=0.70), 1.01 for rivaroxaban versus dabigatran 110 mg (95% CI= , p=0.95), 1.01 for apixaban versus dabigatran 150 mg ( , p=0.93), 0.98 for apixaban versus dabigatran 110 mg (95% CI= , p=0.80) and 0.97 for apixaban versus rivaroxaban ( , p=0.74). The HR of death was 0.97 for dabigatran 150 mg versus 110 mg (95% CI= , p=0.66), as published in the original journal article. In terms of haemorrhagic stroke, rivaroxaban seemed to fare more poorly than dabigatran 150 mg, with an OR of 2.20 (95% Table 5: Relative risk of selected outcomes for the six possible pairwise comparisons of novel oral anticoagulants. Endpoint R vs. D150 R vs. D110 A vs. D150 A vs. D110 A vs. R D150 vs. D110* OR p OR p OR p OR p OR p HR p Stroke or systemic embolism 1.35 ( ) ( ) ( ) ( ) ( ) ( ) Major bleeding 1.10 ( ) ( ) ( ) ( ) ( ) < ( ) 0.04 Intracranial haemorrhage 1.58 ( ) ( ) ( ) ( ) ( ) ( ) 0.19 All-cause mortality 1.04 ( ) ( ) ( ) ( ) ( ) ( ) 0.66 Ischaemic or uncertain stroke 1.19 ( ) ( ) ( ) ( ) ( ) ( ) Haemorrhagic stroke 2.20 ( ) ( ) ( ) ( ) ( ) ( ) 0.67 Myocardial infarction 0.62 ( ) ( ) ( ) ( ) ( ) ( ) 0.88 Gastrointestinal bleeding 0.99 ( ) ( ) ( ) ( ) ( ) ( ) R, rivaroxaban; D150, dabigatran 150 mg; D110, dabigatran 110 mg; A, apixaban; CI, confidence interval; OR, odds ratio; HR, hazard ratio. These results were computed only from information contained in the original journal articles and their appendices as found on the publisher s website (8, 11, 12). *Original results as published for dabigatran 150 mg vs. 110 mg. Schattauer 2012 Thrombosis and Haemostasis 108.3/2012

7 482 Mantha, Ansell: Indirect comparison of dabigatran, rivaroxaban and apixaban Figure 2: Relative risk of selected events for any of the new oral agents in comparison to each other, with 95% confidence interval. Data shown for stroke or systemic embolism and major bleeding. Abbreviations: R, rivaroxaban; D150, dabigatran 150 mg; D110, dabigatran 110 mg, A, apixaban. CI= , p=0.05), even though this value was of borderline statistical significance. On the other hand, rivaroxaban was also associated with a lower occurrence myocardial infarction than either dosage of dabigatran, with an OR of 0.62 (95% CI= , p=0.02) when compared to dabigatran 150 mg and 0.61 (95% CI= , p=0.02) when considering the 110 mg dose. Apixaban appeared to confer a lower risk of gastrointestinal bleeding than dabigatran 150 mg and rivaroxaban, with an OR of 0.59 (95% CI= , p=0.003) and 0.60 (95% CI= , p=0.003), respectively. Of the three new oral anticoagulants, dabigatran is the only one which was reported to be associated with more dyspepsia than warfarin, with about 10% of patients affected. The rate of drug discontinuation as measured after two years of observation was higher for dabigatran in comparison to warfarin (p<0.001); the proportions of patients having stopped the drug were 21.2% and 16.6%, respectively. In ARISTOTLE, 25.3% of patients allocated to apixaban stopped the drug before the end of the study, compared to 27.5% in the warfarin group (p=0.001). In ROCKET AF, 23.7% of patients discontinued the drug before an endpoint event and the termination date in the rivaroxaban group, compared to 22.2% for warfarin; the result of a formal test of hypothesis was not given. Discussion Any indirect comparison like this one relies on the assumption that the trials included in the analysis share a common methodology. The RE-LY, ROCKET AF and ARISTOTLE studies enrolled or randomised all their participants between December 2005 and April 2010, so their respective time frames largely overlap. This certainly contributed to the very similar definitions they used for the primary and secondary outcomes. The use of the same comparator arm in all studies, consisting of treatment with warfarin using a goal INR interval of 2.0 to 3.0 was an essential requirement to go ahead with the indirect comparison. A unique characteristic of the three randomised trials discussed here is their large size, which resulted in rather narrow CIs for the measures of effect reported. This is important, since the standard error (SE) for each element of the comparison adds to the total SE of the final result. The inclusion criteria for the three trials were similar as well, even though ROCKET AF accrued a group of individuals at higher risk of embolic events. If that characteristic had impacted the OR of any given event for rivaroxaban versus warfarin, the indirect comparison for rivaroxaban versus dabigatran or apixaban would have been biased. It is difficult to assess for effect modification, because the information available from the original papers is limited in that regard. The p-value for interaction between treatment group allocation and CHADS 2 score for the primary efficacy endpoint was above the threshold of 0.05 for the three trials. Also, a history of previous embolic event was not shown to interact significantly with the effect of allocated treatment on the primary efficacy endpoint in any of the trials. If the efficacy of rivaroxaban in comparison to warfarin varied markedly across strata of CHADS 2 score, one would expect statistically significant interaction between this parameter and treatment allocation. Higher CHADS 2 scores and a history of ischaemic stroke have also been associated with an increased risk of major bleeding in pa- Thrombosis and Haemostasis 108.3/2012 Schattauer 2012

