Bucher inconsistency estimate and z statistic for a single loop

Compares a direct estimate of the effect of C relative to B with the indirect estimate formed from the A versus B and A versus C evidence, the inconsistency factor IF in the article. The inconsistency is their difference, its variance is the sum of the two variances, and z is referred to the standard normal distribution. The function sqrt is the square root.

Signature

omega = d_BC_dir - d_BC_ind; V_omega = V_BC_dir + V_BC_ind; z = (d_BC_dir - d_BC_ind) / sqrt(V_BC_dir + V_BC_ind)
Inputs
InputsDefinitionUnit
d_BC_dirDirect estimate of the effect of C compared with B from head-to-head trialseffect scale, for example log odds ratio
d_BC_indIndirect estimate d_AC minus d_AB from the Bucher comparison on this pageeffect scale, for example log odds ratio
V_BC_dirVariance of d_BC_dirsquared units of the effect scale
V_BC_indVariance of d_BC_ind, the sum of the two direct variancessquared units of the effect scale
Output
omegaDirect minus indirect estimate of the effect of C relative to Beffect scale, for example log odds ratio
V_omegaVariance of omegasquared units of the effect scale
zInconsistency estimate divided by its standard error, referred to the standard normal distributionno unit

Function

Network consistency function

Expresses every pairwise relative effect in a connected network through effects relative to a common reference treatment A, on a scale where effects add, such as the log odds ratio. Here d_XY is the effect of Y relative to X, the NICE Decision Support Unit convention. Consistency is the assumption that direct and indirect evidence estimate the same d_XY.

Implementations

  • Excel

    Inconsistency, variance, z and two-sided p value

    With the direct and indirect estimates and variances in named cells, the four formulas return omega, its variance, z and the two-sided p value.

    =dBCdir-dBCind; =VBCdir+VBCind; =Omega/SQRT(VarOmega); =2*(1-NORM.S.DIST(ABS(Zstat),TRUE))

Assumptions

  • Three independent sources of evidence

    The three direct estimates come from separate sets of trials. Three-arm trials are excluded from the test because they are internally consistent and would dilute any disagreement.

Worked examples

  • Inconsistency in the TSD 4 HIV loop

    A direct log odds ratio of 0.47 (standard error 0.10) for C against B, compared with the indirect estimate of minus 1.37 (variance 0.4292), gives an inconsistency of 1.84 with variance 0.4392 and z of about 2.78, evidence of inconsistency with p below 0.01, as in Box 1 of TSD 4.

    d_BC_dir = 0.47; d_BC_ind = -1.37; V_BC_dir = 0.01; V_BC_ind = 0.4292; omega = 1.84; V_omega = 0.4392; z = 2.78

Common errors

  • Reading a non-significant test as proof of consistency

    TSD 4 stresses that inconsistency tests are underpowered and will often miss inconsistency that is present. A z value below 1.96 is compatible with agreement but does not establish it, and the clinical assessment of transitivity still applies.

Sources

  • Bucher method for testing inconsistency in a loop

    Dias S, Welton NJ, Sutton AJ, Caldwell DM, Lu G, Ades AE. NICE DSU Technical Support Document 4: Inconsistency in networks of evidence based on randomised controlled trials. Sheffield: Decision Support Unit, ScHARR, University of Sheffield; May 2011, last updated April 2014. Section 3.1 Bucher method for single loops of evidence (the inconsistency estimate omega as direct minus indirect, its variance and the approximate z test), section 3.2 (loops with more edges) and Box 1 worked example with the HIV virologic suppression network.

    View source →

Canonical Identity

Stable URI · Machine-readable · Resolvable · CC BY 4.0