OUP user menu

Adjusted indirect comparison of new oral anticoagulants for stroke prevention in atrial fibrillation

L. Testa, M. Agnifili, R.A. Latini, R. Mattioli, S. Lanotte, F. De Marco, J. Oreglia, A. Latib, S. Pizzocri, M.L. Laudisa, N. Brambilla, F. Bedogni
DOI: http://dx.doi.org/10.1093/qjmed/hcs114 949-957 First published online: 6 July 2012

Abstract

Background: Vit-K antagonists are the therapy of choice to prevent thromboembolic events due to atrial fibrillation since many years. New oral anticoagulants (NOA) showed encouraging results vs. warfarin but there are no data directly comparing different NOA. We performed an adjusted indirect meta-analysis.

Methods: Randomized controlled trials (RCTs) were searched. Efficacy end points were the cumulative rate of thomboembolic stroke (TES) and systemic embolism (SE). Main safety end point was the rate of hemorrhagic stroke (HS).

Results: Three RCTs (50578 patients) were included. Overall, NOA were comparable to warfarin according to the cumulative risk of TES and SE, as well as for TES alone. NOA were associated with a reduced rate of SE [OR 0.64 (0.44, 0.94], P = 0.02]. Compared to warfarin, NOA were associated with a significantly reduced risk of HS [OR 0.43 (0.34, 0.55), P < 0.001, NNT to avoid a HS 153] and all cause death [OR 0.90 [0.84, 0.96], P = 0.03, NNT to save one fatality 43]. Head to head comparison showed that in terms of cumulative rate of TES/SE, as well as of TES, none of the NOA was significantly superior to the others (all Ps > 0.05). Rivaroxaban showed superiority in the prevention of SE. Dabigatran 150 mg/twice daily was associated with the largest reduction in the risk of HS vs. warfarin and vs. other NOA. Overall mortality was quite comparable across NOA.

Conclusions: Overall superiority of NOA over warfarin is largely influenced by the reduction of HS. Dabigatran 150 mg/twice daily seems to have the best risk/benefit profile.

Introduction

Atrial fibrillation (AF) is associated with a 4- to 5-fold increased risk of thromboembolic stroke (TES).1 It accounts for up to 15% of strokes regardless the age. This percentage reaches 30% in people >75 years.2

For more than 50 years, Vitamin K antagonists have been the primary medication to reduce the risk of thromboembolic events in patients with AF. Nevertheless, along with proven clinical efficacy, they have several limitations, including a number of interactions with other drugs and food, and need for regular blood monitoring for dose adjustments.1,2

These drawbacks led to several attempts to find reliable alternative approaches, however, none eventually fulfilled the promise.3

The scenario has dramatically changed in the last few years. As compared to Warfarin, Dabigatran,4 Rivaroxaban5 and Apixaban6 showed encouraging results.

Of note, a ‘head-to-head’ comparison of these new drugs has never been done and is unlikely in the future.

Indirect comparison meta-analyses, adjusted by a common control, enable head-to-head comparisons of two treatments originally compared in controlled trials to a common reference treatment.7 While obviously second to direct comparisons in the hierarchy of clinical evidence, their use has been recommended and their validity scientifically established.7,8 Building upon previous similar works from our group,9,10 we thus aimed to perform an adjusted indirect comparison meta-analysis of new oral anticoagulants (NOA) for stroke prevention in AF.

Methods

Design

The present review was performed according to the Cochrane Collaboration and PRISMA statements.11,12

Search strategy

PubMed was searched with established methods13 for randomized phase-III trials comparing warfarin with NOA in patients with non valvular AF without language restrictions (updated on 1 March 2012; see the Appendix A for further details). In addition, Google Scholar, The Cochrane Library and Scopus were also searched for pertinent citations. References of retrieved studies were checked for additional studies (backward snowballing) and 2008–11 conference proceedings of the American College of Cardiology, American Heart Association and European Society of Cardiology scientific sessions were also manually searched. No language restriction was enforced.