8 Mantha, Ansell: Indirect comparison of dabigatran, rivaroxaban and apixaban 483 tients on chronic oral anticoagulation with warfarin (30, 31). In ROCKET AF, there was no interaction between CHADS 2 score or a history of prior embolism and treatment group allocation in regards to the occurrence of major/non-major clinically relevant bleeding. In ARISTOTLE, CHADS 2 score and prior history of stroke or TIA were not found to interact either with treatment group when it comes to major bleeding. Having been recently started on warfarin and concomitant aspirin intake are associated with a higher risk of bleeding event as well (32, 33). In that regard, the proportion of patients with a history of previous VKA intake and the numbers of those with ongoing aspirin use varied significantly across studies. However, there was no report of interaction between these two variables and drug effect, except for a prior history of VKA use in ROCKET AF, where it seemed to confer a marginally higher rate of major or non-major clinically relevant bleeding to rivaroxaban users in comparison to warfarin, with a HR of 1.09 (95% CI= , p=0.044); interaction data was not reported for major bleeding alone, and the authors considered that drug effect was consistent across all pre-specified subgroups in terms of efficacy and safety. Importantly, the influence of previous warfarin exposure was assessed in a separate paper and no effect was found on the benefits of dabigatran at either dose when compared to warfarin (34). Interaction data for major bleeding and use of aspirin at randomisation was only reported for ARISTOTLE (p=0.40). As for the TTR, it was not shown to interact with treatment effects in ROCKET AF (p=0.736 for interaction). For RE-LY, in centres with a lower TTR there was an advantage for dabigatran over warfarin in terms of major bleeding which was not seen in centres with a better TTR (p=0.03 for interaction) (26). However, the mean TTRs are very similar for RE-LY and ARISTOTLE, at 64% and 62%, respectively. It is difficult to imagine that such a small difference would have resulted in a safety benefit for apixaban. The mean TTR appears to be lower for ROCKET AF, at 55%. It is expected that more patients with a sub-therapeutic INR in the warfarin group of ROCKET AF would increase the apparent efficacy of rivaroxaban in comparison to dabigatran and apixaban when performing an indirect comparison. Finally, the distributions of gender and age are very similar between cohorts and would not be expected to affect the results of the indirect comparison. However, as detailed in Table 2, renal function is difficult to compare across trials and there could be significant differences leading to bias. In aggregate, we believe these data do not suggest that the safety benefit for apixaban versus dabigatran and rivaroxaban observed in this comparison is due to interaction between a patient characteristic and the effect of treatment allocation. The results of the indirect comparison between the three novel agents suggest that apixaban has equivalent efficacy and might be safer than dabigatran 150 mg and rivaroxaban in regard to major bleeding, with a relative risk reduction of about one fourth when compared to dabigatran 150 mg and about one third when compared to rivaroxaban. As with other anticoagulants, the risk of bleeding with rivaroxaban seems to be dose related, at least in the presence of antiplatelet therapy (35). It is possible that if a different dose had been used in ROCKET AF, the result of the comparison with dabigatran would have been different in regards to bleeding complications. The original trials showed that compared to warfarin, dabigatran 150 mg and apixaban are superior to warfarin for prevention of stroke or systemic embolism, with an overall relative risk reduction of about 30%; only dabigatran 150 mg twice a day showed a significant reduction in ischaemic or uncertain stroke. In the case of apixaban, there was a significant improvement in mortality. Administration of dabigatran 150 mg on the other hand was associated with a trend toward improved survival, which was of borderline statistical significance. Gastric intolerance seems to be a problem in a minority of patients taking dabigatran. This has not been reported with rivaroxaban