On 2014 Mar 15, Valeria Fadda commented:

Meta-analysis on novel oral anticoagulants vs warfarin for stroke prevention: re-expressing the results as risk difference

In meta-analysis, the risk ratio (RR) and the risk difference (RD) are two commonly used outcome measures. Each of these indexes has its own advantages and disadvantages [1].

The meta-analysis by Ruff et al. [2] adopted the outcome measure of RR. In particular, the values of pooled RR were estimated according to the random-effect model.

We have re-analysed the results reported in Figure 1 of Ruff et al. [2] by expressing the results (event =stroke or systemic embolism) as RD instead of RR. In our re-analysis, the meta-analysis model was random-effect as implemented in the OMA software [3].

Our results were the following (see Ruff et al.[2] for further details on these data sets): RE-LY (dabigatran vs warfarin): RD= -1.10% (95% confidence interval[CI]: -1.68% to -0.52%); ROCKET AF (rivaroxaban vs warfarin): RD= -0.52% (95%CI: -1.17% to 0.13%); ARISTOTLE (apixaban vs warfarin): RD = -0.59% (95%CI: -1.06% to -0.13%); ENGAGE AF-TIMI48 (edoxaban vs warfarin): RD = -0.58% (95%CI: -1.27% to 0.10%); overall (all new oral anticoagulants vs warfarin): RD =-0.70% (95%CI: -0.99% to -0.41%). Expressing this incremental benefit as RD is advantageous because this analysis provides an absolute estimate of the magnitude of the incremental improvement. This meta-analytical result (RD = -0.70%; 95%CI: -0.99% to -0.41%) is statistically significant, but has a small magnitude.

To better interpret this absolute incremental benefit (i.e. a meta-analytical reduction of 0.70%) estimated from the overall analysis of these trials, a useful reference is given by the pre-specified non-inferiority margin that was used in the statistical power calculations of the pivotal trials comparing dabigatran, rivaroxaban, or apixaban with warfarin (RE-LY, ROCKET AF, and ARISTOTLE, respectively). These trials adopted a non-inferiority margin equal to half of the incremental benefit between warfarin and placebo (6 warfarin-vs-placebo trials) found in the meta-analysis by Hart et al. [4].

As pointed out by Fadda et al. [5], while Hart and co-workers considered, as their outcome measure, only RR and not RD, a meta-analysis of the same trials based on RD can be helpful to interpret these findings. The result of this re-analysis is represented by a pooled RD for warfarin vs. placebo of -5.0% (95%CI: -7.5% to -2.4%; random-effect model). Halving this difference yields a margin, expressed as RD, of -2.5%

References

1. Walter SD. Choice of effect measure for epidemiological data. J Clin Epidemiol. 2000 Sep;53(9):931-9.

2. Ruff CT, Giugliano RP, Braunwald E, Hoffman EB, Deenadayalu N, Ezekowitz MD, Camm AJ, Weitz JI, Lewis BS, Parkhomenko A, Yamashita T, Antman EM. Comparison of the efficacy and safety of new oral anticoagulants with warfarin in patients with atrial fibrillation: a meta-analysis of randomised trials. Lancet. 2013 Dec 3. pii: S0140-6736(13)62343-0. doi: 10.1016/S0140-6736(13)62343-0. [Epub ahead of print]

3. OMA software (Open Meta-Analyst version 4.16.12, Tufts University, U.S., url http://tuftscaes.org/open_meta/).

4. Hart RG, Benavente O, McBride R, Pearce LA. Antithrombotic therapy to prevent stroke in patients with atrial fibrillation: a meta-analysis. Ann Intern Med. 1999 Oct 5;131(7):492-501.

