The outcomes of interest were
clinically significant nonmajor bleeding
and minor bleeding events as we
defined them for this study. We based
this decision on our expectation
that studies would not be adequately
powered or of sufficient duration to
detect thromboembolic events and
on the paucity of reported major or
life-threatening bleeding after elective
dental surgery. The risk of bleeding
with the continuation of warfarin
compared with that for the discontinuation
or reduction of the dose of
warfarin was expressed as a relative
risk (RR) with an associated 95% CI.
For this analysis, an RR < 1.0 favoured
the continuation of the usual warfarin,
a RR > 1.0 favoured the discontinuation
or alteration of the warfarin dose,
whereas an RR of 1.0 indicated equivalence between the
2 groups. CIs that crossed the line of equivalence indicated
that the true RR may be 1 rather than the calculated
value. The χ2 test was used to calculate the percentage
variation across studies caused by their heterogeneity.
Significance for this test was set liberally at p ≤ 0.1, since
in practice, the test often lacks the power to detect interstudy
differences of the treatment effect.16 DerSimonian
and Laird’s random effects model of pooling17 was used
to provide a more conservative estimate of the true effect.
These analyses were carried out with Review Manager 4.2
(Cochrane Collaboration, Oxford).