or apixaban. Finally, all three novel agents performed better than warfarin in regard to the risk of ICH, overall halving the rate of this serious complication which often results in considerable mortality or incapacitation. Limitations Combining the results of different trials always raises questions about the validity of the computed measures of effects; in such an indirect comparison, undetected confounders or effect modifiers can result in an inaccurate estimated measure of effect. For example, the CHADS 2 scores were higher in the ROCKET AF cohort; if the benefit of rivaroxaban versus warfarin was lower for a score of 3 compared to a score of 2 for any endpoint (i.e. effect modification), the indirect comparison with dabigatran and apixaban would have been biased against rivaroxaban. Also, if the combination with aspirin was more detrimental with any of the new drugs in terms of bleeding compared to warfarin, then the comparison for this endpoint would unduly favor apixaban when compared to dabigatran, because more patients in RE-LY were taking aspirin in than in ARISTOTLE. However, in the case of this analysis we have not come across a difference in design or patient population which to our understanding would have caused the estimated ORs for the primary efficacy and safety endpoints to be altered in a major way. Individual patient data meta-analysis of RE- LY, ROCKET AF and ARISTOTLE could certainly adjust for some of the potential confounders. However, such a study would require the sharing of proprietary data and might never be performed. Also, performing multiple comparisons is associated with a higher probability of obtaining a statistically significant result. However, we believe that in spite of those shortcomings it was important to do the analysis in order to help generate hypotheses which can later be tested in formal randomised trials. Conclusion In this indirect comparison, dabigatran and apixaban have similar efficacy compared to warfarin, the current standard of care for the prevention of stroke in patients with non-valvular atrial fibrillation. Dabigatran 150 mg seems to be more effective than rivaroxaban in decreasing the incidence of stroke or systemic embolism. The results suggest an improved safety profile for apixaban Schattauer 2012 Thrombosis and Haemostasis 108.3/2012

9 484 Mantha, Ansell: Indirect comparison of dabigatran, rivaroxaban and apixaban when compared to rivaroxaban and the currently approved dosage of dabigatran. Also, rivaroxaban seems to be associated with more bleeding than dabigatran 110 mg. Finally, RE-LY demonstrated that dabigatran 150 mg was more effective than the 110 mg dosage for prevention of the main efficacy endpoint. However, the analysis presented here is subject to confounding by unmeasured factors and performing multiple comparisons increases the risk of false positive, so the differences observed should not be considered as grounds to administer one drug versus the other and the results should be used strictly for hypothesis generation. This work suggests that a randomised trial comparing apixaban to dabigatran and rivaroxaban would be warranted. Conflicts of interest Dr. Ansell has received consulting fees from Bristol Myers Squibb, Janssen, Boehringer Ingelheim and Daiichi Sankyo, as well as payement for development of educational presentations from Daiichi Sankyo; he sits on the ARISTOTLE trial Executive Committee. Dr. Mantha reports no conflicts of interest. References 1. Go AS, Hylek EM, Phillips KA, et al. Prevalence of diagnosed atrial fibrillation in adults: national implications for rhythm management and stroke prevention: the AnTicoagulation and Risk Factors in Atrial Fibrillation (ATRIA) Study. J Am Med Assoc 2001; 285: Wolf PA, Abbott RD, Kannel WB. Atrial fibrillation as an independent risk factor for stroke: the Framingham Study. Stroke 1991; 22: Hart RG, Pearce LA, Aguilar MI. Meta-analysis: antithrombotic therapy to prevent stroke in patients who have nonvalvular atrial fibrillation. Ann Intern Med 2007; 146: Rosand J, Eckman MH, Knudsen KA, et al. The effect of warfarin and intensity of anticoagulation on outcome of intracerebral hemorrhage. Arch Intern Med 2004; 164: Holbrook AM, Pereira JA, Labiris R, et al. Systematic overview of warfarin and its drug and food interactions. Arch Intern Med 2005; 165: Albers GW, Yim JM, Belew KM, et al. Status of antithrombotic therapy for patients with atrial fibrillation in university hospitals. Arch Intern Med 1996; 156: Connolly SJ, Ezekowitz MD, Yusuf S, et al. Dabigatran versus Warfarin in Patients with Atrial Fibrillation. N Engl J Med 2009; 361: Connolly SJ, Ezekowitz MD, Yusuf S, et al. Newly identified events in the RE-LY trial. N Engl J Med 2010; 363: Freeman JV, Zhu RP, Owens DK, et al. Cost-effectiveness of dabigatran compared with warfarin