Nothing requires declaring whether a result is marginal or conditional
Source note. This entry derives in part from a source document whose inline citations were ChatGPT interface tokens rather than references, and so resolve to nothing. Claims below are attributed to that document and were checked independently where possible; anything that could not be independently sourced is marked in the verification trail.
Statement
No journal or HTA template requires an analysis to declare all four of target population, target estimand, effect scale, and whether the result is marginal or conditional; the marginal versus conditional element in particular is nowhere mandated. Two of the four are already required for NICE submissions: NICE DSU TSD18 (doi:10.1177/0272989X17725740) requires the target population to be stated explicitly and requires indirect comparisons to be carried out on the linear predictor scale. A published PAIC framework also codifies definition of the comparison as an estimand as its first element (Ishak et al. 2025, doi:10.1007/s40273-025-01503-1), and Phillippo et al. 2025 (doi:10.1017/rsm.2025.2) give recommendations for reporting marginal and conditional estimands in the decision target population, so what is missing is a single enforcement point rather than a standard to enforce. Population-superscripted notation likewise already appears in the methods literature, including TSD18; what has not happened is its adoption in applied reports.
Why it is open
The requirements that exist are scattered across a technical support document, a framework paper and a recommendations paper, so there is no one field where a reviewer checks all four and no error signal when one is missing; the marginal versus conditional element has recommendations behind it but no template that demands it. Writing the declaration forces the analyst to fix the target population before seeing the data, which conflicts with the current practice of choosing the adjustment set from whatever the comparator publication reported. Because nothing pushes the population into the name of the estimand in applied reports, two numerically different quantities are published under one label and inherited unlabeled by appraisal templates and economic models. The mismatch changes the number rather than breaking the analysis, so nothing forces the field to correct itself.
What has been tried
| Work | What it contributes |
|---|---|
| Phillippo, Ades, Dias, Palmer, Abrams and Welton 2018, Methods for Population-Adjusted Indirect Comparisons in Health Technology Appraisal, Medical Decision Making (NICE DSU TSD18) | Requires the target population to be stated explicitly and requires indirect comparisons to be carried out on the linear predictor scale with the usual link functions, covering two of the four declaration elements for NICE submissions, and uses population-indexed estimand notation. It does not require a marginal versus conditional declaration. |
| Ishak et al. 2025, A Framework for Reliable, Transparent, and Reproducible Population-Adjusted Indirect Comparisons, PharmacoEconomics | Six-element PAIC reporting and risk-of-bias framework whose first element is definition of the comparison of interest in terms of an estimand. It is a published framework rather than a mandated field in journal or submission templates, and it does not impose population-indexed notation. |
| Phillippo, Remiro-Azocar, Heath, Baio, Dias, Ades and Welton 2025, Effect modification and non-collapsibility together may lead to conflicting treatment decisions, Research Synthesis Methods | Provides published recommendations for declaring and jointly reporting marginal and conditional estimands in the decision target population, the element TSD18 does not cover. |
| cpaic, unreleased development package on GitHub (not on CRAN), commit 9d150e9 | Requires a target profile for effect-modified cML-NMR output and prints the result as a conditional effect at that profile, stating that relative effects from a model with effect modifiers are population-specific. Partial enforcement of the declaration inside one unreleased package. |
Probable solution or research direction
Assemble the declaration from the pieces that exist rather than writing a new standard: TSD18 fixes target population and scale, Ishak et al. 2025 supply a PAIC framework whose first element is estimand definition, and Phillippo et al. 2025 supply the marginal versus conditional recommendation, so the work is binding them into one mandatory field in analysis plans, journal templates and submission templates. Require every reported contrast to identify its population, by superscript or by an equally explicit structured field, since the point is unambiguous identification rather than one notation. Forbid silent reinterpretation of a conditional effect as a marginal one, or of a comparator-trial effect as a decision-target effect, and treat software that already enforces part of this, such as cpaic requiring a target profile for effect-modified cML-NMR output, as a pattern to generalize rather than as a finished solution; cpaic is an unreleased development package on GitHub and is not installable from CRAN.
Verification
Verdict. Overstated. 2 independent auditors judged the claim as stated too strong. No journal or HTA template requires a four-part declaration; in particular nothing requires an analysis to state whether its result is marginal or conditional, even though NICE DSU TSD18 has required an explicit target population and a pre-specified linear-predictor scale since 2016, and population-superscripted notation already exists in the methods literature without being adopted in applied reports.
The source titled this. “No required estimand declaration or population-indexed notation” The title says population-indexed notation is missing, but the audit found it already used in TSD 18 and the methods literature; only applied adoption is absent.
The source said. “Nothing in the publication or submission pipeline requires an analysis to declare its target population, target estimand, effect scale, and whether the result is marginal or conditional, so the literature continues to tolerate analyses that switch interpretations without admitting it. The report also requires population-indexed notation of the form Delta_AB^{P2} versus Delta_AB^{P*}, and asserts that a report should never call both quantities simply the treatment effect.”
What was wrong with it. The claim was unattributed in the source and is now sourced, and both auditors judged it overstated: the literature auditor found that NICE DSU TSD18 (doi:10.1177/0272989X17725740) already requires an explicit target population and a pre-specified linear-predictor scale and already uses population-indexed notation, and that Phillippo et al. 2025 (doi:10.1017/rsm.2025.2) supplies the marginal versus conditional recommendations TSD18 does not cover. A PAIC reporting and risk-of-bias framework established elsewhere in this audit, Ishak et al. 2025 (doi:10.1007/s40273-025-01503-1), makes definition of the comparison as an estimand its first element. GPT-5.6 Sol added that population-superscripted notation is a useful convention rather than a mathematical necessity, since explicit prose or structured fields identify the population equally well.
