Published bypubricaonOctober 8, 2026

Population-Adjusted Indirect Comparisons Are Being Reframed as a Transportability Problem The new estimand-based framework changes how MAIC, STC and anchored comparisons should be designed, reported and defended at HTA

High-Impact Journals

Interesting topics

MAIC and STC have long been treated as statistical fixes: reweight one trial until it resembles another, then report the result. A 2026 preprint by Chandler and Ishak challenges that view. Adjusting for observed differences does not guarantee the result applies to the population an HTA body must decide for [1]. The real question is transportability: can the effect be carried to the decision population, and under which assumptions?

That changes how comparisons are designed, reported and defended at HTA. This article explains what the estimand-based framework adds and what sponsors should do differently before their next submission.

1. Why MAIC and STC Are Being Reconceptualised

Matching-adjusted indirect comparison (MAIC) and simulated treatment comparison (STC) were built to answer a practical question: what would the comparison look like if the trials had enrolled similar patients? Most teams still run them that way, as a reweighting or regression step bolted onto an indirect comparison.

The 2026 preprint by Chandler and Ishak argues that this view is incomplete. Adjusting for observed differences between trials does not by itself guarantee that the result applies to the population an HTA body must decide for [1]. The question the field is now asking is one of transportability: can an effect estimated in one population be carried to another, and under which assumptions?

That shift matters for three reasons:

The target population must be chosen before the analysis, not inherited from whichever comparator trial happened to publish aggregate data.
A report that gives only a hazard ratio and a list of matched covariates no longer says enough.
Assessors at national and EU level have already shown limited patience for unanchored comparisons with weakly justified covariate sets.

This article does not explain what MAIC is. It shows what changes when you treat it as a transportability problem.

2. From Matching Technique to Transportability Problem

The European Commission published the first completed medicinal-product JCA report for tovorafenib (Ojemda) on 9 June 2026. The report provides a practical example of how the JCA process evaluates evidence against the defined assessment scope, including situations in which available clinical evidence does not directly answer every relevant PICO question.

The report describes, it does not decide. The JCA assesses relative clinical effectiveness and safety. It does not give a value judgement, an economic view or a reimbursement recommendation. Those remain national.
Scope drives everything. The assessment answers the PICO questions in the final scope, even where the pivotal trial was designed for a different question. Where the trial does not map to a PICO, the report records the evidence gap rather than closing it.
Methods are held to published guidance. European JCA methodological guidance addresses direct and indirect treatment comparisons and sets out the assumptions and considerations relevant to assessing the validity and applicability of comparative evidence. [2,3] Indirect comparisons without a common comparator require particular methodological justification because they rely on stronger assumptions than anchored comparisons. Sponsors should therefore assess their feasibility against the applicable JCA methodological guidance and the requirements of the relevant national HTA methodology rather than assume that an unanchored analysis will be accepted.

What the first report established	Why it matters to sponsors
Assessment follows the consolidated scope, not the trial protocol	Trial design choices made years earlier are tested against questions the sponsor did not write
Uncertainty is reported openly, including where evidence is missing	A gap in the JCA becomes a visible gap in every national dossier
No economic or value conclusions	National HTA bodies still run their own full assessments
Strict timelines	Late evidence generation cannot be repaired inside the process

3. What an Estimand-Based Framework Adds

An estimand states precisely what treatment effect the analysis is meant to estimate. The ICH E9(R1) addendum set out the attributes in the trial setting: the population, the treatment conditions, the outcome variable, how intercurrent events are handled, and the summary measure.

For an indirect comparison, the same discipline applies, with one addition. The analyst must also say how the estimate moves between populations [1][3].

In practice, that adds four decisions that matching-based protocols often leave implicit:

Which population. The index trial, the comparator trial, or a separately defined decision population.
Which effect measure. Marginal or conditional, and on what scale. Non-collapsible measures such as hazard ratios and odds ratios behave differently across populations even without effect modification [3].
Which transport assumption. What must hold for the estimate to apply outside the population where it was estimated.
Which data. What patient-level and aggregate information can support that assumption.

