Methodologies

Network graph methodology

How network plots represent treatments, direct comparisons, multi-arm evidence, and connectivity without implying effect magnitude or certainty.

10 min readUpdated June 15, 2026

What the graph shows

A network graph displays treatment nodes and direct comparison edges. It is an evidence-structure plot, not an effect-size plot.

Nodes represent treatment labels after harmonization. Edges represent at least one direct within-study comparison. Edge thickness or labels may reflect direct evidence volume depending on plot settings and package output.

Default assumptions

  • The graph uses the treatment labels supplied by extraction setup or NMA input.
  • Reference treatment affects model estimates and forest orientation but does not by itself determine graph connectivity.
  • The graph is generated only for requested NMA modules.
  • Disconnected or sparse graphs are methodological findings, not rendering failures.

What the graph cannot show

  • It does not prove transitivity.
  • It does not show risk of bias or certainty unless those are added as separate annotations.
  • It does not prove that treatment grouping decisions were clinically valid.
  • It does not show whether direct and indirect evidence agree.

Reporting checklist

  1. Include the graph or a direct comparison table in NMA reports.
  2. Report disconnected components or sparse edges.
  3. Explain treatment grouping and reference treatment choices.
  4. Use the graph to justify which inconsistency diagnostics are possible.