Study selection

Citations were first scanned at the title/abstract level. Shortlisted studies were then retrieved in full text.

Specifically, inclusion criteria were randomized allocation, controlled comparisons against a vitamin K antagonist, patients with AF, follow-up of at least 1 year and intention-to-treat analysis.

Abstraction and appraisal

The primary efficacy end-point was the cumulative rate of TES and systemic embolisms (SE) at the longest available follow-up. The primary safety end-point was the rate of hemorrhagic stroke (HS) at the longest follow up. Rates of extracranial major bleeding (EMB), all cause mortality and myocardial infarction (MI) were also computed.

In addition, study validity was appraised according to the risk of bias tool recommended by The Cochrane Collaboration.

Analysis

Review Manager 5.0 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, Denmark, http://ims.cochrane.org/revman) and Indirect Meta-analysis Tool (METCARDIO, Turin, Italy, http://www.metcardio.org/) were used for analysis. Outcomes were initially pooled using fixed-effect odds ratios (OR) with 95% confidence intervals (CI) and verified, in the presence of significant heterogeneity, by means of random effect in order to reduce possible sources of bias.

Adjusted indirect comparison of pooled estimates were then performed according to Song et al.8 Specifically, we generated, from fixed-effect ORs comparing dabigatran, apixaban and rivaroxaban vs. warfarin, interaction ORs for: (i) dabigatran vs. apixaban; (ii) dabigatran vs. rivaroxaban and; (iii) apixaban vs. rivaroxaban with pertinent 95% CI and Z scores for two-tailed hypothesis testing (P-significant if P < 0.05).Specifically, these interaction ORs are calculated according to the formulas reported in Appendix B, where ‘ln’ is the natural logarithm, and ‘var’ is the variance. Both doses of dabigatran have been considered. Trial inconsistency was assessed with I2. Given the few shortlisted studies, publication bias was not appraised. We also calculated the number needed to treat (NNT) or to harm (NNH) as the inverse of the absolute risk reduction (ARR), i.e. 1/ARR.

Results

From 117 citations we finally retrieve three randomized controlled trials (RCTs) totaling 50578 patients allocated to warfarin or NOA.46 Table 1 lists main features of included studies while Table 2 reports the quality assessment according to Cochrane/PRISMA statements.11,12

View this table:
Table 1

Main clinical features of patients treated with NOA

Dabigatran 110 mg/bidDabigatran 150 mg/bidRivaroxaban 20 mg/dieaApixaban 5 mg/bid
Age (year)71.4 ± 8.671.5 ± 8.873 (65–78)70 (63–76)
Male sex (%)64.363.239.735.5
Type of AF (%)
    Persistent/permanent67.867.481.184.9
    Paroxysmal/new onset32.132.618.915.1
CHADS2-score (%)
    0–132.632.2034
    234.735.21335.8
    3–632.732.68730.2
Previous stroke/TIA/SE (%)19.920.354.919.2
Previous MI (%)16.816.916.614.5
Diabetes (%)23.423.140.425
Previous use of VKA (%)50.150.262.357.1
Previous use of Aspirin (%)4038.736.3731.3
  • MI: myocardial infarction; TIA: transient ischemic attack; VKA: vitamin K antagonists. aIn patients with creatinine clearance of 30–49 ml/min, the dose of Rivaroxaban was 15 mg/die.

View this table:
Table 2

Risk of bias assessment of included studies

Adequate sequence generation?Allocation concealment used?Blinding?Concurrent therapies similar?Incomplete outcome data addressed?Uniform and explicit outcome definitions?Free of selective outcome reporting?Free of other bias?Overall risk of bias?
ARISTOTLEYes (computed generated sequence)Yes (interactive voice response system)Double blind, double dummyYes (differences <1%)YesYesYesYesLow
RE-LYYes (computed generated sequence)Yes (interactive voice response system)Double blindYes (differences <1%)YesYesYesYesLow
ROCKET-AFYes (computed generated sequence)Yes (interactive voice response system)Double blind, double dummyYes (differences <1%)YesYesYesYesLow

Pair-wise overall comparison of NOA vs. warfarin

As for efficacy end-points, overall meta-analytic pooling showed that NOA were comparable to warfarin according to the cumulative risk of TES and SE [OR 0.92 (0.83, 1.02) P = 0.1] as well as for TES alone [OR 0.93 (0.83, 1.05], P = 0.24].