for stroke prevention in atrial fibrillation. Ann Intern Med 2011; 154: Sorensen SV, Kansal AR, Connolly S, et al. Cost-effectiveness of dabigatran etexilate for the prevention of stroke and systemic embolism in atrial fibrillation:a Canadian payer perspective. Thromb Haemost 2011; 105: Patel MR, Mahaffey KW, Garg J, et al. Rivaroxaban versus warfarin in nonvalvular atrial fibrillation. N Engl J Med 2011; 365: Granger CB, Alexander JH, McMurray JJ, et al. Apixaban versus warfarin in patients with atrial fibrillation. N Engl J Med 2011; 365: Stangier J, Rathgen K, Stahle H, et al. The pharmacokinetics, pharmacodynamics and tolerability of dabigatran etexilate, a new oral direct thrombin inhibitor, in healthy male subjects. Br J Clin Pharmacol 2007; 64: Mueck W, Becka M, Kubitza D, et al. Population model of the pharmacokinetics and pharmacodynamics of rivaroxaban--an oral, direct factor xa inhibitor--in healthy subjects. Int J Clin Pharmacol Ther 2007; 45: He K, Luettgen JM, Zhang D, et al. Preclinical pharmacokinetics and pharmacodynamics of apixaban, a potent and selective factor Xa inhibitor. Eur J Drug Metab Pharmacokinet 2011; 36: Stangier J, Clemens A. Pharmacology, pharmacokinetics, and pharmacodynamics of dabigatran etexilate, an oral direct thrombin inhibitor. Clin Appl Thromb Hemost 2009; 15 (Suppl 1): 9S-16S. 17. Laux V, Perzborn E, Kubitza D, et al. Preclinical and clinical characteristics of rivaroxaban: a novel, oral, direct factor Xa inhibitor. Semin Thromb Hemost 2007; 33: Wong PC, Pinto DJ, Zhang D. Preclinical discovery of apixaban, a direct and orally bioavailable factor Xa inhibitor. J Thromb Thrombolysis 2011; 31: Huisman MV, Lip GYH, Diener HC, et al. Dabigatran etexilate for stroke prevention in patients with atrial fibrillation: Resolving uncertainties in routine practice. Thromb Haemost 2012; 107: Pengo V, Crippa L, Falanga A, et al. Questions and answers on the use of dabigatran and perpectives on the use of other new oral anticoagulants in patients with atrial fibrillation. Thromb Haemost 2011; 106: Goldhaber SZ. What s the go to anticoagulant for stroke prevention in atrial fibrillation? Thromb Haemost 2012; 107: Banerjee A, Lane DA, Torp-Pedersen C, et al. Net clinical benefit of new oral anticoagulants (dabigatran, rivaroxaban, apixaban) versus no treatment in a real world atrial fibrillation population: A modelling analysis based on a nationwide cohort study. Thromb Haemost 2012; 107: Ezekowitz MD, Connolly S, Parekh A, et al. Rationale and design of RE-LY: randomized evaluation of long-term anticoagulant therapy, warfarin, compared with dabigatran. Am Heart J 2009; 157: Rivaroxaban-once daily, oral, direct factor Xa inhibition compared with vitamin K antagonism for prevention of stroke and Embolism Trial in Atrial Fibrillation: rationale and design of the ROCKET AF study. Am Heart J 2010; 159: Lopes RD, Alexander JH, Al-Khatib SM, et al. Apixaban for Reduction In Stroke and Other ThromboemboLic Events in Atrial Fibrillation (ARISTOTLE) trial: Design and rationale. Am Heart J 2010; 159: Wallentin L, Yusuf S, Ezekowitz MD, et al. Efficacy and safety of dabigatran compared with warfarin at different levels of international normalised ratio control for stroke prevention in atrial fibrillation: an analysis of the RE-LY trial. Lancet 2010; 376: Glenny AM, Altman DG, Song F, et al. Indirect comparisons of competing interventions. Health Technol Assess 2005; 9: 1 134, iii-iv. 28. Altman DG, Bland JM. Interaction revisited: the difference between two estimates. Br Med J 2003; 326: Bland JM, Altman DG. Statistics notes. The odds ratio. Br Med J 2000; 320: Pisters R, Lane DA, Nieuwlaat R, et al. A novel user-friendly score (HAS-BLED) to assess 1-year risk of major bleeding in patients with atrial fibrillation: the Euro Heart Survey. Chest 2010; 138: Poli D, Antonucci E, Marcucci R, et al. Risk of bleeding in very old atrial fibrillation patients on warfarin: relationship with ageing and CHADS2 score. Thromb Res 2007; 121: Landefeld CS, Goldman L. Major bleeding in outpatients treated with warfarin: incidence and prediction by factors known at the start of outpatient therapy. Am J Med 1989; 87: Shireman TI, Howard PA, Kresowik TF, et al. Combined anticoagulant-antiplatelet use and major bleeding events in elderly atrial fibrillation patients. Stroke 2004; 35: Ezekowitz MD, Wallentin L, Connolly SJ, et al. Dabigatran and Warfarin in Vitamin K Antagonist-Naive and -Experienced Cohorts With Atrial Fibrillation. Circulation 2010; 122: Mega JL, Braunwald E, Wiviott SD, et al. Rivaroxaban in patients with a recent acute coronary syndrome. N Engl J Med 2012; 366: Fadda V, Maratea D, Trippoli S, et al. Network meta-analysis. Results can be summarised in a simple figure. Br Med J 2011; 342: d1555. Thrombosis and Haemostasis 108.3/2012 Schattauer 2012