Table 2: A Stepwise Decision Framework for Selecting and Applying Population-Adjustment Methods

Question	If the answer is unclear
Who is the HTA decision population?	Define it with the clinical and HTA teams before touching data
Is the evidence anchored or unanchored?	Establish the network structure first
Which effect measure suits that population?	Choose it before choosing the method
Can the method estimate it in that population?	Change the method or narrow the claim
Which assumptions does transport need?	List them and test each as far as the data allow

4. Target Population Definition: The First Design Decision

The target population is the first decision and the one most often skipped.

In a typical MAIC, the weighted individual patient data are reshaped to match the comparator trial’s published baseline means. The target is whoever was enrolled in that trial. That may or may not be the population a reimbursement body is deciding for.

The transportability view asks the sponsor to name the target first. Options include:

the comparator trial population, which is simple but may be unrepresentative;
the index trial population, which sponsors often prefer but which may not be what the HTA body wants;
a defined decision population, for example patients eligible under the proposed label or treated in a particular health system.

Each choice produces a different answer. In the anchored setting, methods that estimate effects in any chosen target population, such as multilevel network meta-regression, were developed for exactly this purpose [4]. Pairwise weighing methods are mostly limited to the comparator population, except in special cases where the marginal effect can be transported directly [1].

A sound protocol records the target population, the reason for choosing it, and what the HTA body is likely to expect. If the sponsor’s preferred population differs from the HTA body’s, say so in the dossier.

5. Effect Modifiers and Missing Covariates Under the New Framework

Every population-adjusted method rests on the same assumption: the variables that modify the treatment effect, and in unanchored settings the prognostic variables too, have all been identified and adjusted for.

The framework adds three points to that familiar requirement.

Covariates that are not obvious modifiers can still matter. Remiro-Azócar showed that on the scale used for non-collapsible measures, variables not normally described as effect modifiers can modify the population-level effect [3]. A covariate list built only from clinical judgement can therefore miss variables the estimand makes relevant.
Anchored comparisons call for adjusting all effect modifiers. TSD 18 recommends adjusting for all effect modifiers whether they appear imbalanced.
Missing covariates are a transportability limitation, not just a power problem. If the comparator publication does not report a key variable, the assumption cannot be tested, and the estimate cannot be defended as applying to the target.

A practical covariate protocol should:

pre-specify the effect modifier and prognostic factor list from literature and clinical input, before seeing outcome data;
document which variables are unavailable in the comparator and what that implies;
State how the unmeasured variables could shift the result and in which direction.

6. Anchored vs Unanchored Comparisons: Different Estimands

Anchored and unanchored comparisons do not answer the same question, and the framework keeps them apart.

An anchored comparison uses a common comparator arm. Randomisation within each trial protects the relative effect, so the analysis needs to deal with effect modification. An unanchored comparison, typically built on single-arm evidence, has no such protection. Differences in outcomes between studies could reflect the treatment or just differences in patients, so the analysis needs to account for both effect modifiers and prognostic factors. This is a far stronger assumption.

Table3: Comparison of Anchored and Unanchored Population-Adjustment Approaches

Feature	Anchored	Unanchored
Typical evidence	Trials sharing a common comparator	Single-arm or disconnected studies
What must be adjusted?	Effect modifiers	Effect modifiers and prognostic factors
Assumption strength	Strong	Much stronger
Typical methods	MAIC, STC, ML-NMR [4][6][7]	Unanchored MAIC, STC, ML-UMR [2][6][7]
HTACG stance	Acceptable if requirements are met	Regarded as low-certainty evidence

Unanchored comparisons are subject to particularly strong scrutiny because they rely on substantially stronger assumptions [1]. A design that does not say which kind of comparison it is, and what that implies for the estimand, is incomplete.

7. ML-UMR: A New Method for Unanchored Comparisons

Chandler and Ishak proposed multilevel unanchored meta-regression (ML-UMR) as an extension of multilevel network meta-regression to the unanchored setting. It is a Bayesian outcome-regression framework that combines individual patient data with aggregate data from fully disconnected evidence. The authors describe it as a generalisation of unanchored STC [2].

What it offers:

It can estimate relative effects in a pre-specified target population, not only the comparator population [1].
It can handle two or more treatments, so it is not limited to pairwise comparisons.
It makes its central assumption explicit: the shared prognostic factor assumption, which requires prognostic factors to act the same way across treatments.
The authors describe a way to relax that assumption using subgroup aggregate data, when comparators report it [2].

