Feasibility first
The most important step in an indirect comparison happens before any model is fitted: assessing whether studies are similar enough to compare. We map populations, outcome definitions, follow-up and potential effect modifiers across the network, and decide which comparisons are defensible, anchored or unanchored, and on which scale.
What we deliver
- Feasibility assessments and evidence network mapping
- Network meta-analysis, fixed and random effects, Bayesian and frequentist
- Component NMA for combination treatments
- Matching-adjusted indirect comparisons (MAIC) and simulated treatment comparisons (STC)
- Multilevel network meta-regression (ML-NMR)
- Time-to-event NMA, including fractional polynomial and piecewise approaches
- Sensitivity analyses, inconsistency checks and transparent reporting
Built for NICE and the JCA
NICE DSU Technical Support Document 18 sets out recommendations for population-adjusted methods, and the EU HTA Coordination Group has published methodological guidelines on direct and indirect comparisons for joint clinical assessment. We design analyses that satisfy both. Read our in-depth analysis: population-adjusted indirect comparisons for HTA.
Linked to your review and model
Our indirect comparisons draw on systematic reviews designed for them, and feed directly into the economic model, so the evidence chain is consistent from search to cost per QALY.
When an indirect comparison matters
Many new medicines reach HTA with a pivotal trial against placebo or against a single active comparator, while the scope lists several comparators. NICE's technology appraisal and highly specialised technologies manual (PMG36) asks for a network meta-analysis, where appropriate, when the technologies of interest have not been compared within a single randomised trial. It notes that the network should ideally contain all the technologies identified as interventions or appropriate comparators in the scope, and that NICE prefers the methods set out in the Decision Support Unit evidence synthesis series of Technical Support Documents.
The EU joint clinical assessment widens the question. Member States set out their needs through the PICO survey in the scoping process, and each PICO in the consolidated assessment scope may call for its own analysis. The HTA Coordination Group's methodological and practical guidelines on direct and indirect comparisons, both adopted on 8 March 2024, describe how those analyses are conducted and reported. National agencies such as SMC and NCPE then draw on the same comparative evidence for their own decisions, so a well-designed network can serve several submissions at once. See EU JCA dossiers and one evidence plan for NICE and the JCA.
Choosing a method: what each one needs
| Method | Data it uses | Where it fits | Central assumption |
|---|---|---|---|
| Bucher comparison or network meta-analysis | Aggregate results from randomised trials linked through common comparators | Connected networks of broadly similar trials | Similarity, homogeneity and consistency of relative effects across the network |
| Anchored MAIC | Individual patient data (IPD) for one trial, published aggregate data for the other, and a shared comparator | Two trials with imbalanced effect modifiers; the estimate applies to the population of the aggregate-data trial | Conditional constancy of relative effects: all effect modifiers are reported for the aggregate-data trial and adjusted for |
| Anchored STC | As for MAIC, using an outcome regression model | As for MAIC, where a regression approach suits the outcome | The outcome model is correctly specified for the effect modifiers |
| ML-NMR | IPD from at least one trial combined with aggregate data across a network | Larger connected networks with imbalanced effect modifiers; estimates can be produced for a chosen target population | A shared model for how effect modifiers act across the network |
| Unanchored comparison | Single-arm or disconnected evidence | Where no shared comparator exists | Treated as non-randomised evidence; relies on conditional constancy of absolute effects, so all prognostic factors and effect modifiers need adjusting for |
EU guidance puts anchored comparisons first: the HTA Coordination Group guideline states that anchored indirect comparisons preserve randomisation and should always be used in preference to unanchored methods, and that unanchored comparisons call for methods for non-randomised evidence. PMG36 adds that population adjustment needs individual patient data from at least one trial in the comparison or network, and refers to NICE DSU TSD 18, which notes that unanchored comparisons need adjustment for all effect modifiers and prognostic variables. Our analysis of population-adjusted indirect comparisons sets out how these choices play out in practice.
What a strong indirect comparison looks like
Assessors value comparisons they can follow, reproduce and test. In our experience, the strongest analyses share a few qualities:
- A documented feasibility assessment: The HTA Coordination Group frames all evidence synthesis around exchangeability, examined through similarity, homogeneity and, for indirect comparisons, consistency. A clear record of how populations, definitions and follow-up compare makes that case visible.
- A network that matches the scope: Trials that connect relevant treatments are included even when they also contain other arms, as PMG36 describes.
- Trial-level transparency: A table of baseline characteristics for each included trial, and sensitivity analyses that include or exclude trials whose relevance is uncertain.
- Inconsistency explored and explained: Where direct and indirect evidence differ, the analysis sets out possible reasons and how they were investigated.
- Analyses planned in advance: A base case and sensitivity analyses defined before results are seen are easier for every reader to trust.
- Code and outputs that can be rerun: Results that an external assessor can reproduce and adapt to alternative assumptions.