Reframing Population-Adjusted Indirect Comparisons as a Transportability Problem: An Estimand-Based Perspective and Implications for Health Technology Assessment

Conor Chandler; Jack Ishak

REVIEW

Reframing Population-Adjusted Indirect Comparisons as a Transportability Problem: An Estimand-Based Perspective and Implications for Health Technology Assessment

T0 review · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Population-adjusted indirect comparisons identify comparator-population effects, and the shared effect modifier assumption alone does not make them transportable to other populations.

desk verdict Useful two-step transportability frame and a correct warning about marginal OR/HR, but the conditional-transportability guidance quietly assumes SPFA and overstates what SEMA alone buys you. read the letter →

arxiv 2602.17041 v4 pith:VUO4FW6M submitted 2026-02-19 stat.ME

Conor Chandler , Jack Ishak This is my paper · ORCID

classification stat.ME

keywords transportabilitypopulation-adjustedindirectcomparisonsmatching-adjustedcomparisonsimulatedtreatmentcollapsibilitysharedeffectmodifierassumptionestimandshealthtechnologyassessment

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Health technology assessment often relies on indirect comparisons when no head-to-head trials exist. The paper argues that the common adjustment methods—matching-adjusted and simulated treatment comparisons—only pin down the treatment effect in the comparator trial's population, not in any other population. It shows that the shared effect modifier assumption, widely cited as justifying transport of these effects, is not enough: for non-collapsible measures such as odds ratios and hazard ratios, the marginal effect changes with the covariate distribution even when the assumption holds. If accepted, this reframing changes how PAIC results should be reported and used in cost-effectiveness modeling: population-specific estimates need an explicit additional transport step before being applied to decision-relevant populations.

What carries the argument

The central object is the estimand-based decomposition of a pairwise anchored indirect comparison into two transport steps: first, conditional transport of the index-trial effect to the comparator population; second, implicit direct transport of the resulting active-to-active effect to the decision population. The technical workhorse is the structural model g(E[Y_t|X]) = m(x) + δ_t + φ(x)1{t≠A}, with a shared prognostic function m and shared effect modification φ under SEMA. Propositions A1–A2 establish when contrasts are invariant under covariate-distribution shifts; Propositions B1–B3 establish contrast-induced direct collapsibility, for example the log risk ratio case where the active-to-

What would settle it

Simulate an anchored comparison with SEMA and scale alignment holding by construction for a marginal odds ratio or hazard ratio, letting the index and comparator populations differ in the mean of an effect modifier; if the true marginal active-active effect is identical in every population, the paper's central claim is false, and if it varies, the claim is confirmed.

Watch

Extended reading notes

Core claim

The paper's central claim is that pairwise MAIC and STC identify active-to-active contrasts defined in the comparator population, and that these contrasts are not generally portable. SEMA ensures only that effect-modification terms cancel on a chosen scale; it does not make the marginal estimand invariant. Direct transportability of a marginal effect requires three joint conditions: SEMA; effect modification and the effect measure on the same linear predictor scale; and a collapsible measure. For conditional effects, SEMA plus scale alignment suffice. The paper proves this in formal propositions and demonstrates with simulations that mean differences and log risk ratios transport under SEMA,

Load-bearing premise

The formal results rest on the structural model in which the same baseline prognosis function applies to all treatments and the two active treatments share an identical effect-modification function on the model's scale; if those functions differ across treatments, the cancellation that produces transportable contrasts does not occur.