NOA were associated with a reduced rate of SE [OR 0.64 (0.44, 0.94], P = 0.02], although this analysis is affected by moderate inconsistency, I2 50%, (Figure 1). Exact number of events/patients is shown in the figures.

Figure 1.

Overall meta-analytic comparison: efficacy endpoints. Single study fixed-effect odds ratios and 95% CI are shown by squares and horizontal lines. Overall and group pooled odds ratios with 95% CI are shown by diamonds. SE: systemic embolism, TES: thromboembolic stroke, VKA: vitamin K antagonist.

As for safety end-points, NOA showed large superiority in terms of HS [OR 0.43 (0.34, 0.55), P < 0.001, NNT to save one HS 153] but not in terms of EMB [OR 0.98 (0.91, 1.07), P = 0.69] (Figure 2).

Figure 2.

Overall meta-analytic comparison: safety endpoints. Single study fixed-effect odds ratios and 95% CI are shown by squares and horizontal lines. Overall and group pooled odds ratios with 95% CI are shown by diamonds. EMB: extracranial major bleeding; HS: hemorrhagic stroke. VKA: vitamin K antagonist.

NOA were superior to warfarin for the risk of all cause death [OR 0.90 [0.84, 0.96], P = 0.03, NNT to save one fatality 43] while being comparable for the risk of MI (Figure 3).

Figure 3.

Overall meta-analytic comparison for the risk of all cause death and MI. Single study fixed-effect odds ratios and 95% CI are shown by squares and horizontal lines. Overall and group pooled odds ratios with 95% CI are shown by diamonds. VKA: vitamin K antagonist.

Analyses of EMB and MI showed significant inconsistency, I2 73%, thus a further computation has been performed applying random effect model in order to reduce the potential impact of bias. On the other hand, results obtained with both fixed effect model and random effect model were quite consistent: OR with Random effect model for EMB: 0.98 (0.84–1.14), P = 0.82; OR with random effect model for MI: 1.04 (0.8–1.42), P = 0.67.

Adjusted indirect meta-analysis of NOA

Head to head comparisons for efficacy end points (Table 3) showed that in terms of cumulative rate of TES/SE, as well as of TES, none of the NOA was significantly superior to the others. Rivaroxaban showed superiority in the prevention of SE compared to all the other NOA.

View this table:
Table 3

Head to head adjusted indirect comparison of NOA

ORZ-scoreP
Apixaban vs. Dabigatran 110 mg
    TES/SE0.8 (0.6–1.1)1NS
    TES0.82 (0.6–1.05)1.5NS
    SE1.1 (0.39–3.1)0.2NS
    HS1.6 (0.8–3.3)1.3NS
    EMB0.8 (0.6–1.05)1.4NS
    All cause death1.03(0.8–1.23)0.3NS
    MI0.6 (0.4–0.9)1.990.04
Apixaban vs. Dabigatran 150 mg
    TES/SE1.16 (0.85–1.59)0.95NS
    TES1.19 (0.8–1.6)1.05NS
    SE1.03 (0.3–2.9)0.06NS
    HS1.9 (0.9–4.1)1.780.06
    EMB0.7 (0.59–0.92)2.60.03
    All cause death1.06 (0.88–1.28)0.7NS
    MI0.6 (0.4–0.96)2.10.04
Apixaban vs. Rivaroxaban
    TES/SE1 (0.7–1.3)1NS
    TES0.9 (0.7–1.3)0.6NS
    SE3.8 (1.1–1.59)2.20.03
    HS0.8 (0.4–1.59)0.4NS
    EMB0.7 (0.56–0.88)30.001
    All cause death1.13 (0.9–1.4)1NS
    MI1.1 (0.7–1.6)0.4NS
Dabigatran 110 mg vs. Rivaroxaban
    TES/SE1.18 (0.89–1.57)1.18NS
    TES1.2 (0.87–1.66)1.13NS
    SE3.4 (1–11.7)1.960.05
    HS0.5 (0.2–1.13)1.6NS
    EMB0.8 (0.67–1.05)1.5NS
    All cause death1.09 (0.8–1.37)0.79NS
    MI1.7 (1.12-2.6)2.50.004
Dabigatran 150 mg vs. Rivaroxaban
    TES/SE0.85 (0.65–1.11)1.1NS
    TES0.82 (0.59–1.15)1.09NS
    SE3.6 (1.08–12.5)20.04
    HS0.4 (0.2–0.9)20.04
    EMB0.9 (0.76–1.19)0.3NS
    All cause death1.06 (0.8–1.3)0.4NS
    MI1.76 (1.1–2.6)2.60.004
  • EMB: extracranial major bleeding; HS: hemorrhagic stroke; MI: myocardial infarction; OR: odds ratio; SE: systemic embolism; TES: thromboembolic stroke.

