← allmeta

Workflow · PRISMA-NMA

Network meta-analysis

From a connected network of trials to a defensible ranking. The order matters: check transitivity and inconsistency before reporting a SUCRA. Skip those checks and your ranking is fiction.

  1. Frame & search (as for any systematic review)

    The PICO must be specific about treatments and outcomes. Comparators must form a connected network — at least one trial per loop closure.

    If the network has disconnected components, NMA is impossible — fall back to pairwise.

    PICO Formulator → Search Translator →

  2. Risk of bias & transitivity

    RoB 2 per study. Then assess transitivity: are effect modifiers (age, baseline severity, follow-up) similar across comparisons? If not, indirect comparison is invalid.

    Salanti G. Stat Methods Med Res 2008;17:279 and Res Synth Methods 2012;3:80–97 — transitivity is the central NMA assumption. A non-transitive network produces precise but biased pooled effects.

    RoB 2 → Traffic-light plot → Learn the theory: Advanced MA course (NMA module)

  3. Pairwise checks first (Bucher)

    For each closed loop, run a Bucher indirect comparison and check it agrees with the direct estimate. Big disagreements signal inconsistency before you ever fit the network model.

    Cheap diagnostic. Catches problems that the global model averages out.

    Bucher Indirect →

  4. Fit the network & visualise

    Frequentist (graph-theoretic, contrast-level random effects) or Bayesian. Either way, draw the network with edge widths proportional to the number of trials per comparison.

    A pretty SUCRA hides a lot. The network plot is the first thing a methodological reviewer looks at.

    NMA → Bayesian NMA → Component NMA → WebR (netmeta) →

  5. Test inconsistency — globally and locally

    Design-by-treatment interaction test for global inconsistency. Node-splitting for each loop. If either is significant, the ranking is suspect — investigate which design or comparison is the source.

    Per Cochrane Handbook §11.4: "Always test consistency before interpreting." Skip this and your ranking can't be defended.

    Global inconsistency → Node-splitting →

  6. Certainty & report (CINeMA + PRISMA-NMA)

    Use CINeMA to grade certainty across the 6 NMA-specific domains (within-study bias, reporting, indirectness, imprecision, heterogeneity, incoherence). Report SUCRA only alongside the certainty rating.

    Nikolakopoulou A et al. PLoS Med 2020;17:e1003082 (CINeMA framework). A treatment can rank #1 with very low certainty — readers must see both. The SUCRA-without-certainty figure has been a recurring red-flag in NMA peer review.

    CINeMA → GRADE SoF →

New to NMA? Start with the Advanced Meta-Analysis course (NMA module). For pairwise reviews, see systematic review; for diagnostic accuracy, see DTA.