Editorial extensions

If this is right

Pairwise MAIC/STC results on odds-ratio or hazard-ratio scales should be labelled as comparator-population estimands; applying them to the index or real-world population requires explicit re-standardization or new assumptions.
HTA guidance that treats SEMA as sufficient for transporting marginal relative effects is too permissive; the transport step fails exactly for the non-collapsible measures most commonly submitted.
Cost-effectiveness models that feed MAIC/STC hazard or odds ratios into a decision model defined for another population risk transport bias that is not captured in the statistical confidence intervals.
For collapsible, scale-aligned measures (mean differences, log risk ratios under a log-link SEMA), direct transport is justified; analysts can design analyses around these measures when clinically reasonable.
Network-based methods that estimate effects in a pre-specified target population can avoid the implicit second step, but only under stronger network-wide assumptions about shared prognostic and effect-modification functions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper: the same two-step logic applies to any indirect comparison or network meta-analysis applied in an economic model; a marginal hazard ratio from an NMA is also population-specific, so the transport-bias concern is broader than MAIC/STC.
Beyond the paper: divergent MAIC results in different sponsors' submissions may be explained as different estimands anchored to different comparator populations, suggesting that consistency checks should compare target-population standardized effects rather than raw estimates.
Beyond the paper: a practical test for HTA reviews would be to request the covariate distributions of the comparator trial and run a model-based re-standardization to the decision population; if the effect changes materially, the submission should report both estimates.
Beyond the paper: the scale-alignment condition suggests an actionable design choice—specify effect modification on the same scale as the decision-relevant effect measure, or choose a collapsible measure—to make transportability claims more defensible.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear colleague,

The paper is worth your time, but read the conditional-transportability claims with a grain of salt. The core message — that in anchored MAIC/STC the active-to-active effect is identified in the comparator population, and moving it to a decision population is a separate, assumption-laden transport step — is correct and important. The two-step decomposition is genuinely useful, and the simulations cleanly show that marginal ORs and HRs are not directly transportable even under SEMA. That negative result lands against current NICE and EU HTA guidance, and it is the main contribution.

What is actually new: the explicit two-step formalization and the anchored active–active propositions (A1–A2, B1–B3) characterizing when conditional vs. marginal contrasts are directly transportable. The algebra checks out, and the illustrative examples are transparent population-level calculations with code in the supplement. The distinction between collapsibility and transportability in Appendix D is also well put.

The soft spot is real. The conditional-transportability proofs rest on Equation (A1), which assumes a single prognostic function m(x) common to all treatments — that is SPFA. Without SPFA, the conditional B-vs-C contrast at x is (m_B(x)-m_C(x)) + (δ_B−δ_C), which need not be constant in x. SEMA does not imply m_B=m_C. So Section 6.4 and Table 1 overstate what SEMA alone achieves for conditional estimands. Appendix A acknowledges the common m(x) as "standard," but then claims the only structural assumptions are additivity and shared φ(x); that claim is inconsistent. This is not fatal for the central negative result about non-collapsible marginal measures — that holds regardless — but it is load-bearing for the paper's positive guidance that conditional estimands on the linear predictor scale are directly transportable under SEMA. The guidance needs a caveat listing SPFA as a required assumption, and the "regardless of collapsibility" phrasing should be qualified.

Novelty is moderate: Remiro-Azócar and Phillippo et al. already established the population dependence of non-collapsible marginal measures. The paper's contribution is the explicit transport decomposition and the HTA decision framing, which is enough to be a useful methodological paper.

Who it's for: statisticians and HTA methodologists who use or review MAIC/STC, and anyone writing guidance about indirect comparisons. It deserves a serious referee; the SPFA issue is fixable in revision.

Overall: accept with revisions.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the transportability results are self-contained mathematical deductions; self-citations are background, and the SPFA caveat is an assumption-reporting issue, not a circular reduction.

full rationale

The paper's central derivation (Propositions A1–A2 and B1–B3) is a set of conditional mathematical implications from an explicitly specified structural model, Eq. (A1): g(E[Y_t|X]) = m(x) + δ_t + φ(x)1{t≠A}. The claim that SEMA alone does not make marginal OR/HR effects directly transportable is derived by computing marginal contrasts under this model and using the standard property that non-collapsible measures do not reduce to a covariate-independent contrast. No fitted constants are used to generate the illustrative results; the simulations are constructed examples, not empirical predictions. The two self-citations (Ishak et al. 2025, ref 19; Chandler & Ishak 2025, ref 77) provide background and an extension (ML-UMR), but the transportability proof does not rest on them; the load-bearing guidance citations (NICE DSU TSD 18; Phillippo et al.) are external. The skeptic's SPFA point is a genuine assumption-reporting issue: Prop. A1's Eq. (A1) sets a common m(x), and Appendix A acknowledges this as SPFA, yet Section 6.4 and Table 1 state conditional transportability under SEMA plus scale alignment without listing SPFA, and Appendix A says 'the only structural assumptions required are additivity on the g-scale and sharing of φ(x) across active treatments under SEMA.' If m(x) differed across active treatments, the x-dependence would not cancel even for conditional contrasts, so the Section 6.4 summary overstates what SEMA alone achieves. However, this is an omitted-assumption/overstatement, not a circular reduction: the formal proposition explicitly includes the common-m assumption, and the core negative result for non-collapsible marginal measures does not rely on SPFA. Therefore, under the hard rules requiring a specific reduction of a prediction to its inputs, no circular step qualifies.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The central transportability claims are derived from the structural model in Eq. (A1) plus SEMA/SPFA and scale alignment. No new physical or statistical entities are postulated; the simulations use hand-set constants rather than fitted parameters. The burden falls on whether the structural model and assumptions hold in real PAIC applications.