Head to head comparisons for safety end points showed that Dabigatran 150 mg/twice daily was associated with a further reduced risk of HS compared to Rivaroxaban. On the other hand Apixaban showed a reduced risk of EMB compared to Dabigatran 150 mg/twice daily as well as to Rivaroxaban.

Overall mortality was quite comparable across NOA.

Risk of MI was higher with both doses of Dabigatran compared to other NOA.

Discussion

Robust evidences of efficacy entitled warfarin and other vitamin K antagonists as the drugs of choice to prevent thromboembolic events in patients with AF since decades,1 despite the well known limitations. The latter drove the need for valid alternative drugs which should ideally be at least as effective as warfarin but also easier to manage.

Recent publication of RE-LY,4 ARISTOTLE5 and ROCKET-AF6 trials built up the basis for an epochal change.

These trials share somewhat similar conclusions.

Primary efficacy end-point at 1 year was consistently the cumulative rate of any stroke and SE and all the trials reached statistically significance for the primary hypothesis. Specifically, RE-LY and ARISTOTLE trials showed ‘superiority’ while ROCKET-AF reached the expected ‘non-inferiority’ compared to warfarin. These results were substantially conditioned by a large and consistent reduction of HS rate, although inclusion of HS in the efficacy evaluation could be questioned considering that HS is a side-effect of anticoagulants, perhaps an ominous complication. Notably, only Dabigatran 150 mg/twice daily significantly reduced the risk of TES compared to warfarin.

The aforementioned trials however differ in several ways, thus interpretation must use much caution.

Dabigatran is a direct Thrombin inhibitor that is administered twice daily, while apixaban and rivaroxaban are direct Factor Xa inhibitors, the former administered twice daily, the latter once daily.

In the RE-LY trial, allocation to either drugs was not concealed, while in ARISTOTLE and ROCKET-AF a double blind fashion was achieved. Study populations were also different as in the ROCKET-AF trial patients with 0–1 CHADS score were not included (Table 1).

Notably, although ROCKET-AF trial enrolled a higher risk population, mean percentage of time in which INR was in the therapeutic range was the lowest, i.e. 55% vs. 62% in the ARISTOTLE trial and 65% in the RE-LY trial. Time into therapeutic range is strongly associated with the risk of adverse events.14

Dabigatran and Rivaroxaban has already received approval for the prevention of thromboembolic events in patients with AF.1517

In a recent analysis, despite the much higher cost of the drug itself, Dabigatran showed even cost-effectiveness compared to warfarin when accounting for costs of regular monitoring.18 On the other hand, generic warfarin is expected to be far less expensive.

Practitioners and patients will be soon facing the question of which is the right medication, perhaps without any direct comparison (see Table 4 for Pharmacokinetics of the novel oral anticoagulants and use in specific metabolic conditions).