5. Fadda V, Gatto R, Maratea D. Incremental benefit of warfarin over placebo in patients with atrial fibrillation (Comment), PubMed Commons, available at http://www.ncbi.nlm.nih.gov/pubmed/10507957#cm10507957_2944

On 2014 Mar 15, Valeria Fadda commented:

Meta-analysis on novel oral anticoagulants vs warfarin for stroke prevention: re-expressing the results as risk difference

In meta-analysis, the risk ratio (RR) and the risk difference (RD) are two commonly used outcome measures. Each of these indexes has its own advantages and disadvantages [1].

The meta-analysis by Ruff et al. [2] adopted the outcome measure of RR. In particular, the values of pooled RR were estimated according to the random-effect model.

We have re-analysed the results reported in Figure 1 of Ruff et al. [2] by expressing the results (event =stroke or systemic embolism) as RD instead of RR. In our re-analysis, the meta-analysis model was random-effect as implemented in the OMA software [3].

Our results were the following (see Ruff et al.[2] for further details on these data sets): RE-LY (dabigatran vs warfarin): RD= -1.10% (95% confidence interval[CI]: -1.68% to -0.52%); ROCKET AF (rivaroxaban vs warfarin): RD= -0.52% (95%CI: -1.17% to 0.13%); ARISTOTLE (apixaban vs warfarin): RD = -0.59% (95%CI: -1.06% to -0.13%); ENGAGE AF-TIMI48 (edoxaban vs warfarin): RD = -0.58% (95%CI: -1.27% to 0.10%); overall (all new oral anticoagulants vs warfarin): RD =-0.70% (95%CI: -0.99% to -0.41%). Expressing this incremental benefit as RD is advantageous because this analysis provides an absolute estimate of the magnitude of the incremental improvement. This meta-analytical result (RD = -0.70%; 95%CI: -0.99% to -0.41%) is statistically significant, but has a small magnitude.

To better interpret this absolute incremental benefit (i.e. a meta-analytical reduction of 0.70%) estimated from the overall analysis of these trials, a useful reference is given by the pre-specified non-inferiority margin that was used in the statistical power calculations of the pivotal trials comparing dabigatran, rivaroxaban, or apixaban with warfarin (RE-LY, ROCKET AF, and ARISTOTLE, respectively). These trials adopted a non-inferiority margin equal to half of the incremental benefit between warfarin and placebo (6 warfarin-vs-placebo trials) found in the meta-analysis by Hart et al. [4].

As pointed out by Fadda et al. [5], while Hart and co-workers considered, as their outcome measure, only RR and not RD, a meta-analysis of the same trials based on RD can be helpful to interpret these findings. The result of this re-analysis is represented by a pooled RD for warfarin vs. placebo of -5.0% (95%CI: -7.5% to -2.4%; random-effect model). Halving this difference yields a margin, expressed as RD, of -2.5%

References

1. Walter SD. Choice of effect measure for epidemiological data. J Clin Epidemiol. 2000 Sep;53(9):931-9.

2. Ruff CT, Giugliano RP, Braunwald E, Hoffman EB, Deenadayalu N, Ezekowitz MD, Camm AJ, Weitz JI, Lewis BS, Parkhomenko A, Yamashita T, Antman EM. Comparison of the efficacy and safety of new oral anticoagulants with warfarin in patients with atrial fibrillation: a meta-analysis of randomised trials. Lancet. 2013 Dec 3. pii: S0140-6736(13)62343-0. doi: 10.1016/S0140-6736(13)62343-0. [Epub ahead of print]

3. OMA software (Open Meta-Analyst version 4.16.12, Tufts University, U.S., url http://tuftscaes.org/open_meta/).

4. Hart RG, Benavente O, McBride R, Pearce LA. Antithrombotic therapy to prevent stroke in patients with atrial fibrillation: a meta-analysis. Ann Intern Med. 1999 Oct 5;131(7):492-501.

5. Fadda V, Gatto R, Maratea D. Incremental benefit of warfarin over placebo in patients with atrial fibrillation (Comment), PubMed Commons, available at http://www.ncbi.nlm.nih.gov/pubmed/10507957#cm10507957_2944