free parameters (1)

Simulation scenario constants (β0, β1, γB, γC, β2,B, β2,C, μX range, Weibull ν) = Hand-set, e.g., β0=20, β1=10, γB=10, γC=5, β2=2 (§7.1); β2=log(0.9), ν=1.5 (§7.3)
These are illustrative data-generating choices, not fitted parameters. They instantiate SEMA/no-SEMA scenarios but do not determine the qualitative transportability conclusions.

assumptions (5)

standard math Rubin causal model with potential outcomes Y_t for t∈{A,B,C}; consistency and positivity.
Section 2.1; the estimand framework presupposes the standard causal inference counterfactual assumptions.
domain assumption Conditional transportability in Step 1: all relevant effect modifiers/prognostic factors are measured and correctly modeled.
Sections 4 and 11; the paper explicitly conditions on successful Step 1: 'Our analysis intentionally conditions on successful completion of the first transport step.'
ad hoc to paper Shared prognostic factor assumption (SPFA): the baseline prognostic function m(x) is common to all treatments in Eq. (A1).
Appendix A, Eq. (A1); required for the X-dependence to cancel in η_B(x) − η_C(x). The paper notes this is 'standard' but does not test it.
domain assumption Shared effect modifier assumption (SEMA): the same φ(x) applies to B and C relative to A on the g-scale.
Section 5; the central PAIC assumption, but the paper shows it is not sufficient for non-collapsible marginal measures.
ad hoc to paper Scale alignment: h∘g^{-1} = I on the relevant range, i.e., the effect measure operates on the linear predictor scale.
Propositions A1/A2 and Section 6.3; an analytical condition required for direct transportability, not an empirical fact.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Reframing Population-Adjusted Indirect Comparisons as a Transportability Problem: An Estimand-Based Perspective and Implications for Health Technology Assessment." pith.science (2026). https://pith.science/paper/VUO4FW6M

@misc{pith2026260217041,
  author       = {Pith},
  title        = {Pith review of: Reframing Population-Adjusted Indirect Comparisons as a Transportability Problem: An Estimand-Based Perspective and Implications for Health Technology Assessment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VUO4FW6M}},
  note         = {Machine review of arXiv:2602.17041}
}

read the original abstract

Population-adjusted indirect comparisons (PAICs) are widely used to synthesize evidence when randomized controlled trials enroll different patient populations and head-to-head comparisons are unavailable. Although PAICs adjust for observed population differences across trials, adjustment alone does not ensure transportability of estimated effects to decision-relevant populations for health technology assessment (HTA). We examine and formalize transportability in PAICs from an estimand-based perspective. We distinguish conditional and marginal treatment effect estimands and show how transportability depends on effect modification, collapsibility, and alignment between the scale of effect modification and the effect measure. Using illustrative examples, we demonstrate that even when effect modifiers are shared across treatments, marginal effects are generally population-dependent for commonly used non-collapsible measures, including hazard ratios and odds ratios. Conversely, collapsible and conditional effects defined on the linear predictor scale exhibit more favorable transportability properties. We further show that pairwise PAIC approaches typically identify effects defined in the comparator population and that applying these estimates to other populations entails an additional, often implicit, transport step requiring further assumptions. This has direct implications for HTA, where PAIC-derived effects are routinely applied within cost-effectiveness and decision models defined for different target populations. Our results clarify when applying PAIC-derived treatment effects to desired target populations is justified, when doing so requires additional assumptions, and when results should instead be interpreted as population-specific rather than decision-relevant, supporting more transparent and principled use of indirect evidence in HTA and related decision-making contexts.