View this table:
Table 4

Pharmacokinetics of the novel oral anticoagulants and use in specific metabolic conditions

DabigatranApixabanRivaroxaban
TargetFactor IIaFactor XaFactor Xa
Dose for AF75–150 mg/twice daily5 mg/twice daily20 mg/daily
Effect of foodMay delay (but not limit) absorptionNoneNone
T1/212–17 h12 h5–9 h
Tmax1 h1–3 h2–4 h
MetabolismActivation by esterases—renalHepatic (CYP3A4—major)Hepatic (CYP3A4—major)—renal
Renal impairmentUse 75 mg twice daily if CrCl = 15–30 ml/min*m2Use 2.5 mg twice daily if SCr ≥ 1.5 mg/dlUse 15 mg daily if CrCl = 30–49 ml/min*m2—avoid use if CrCl <30 ml/min*m2
Hepatic impairmentN/AAvoid if severe (Child-Pugh C)Avoid use in moderate or severe (Child-Pugh B-C)
Drug interactionsP-glycoprotein inhibitors or inducersCYP3A4 inhibitors or inducers; P-glycoprotein inhibitors or inducersCYP3A4 inhibitors or inducers; P-glycoprotein inhibitors or inducers
Overdose managementUnknown (can be dialyzed)UnknownUnknown (likely not dialyzable—possibly protrombin complex concentrate
  • CrCl = creatinine clearance; CY = cytochrome; N/A = not applicable or available; SCr = serum creatinine; T1/2: half-life, TMAX: assumption to peak plasmatic concentration time.

From a practical point of view, it is conceivable that the individual patient in which INR value has been adequate for long time does not really need to switch to NOA.

This consideration is corroborated by data from a recent meta-analysis showing that the annual rate of stroke in patients treated with warfarin has decreased from 2.09% in earlier studies to 1.66% in RCTs performed in the last 6 years, as a result of an improved quality of anticoagulation and a greater proportion of time spent in therapeutic range.19

In our overall analysis NOA were equivalent to warfarin according to the cumulative risk of TES and SE as well as for TES alone. On the other hand, NOA were associated with 36% reduced rate of SE and an impressive 57% reduction in HS. Consistently, NOA were associated with a reduced risk of all-cause mortality.

From the adjusted comparisons across different NOA, we observed that none of them was superior to the others in terms of the cumulative rate of TES/SE as well as of TES alone. Among others, Rivaroxaban appeared particularly protective against SE.

Dabigatran 150 mg/twice daily was associated with the largest reduction of HS vs. warfarin and, consistently, it was significantly superior to Dabigatran 110 mg/twice daily and Rivaroxaban, while achieving a trend of superiority compared to Apixaban.

The latter showed the lowest risk of EMB. Mortality rate showed no significant differences across NOA.

Both in absolute terms compared to warfarin and according to head to head adjusted comparisons, Dabigatran 150 mg/twice daily could be reasonably entitled as having the best risk/benefit profile as the higher risk of MI initially observed in the first publication of the RE-LY trial did not offset the significantly lower risk of all-cause mortality compared to warfarin. Moreover, a detailed post hoc analysis of the RE-LY trial including the assessment of silent MI showed no significant difference between Dabigatran 150 mg/twice daily and warfarin.20

Of note, a recent meta-analysis including trials of dabigatran vs. warfarin in all the clinical settings in which Dabigatran has been tested reported an absolute risk increase of 0.27% with respect to warfarin.21

Use of NOA is not immune to risks. The short half-life of NOA, while possibly preventing excessive bleeding, requires careful management as missing just one or two doses could lead to insufficient anticoagulation. The lack of need for routine monitoring is intriguing but could prevent the adequate dose adjustments in case of bleeding or thrombosis, renal or liver insufficiency, advanced age, or treatment with concomitant interacting medications, for instance anti-platelet drugs.22

Availability of possible antidotes will also be an issue. Prothrombin complex concentrate has been shown able to immediately and completely reverse the anticoagulant effect of rivaroxaban in healthy subjects, while having no influence on the anticoagulant action of dabigatran, that is, on the other hand, dialyzable.23

Noteworthy, the near future will see further complexity as newer agents such as Edoxaban and Betrixaban, both oral Factor Xa Inhibitors, are still under investigation.24,25 The latter, as being the only new anticoagulant excreted almost unchanged through bile (renal excretion <5%), appears particularly promising for patients with advanced renal failure.