What to keep in mind:

The 2026 preprint is a methods proposal supported by simulation and has not completed peer review at the time of writing [2].
It remains subject to the strong limitations of any unanchored comparison, and the data needed to relax its main assumption are often not reported [1].
No HTA body has formally endorsed it. For now, it is best used as a sensitivity or supportive analysis next to a conventional approach, with the choice justified in the protocol.

8. How the Reframing Changes What You Report

A matching-based report usually shows the covariate list, the weights, the effective sample size and the final hazard ratio. An estimand-based report keeps all of that and adds the reasoning that connects it to the decision.

Table 4: Reporting Elements for Population Adjustment Under Matching and an Estimand Framework

Report element	Under matching	Added under the estimand framework
Objective	Compare drug A with drug B	State the estimand: population, measure, handling of events
Population	Comparator trial inclusion criteria	Target population and why it was chosen
Covariates	List of matched variables	Basis for selection; those missing; how they might bias the result
Method	MAIC or STC and software	Why the method can estimate this estimand in this population
Result	Adjusted effect and confidence interval	Population in which the effect applies
Limitations	General caveats	Transport assumptions and what would violate them

The test for a good report is whether an assessor can answer three questions from the document alone: which population does this apply to, what assumptions carry it there, and what would make it wrong.

9. Sensitivity Analyses Under the Transportability Framework

Sensitivity analysis in the matching tradition often means trying a different covariate set. Under the new framework, it should test the assumptions that carry the transport.

Useful analyses include:

Covariate set variation. Re-run with larger and smaller sets of effect modifiers and prognostic factors, and report how the estimate moves.
Alternative target populations. Estimate the effect in more than one defined population where the method allows and compare.
Alternative effect measures. Compare marginal and conditional estimates and different scales, especially where non-collapsibility is a concern [3].
Assumption relaxation. In unanchored work, relax the shared prognostic factor assumption where subgroup data allow [2].
Quantitative bias analysis. Estimate how strong an unmeasured variable would need to be to change the conclusion.
Overlap checks. Report effective sample size and weight distributions, and flag cases where the populations barely overlap [5].

Each analysis should have a reason in the protocol. Ten sensitivity runs with no stated purpose gives assessors more to criticise, not less.

10. Current HTA Body Positions: NICE, IQWiG, EUnetHTA

There is a gap between what methodologists now recommend and what HTA bodies accept. Closing it is the sponsor’s job.

Table 5. HTA Agency Positions on Population-Adjusted Treatment Comparisons

Body	Position
NICE	TSD 18 sets out MAIC and STC, their assumptions and when population adjustment is justified. It was not an endorsement of MAIC for routine use.
IQWiG	Methods favour randomised evidence and adjusted indirect comparisons. Comparisons without randomised anchoring are treated with strong caution.
HTACG (EU JCA)	Anchored comparisons are acceptable if methodological requirements are met. Population-adjusted and unanchored methods are accepted only conditionally and viewed critically.

The first JCA reports show how this plays out. A commentary on the first report describes assessor criticism of an unanchored MAIC, centred on how covariates were selected, and notes that HTACG regards unanchored MAIC as one of the weakest forms of relative efficacy evidence.

Sponsors should check the current wording of each body’s guidance before submission. Positions are changing and this table gives an outline only.

The transportability framework is not yet part of any of these documents. Its value is as a structure for answering the criticisms assessors already make: unclear target population, weak covariate justification and untested assumptions.

11. Pubrica MAIC/STC Design Checklist (Estimand-Based)

The checklist below turns the framework into questions a team can answer before locking a protocol.

Table 6: Quality-Control Checklist for Population-Adjusted Treatment Comparisons

Area	Check
Estimand	Population, treatment, outcome, intercurrent events and summary measure are stated
Estimand	Marginal or conditional effect is chosen and justified
Target population	Defined before analysis, with the reason and the likely HTA expectation
Evidence structure	Anchored or unanchored is stated, with the consequences for assumptions
Effect modifiers	Pre-specified list with literature or clinical basis
Prognostic factors	Included for unanchored comparisons
Missing covariates	Listed, with their likely direction of bias
Method fit	The method can estimate the chosen estimand in the chosen population
Overlap	Effective sample size and weight diagnostics reported
Sensitivity	Covariate sets, populations, effect measures and assumptions tested
Reporting	Target population, assumptions and limits stated in plain language
HTA alignment	Wording checked against NICE, IQWiG and HTACG guidance

This checklist is Pubrica’s own working tool, derived from the published framework and not an accepted standard. Run it at protocol stage, not after results are in.