Revised by the full-text reading. Applied 2026-07-26 full-text reading of the consolidated library (687 papers).
- closure-upheld
- 5 findings from the reading assert this problem is open; the closure was re-examined and stands Put to an independent reviewer with the closing auditors’ reasoning, the work they relied on, and the new findings in view together. The new findings reinforce the closure’s residual: no cited checklist or decision-facing framework requires declaration of marginal versus conditional status, and estimand uptake in evidence synthesis remains limited. They also independently confirm the closure’s qualification that target population and effect scale have existing reporting requirements or checklist items. The 2026 JCA framework is later than some closure evidence, but it does not supersede or contradict TSD18’s requirements and merely lacks the missing enforcement element already identified. Thus the current statement correctly distinguishes scattered partial standards from the absent single four-part mandate. (reviewer confidence: high)
- Flagged for reopening
-
6 findings from the reading assert this problem is open while the registry carries
overstatedFlagged, not applied: reversing an earlier audit’s closure needs the closure’s own evidence re-examined alongside this, which the reading did not do. See documentation/audit/reading/PROPOSED.md. - Literature and prior-art check
- partially-solved / overstated. Also SOURCE-A-UNSOURCED. Reading the full text of Phillippo et al. 2018 Med Decis Making (the published NICE DSU TSD18) contradicts the framing that ‘nothing in the publication or submission pipeline requires an analysis to declare its target population … [and] effect scale’. Recommendation 6 requires the target population to be explicitly stated; Recommendation 5 requires the scale: ‘Indirect comparisons should be carried out on the linear predictor scale, with the same link functions that are usually employed for those outcomes.’ TSD18 is the reference document for NICE submissions, so an enforcement point exists for two of the four declaration elements. That same paper already uses population-indexed notation of exactly the requested form, writing estimands with a population superscript and stating the transport relation between two populations P and Q, so ‘no established standard uses source-target notation’ is too strong for the methodological literature even if true of applied reports. What genuinely remains undeclared and unenforced is the marginal-versus-conditional element; Phillippo et al. 2025 RSM is the closest thing to a normative recommendation there. Cites doi:10.1177/0272989X17725740: Supplies an existing enforcement point for two of the four declaration elements (target population, effect scale) and already uses population-indexed estimand notation.. Cites doi:10.1017/rsm.2025.2: Provides published recommendations for declaring and jointly reporting marginal and conditional estimands in the decision target population, the element TSD18 does not cover..
- GPT-5.6 Sol, technical and source-code lens
- partially-solved / overstated. C1 is a placeholder. C2, C3, C5, C6, C9, and C10 are claims about current practice, prevalence, or standards that cannot be established from technical correctness alone. C4 is a sound minimum declaration, although a full estimand also needs outcome, treatment conditions, time horizon, and handling of intercurrent events where relevant. C7 is useful notation, not a mathematical necessity; equally explicit prose or structured fields can identify the population. C8 is a sound communication rule unless the quantities have been shown to coincide and that fact is stated. The absolute statement that nothing in the pipeline requires target or scale declaration is false, and cpaic itself requires a target for effect-modified cML-NMR output and labels it as conditional. Existing guidance and package behavior partly solve declaration; universal four-part enforcement and routine source-target notation remain operational issues. Cites doi:10.1177/0272989X17725740: Requires explicit target-population declaration and specifies the synthesis scale, covering two of the proposed four fields.. Cites doi:10.1017/rsm.2025.2: Supplies an explicit framework for distinguishing and reporting marginal and conditional estimands.. Cites choxos/cpaic@9d150e9cad88f2e6fbcd94dbb92c3dee3b7c2dca: Requires a target profile for effect-modified cML-NMR and prints the result as a conditional effect at that profile..
- Decision path
-
R3:partially-solved(n=2) → R8:overstated-majority(n=2)
How the evidence falls in time
6 papers in the reviewed corpus bear on this problem, published 2011 to 2026. Partial progress is on the record, and later work still calls this open. Both belong in view.
This problem is recurrent: three or more papers spanning at least eight years assert it independently. Sustained restatement by authors who mostly do not cite each other is the strongest evidence this reading can offer that a gap is real rather than one group’s framing.
| Year | Says | Paper |
|---|---|---|
| 2011 | confirms it is open | Interpreting indirect treatment comparisons and network meta-analysis for health-care decision making: report of the ISPOR Task Force on Indirect Treatment Comparisons Good Research Practices: part 1 |
| 2013 | confirms it is open | Evidence synthesis for decision making 7: a reviewer’s checklist |
| 2017 | partly addresses it | Methods for Population-Adjusted Indirect Comparisons in Health Technology Appraisal |
| 2024 | confirms it is open | Broad versus narrow research questions in evidence synthesis: A parallel to (and plea for) estimands |
| 2025 | confirms it is open | Effect modification and non-collapsibility together may lead to conflicting treatment decisions: A review of marginal and conditional estimands and recommendations for decision-making |
| 2026 | confirms it is open | Evidence Synthesis via Indirect Treatment Comparisons in the European Framework of Joint Clinical Assessment |
Source
Derived from A:L80, A:L214, C:L300-306, C:L219 of the reviewed corpus.