Limitations

Limitations of meta-analyses are well known. Adjusted indirect comparison meta-analysis is an established statistical technique which, together with network meta-analysis, belongs to the wider family of multiple treatment meta-analyses. Despite being usually consistent with results of direct comparisons, they are occasionally conflicting with head-to-head randomized trials.7,8 In particular, substantial differences between direct and indirect comparisons may be anticipated in the absence of comparability of patients and/or interventions, and when similarity, consistency and homogeneity assumptions are not met. This is not the case of our analysis as the only true difference across studies was the inclusion of a slightly higher risk population in the ROCKET-AF trial.

Moreover, there is evidence that, at least in selected cases, adjusted indirect comparison meta-analyses may be less biased than direct comparisons for evaluating new interventions.24

Conclusions

A randomized head to head direct comparison of NOA is unlikely to be done. Present data, while providing numbers and figures of practical utility, contribute to clarify the complex scenario that is progressively emerging from recent literature. However, literature has always a difficult translation into real world. Only the long-term use of these new agents in a real-world clinical setting will demonstrate how they compare with vitamin K antagonists in terms of efficacy, safety and cost.

Nonetheless, the new era of oral anticoagulation for stroke prevention in AF has clearly begun.

Appendix A

PubMed was searched on 30th October 2011 according to Biondi-Zoccai et al.,13 using the following highly sensitive strategy: (RCT[pt] OR controlled clinical trial[pt] OR randomized controlled trials[mh] OR random allocation[mh] OR double-blind method[mh] OR single-blind method[mh] OR clinical trial[pt] OR clinical trials[mh] OR (clinical trial[tw] OR ((singl*[tw] OR doubl*[tw] OR trebl*[tw] OR tripl*[tw]) AND (mask*[tw] OR blind [tw])) OR (latin square[tw]) OR placebos[mh] OR placebo*[tw] OR random*[tw] OR research design[mh:noexp] OR follow-up studies [mh] OR prospective studies[mh] OR cross-over studies[mh] OR control*[tw] OR prospectiv*[tw] OR volunteer*[tw]) NOT (animal[mh] NOT human[mh]) NOT (comment[pt] OR editorial[pt] OR metaanalysis[pt] OR practice-guideline[pt] OR review[pt])) AND (warfarin AND (loading OR dabigatran OR apixaban OR rivaroxaban)), retrieving a total of 117 citations.

Appendix B

  1. ln (ORdabigatran vs. apixaban) = ln (ORdabigatran vs. warfarin) − ln (ORapixaban vs. warfarin), and var [ln (ORdabigatran vs. apixaban)] = var [ln (ORdabigatran vs. warfarin)] + var [ln (ORapixaban vs. warfarin)]

  2. ln (ORdabigatran vs. rivaroxaban) = ln (ORdabigatran vs. warfarin) − ln (ORrivaroxaban vs. warfarin), and var [ln (ORdabigatran vs. rivaroxaban)] = var [ln (ORdabigatran vs. warfarin)] + var [ln (ORrivaroxaban vs. warfarin)]

  3. ln (ORapixaban vs. rivaroxaban) = ln (ORapixaban vs. warfarin) − ln (ORrivaroxaban vs. warfarin), and var [ln (ORapixaban vs. rivaroxaban)] = var [ln (ORapixaban vs. warfarin)] + var [ln (ORrivaroxaban vs. warfarin)]

Acknowledgements

We would like to thank Ms Giulia D’ Agostino for her irreplaceable support.

Conflict of interest: None declared.

References

View Abstract