13 Research methods, statistics, critical review and evidence-based practice13.4 Systematic reviews
13.4.6 Network meta-analysis
2 min read1 figure
Written by Resident Doctor Committee · Published
New in the 2027 syllabus
The College added this topic for exams from January 2027.
Comparing several treatments
A network meta-analysis compares several treatments at once. It combines direct and indirect comparisons across a connected set of interventions. From these, it estimates the relative effect of each treatment against the others.
Direct evidence compares two treatments within the same randomised trials.
Indirect evidence links two treatments through a shared comparator, such as placebo.
In the network diagram, each node is a treatment and each edge (line) is a direct comparison in the included studies. Check the legend to see what line thickness and node size mean.
Assumptions
Concept | Meaning | Appraisal question |
|---|---|---|
Transitivity | The comparisons are similar enough in their effect modifiers | Could participants plausibly have been randomised to any treatment being compared? |
Coherence, or consistency | Direct and indirect estimates agree, allowing for sampling uncertainty | Do different routes through the network give compatible effects? |
Heterogeneity | Effects vary within one particular comparison | How similar are the trials comparing the same pair? |
Effect modifiers include baseline severity, previous treatment and follow-up. If they are spread unevenly across comparisons, indirect estimates can be misleading.
Disconnected networks cannot give every relative effect without further assumptions to link the separate parts.
A non-significant inconsistency test does not prove coherence, especially when the network is sparse.
Illustrative indirect comparison
Assume the trials are comparable, use the same outcome and follow-up, and express risk ratios in the stated direction.
Comparison | Risk ratio for relapse |
|---|---|
A versus placebo | 0.60 |
B versus placebo | 0.80 |
Use the same reference treatment and outcome direction in both comparisons.
Indirect risk ratio for A versus B = 0.60 / 0.80 = 0.75.
This suggests a lower relative risk of relapse with A, but only if transitivity holds. The uncertainty must include the uncertainty from both estimates.
Reading rankings
A ranking shows the relative order for one particular outcome. Coming first does not mean a treatment has a clinically worthwhile advantage.
Rankings can be very uncertain when the evidence is sparse, imprecise or biased.
Weigh benefits, harms and certainty together. Use effect estimates and their intervals, not the rank alone.
Multi-arm trials produce comparisons that are correlated with each other. The analysis must allow for this, rather than counting their shared control group repeatedly as independent evidence.