12. How Pubrica Would Design and Defend a Population-Adjusted Comparison

A design review with Pubrica follows four steps.

Define the decision. We confirm the HTA body, the population it will assess and the comparators in scope.
Build the estimand. We write the estimand in full and agree on the target population with your clinical and HTA teams.
Test the assumptions. We check covariate availability, overlap and the strength of the transport assumptions, and recommend a primary method with supportive analyses.
Prepare the defence. We draft the protocol, the report and the response to likely assessor questions in language that matches the guidance.

If your protocol still frames the analysis as matching, a design review is the quickest way to find where the estimand is underspecified before an assessor does.

Get expert support from Pubrica for addressing reviewer comments, preparing point-by-point rebuttals, and strengthening manuscripts for journal resubmission. Prepare your revised manuscript with confidence.

Conclusion

What is the transportability problem in population-adjusted indirect comparisons?

Transportability asks whether a treatment effect estimated in one study population can be validly applied to the population relevant to the HTA decision. The framework shifts the focus from simply balancing trial populations to defining the target population, effect measure, and assumptions required to carry the estimate to that population.

Why is the target population important in MAIC and STC?

The target population should be defined before the analysis because the comparator trial population may not represent the population an HTA body is assessing. It may be the comparator trial population, the index trial population, or a separately defined decision population.

What is the difference between anchored and unanchored indirect comparisons?

An anchored comparison uses a common comparator and primarily requires adjustment for treatment-effect modifiers. An unanchored comparison does not have this protection and requires consideration of both effect modifiers and prognostic factors, making its assumptions substantially stronger.

What additional information should be reported in an estimand-based analysis?

An estimand-based report should specify the target population, effect measure, assumptions used for transportability, selection and availability of covariates, and the limitations of applying the estimate to the target population. This allows assessors to understand where the treatment effect applies and what assumptions could invalidate the result.

How should sponsors prepare population-adjusted indirect comparisons for HTA submissions?

Sponsors should define the decision population and estimand before analysis, identify effect modifiers and prognostic factors, assess missing covariates and population overlap, select a method that can estimate the chosen estimand, and conduct sensitivity analyses around key transportability assumptions. The approach should also be checked against current HTA guidance.

References

Chandler, C., & Ishak, J. (2026). Reframing population-adjusted indirect comparisons as a transportability problem: An estimand-based perspective and implications for health technology assessment (arXiv:2602.17041). arXiv. https://arxiv.org/abs/2602.17041
Chandler, C., & Ishak, J. (2026). Anchors away: Navigating unanchored indirect comparisons with multilevel unanchored meta-regression (ML-UMR) (arXiv:2606.20341). arXiv. https://doi.org/10.48550/arXiv.2606.20341
Remiro-Azócar, A. (2022). Target estimands for population-adjusted indirect comparisons. Statistics in Medicine, 41(28), 5558-5569. https://doi.org/10.1002/sim.9587
Phillippo, D. M., Dias, S., Ades, A. E., Belger, M., Brnabic, A., Schacht, A., Saure, D., Kadziola, Z., & Welton, N. J. (2020). Multilevel network meta-regression for population-adjusted treatment comparisons. Journal of the Royal Statistical Society: Series A, 183(3), 1189-1210. https://doi.org/10.1111/rssa.12579
Remiro-Azócar, A., Heath, A., & Baio, G. (2021). Methods for population adjustment with limited access to individual patient data: A review and simulation study. Research Synthesis Methods, 12(6), 750-775. https://doi.org/10.1002/jrsm.1511
Signorovitch, J. E., Sikirica, V., Erder, M. H., Xie, J., Lu, M., Hodgkins, P. S., Betts, K. A., & Wu, E. Q. (2012). Matching-adjusted indirect comparisons: A new tool for timely comparative effectiveness research. Value in Health, 15(6), 940-947. https://doi.org/10.1016/j.jval.2012.05.004
Caro, J. J., & Ishak, K. J. (2010). No head-to-head trial? Simulate the missing arms. PharmacoEconomics, 28(10), 957-967. https://doi.org/10.2165/11537420-000000000-00000

pubrica

AI and ML Services

Our Quality