← allmeta

NMA Inconsistency

Detect inconsistency between direct and indirect evidence in network meta-analysis. Performs node-splitting (Dias et al. 2010) on every comparison with a closed loop, and a global design-by-treatment test (Higgins et al. 2012), restricted to independent two-arm studies.

Input

Try real data (one click):

Node-split per comparison

ComparisonDirectIndirectDifferencezpVerdict

Design-by-treatment (global)

Net-split forest

Reading the plot: direct and indirect whiskers should overlap for a consistent network. Non-overlapping whiskers with |z| > 1.96 indicate a specific loop with inconsistency. The global design-by-treatment test is the omnibus check.

Interpretation, sensitivity and reporting

What the results mean

Network graph

Network of treatments and direct comparisons

Leave-one-study-out sensitivity

Each study is removed in turn and the whole analysis is refitted with the selected model (random-effects τ² is re-estimated for every refit).

Change in global Q when each study is removed
Refit without each study. Select a column heading to sort, or a study to see its node-splits. Highlighted rows change a conclusion at the 0.05 level.

Fixed-effect vs random-effects

The same tests under the three models (p-values; a dash means not estimable).
QuantityFixed-effectDerSimonian-LairdPaule-MandelConclusion at 0.05

Methods and Results text

The R script refits this analysis with netmeta and metafor in native R, compares every number shown here and stops with an error on any mismatch.