No procedure shows an unanchored adjustment set is complete
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
Anchored versus unanchored is a primary identification distinction: Phillippo et al. (doi:10.1177/0272989X17725740) state that unanchored comparisons make much stronger assumptions that are widely regarded as infeasible, and NICE DSU TSD 18 calls the requirement that all effect modifiers and prognostic factors be accounted for largely impossible to meet. It is not the only axis that matters, because MAIC, STC and ML-NMR still differ in estimand, model dependence, extrapolation and uncertainty, and anchored adjustment is not modifier-only on a non-collapsible scale, where shifting the distribution of a purely prognostic covariate moves the marginal odds ratio even with a constant conditional odds ratio and no interaction. What is unresolved is completeness: unanchored identification needs a measured set sufficient to block the study-selection paths for absolute potential outcomes, that set is fixed by a scientific claim rather than by the data, and no accepted procedure demonstrates that a proposed set is complete. ML-UMR (arXiv:2606.20341) extends unanchored comparison beyond the pairwise case and makes its assumptions explicit for sensitivity testing, which is an estimation and reporting advance rather than a completeness result.
Why it is open
With a common comparator, randomization still identifies the within-trial contrast and the analyst is fixing an imbalance in modifiers of that contrast; once the anchor is gone, study membership and treatment are inseparable and every cross-study difference in absolute outcomes loads onto the treatment effect. That is a change in what must be known rather than in how well it is estimated, so no choice of estimator recovers it and larger samples do not help. Because the required set is a claim about what drives cross-study outcome differences, any proposed list is an argument rather than a result, and method debates conducted at the estimator level let an analysis lose the anchor with no corresponding change in what is reported or how the result is labeled.
What has been tried
| Work | What it contributes |
|---|---|
| Phillippo et al. 2018, Medical Decision Making | sets out population-adjustment methods for HTA and distinguishes anchored from unanchored comparisons according to whether a common comparator arm is used, stating that unanchored comparisons make much stronger assumptions that are widely regarded as infeasible |
| NICE DSU Technical Support Document 18: Methods for Population-Adjusted Indirect Comparisons in Submissions to NICE | official guidance stating that unanchored MAIC and STC assume all effect modifiers and prognostic factors are accounted for, an assumption it calls very strong and largely considered impossible to meet, and that anchored approaches are preferred whenever a connected network exists |
| Chandler and Ishak 2026, Anchors Away: Navigating Unanchored Indirect Comparisons with Multilevel Unanchored Meta-Regression (ML-UMR), arXiv:2606.20341 | extends unanchored comparison beyond the pairwise case and makes its assumptions explicit so they can be varied in sensitivity analysis; it does not supply a procedure for demonstrating that an unanchored adjustment set is complete |
Probable solution or research direction
Make anchored versus unanchored the primary axis for guidance, reporting and appraisal, and route unanchored analyses into a transportability frame with explicit residual-confounding analysis instead of treating them as a variant of the same estimator family. State the required adjustment set as a causal restriction attached to the declared estimand and effect scale, separating modifiers of the randomized contrast from the prognostic drivers of cross-study outcome differences, and say where non-collapsibility makes prognostic standardization necessary even in the anchored case. Treat completeness as a target for quantitative bias analysis, partial identification, negative controls, proximal methods and hidden-truth calibration, and allow no strong causal reading of an unanchored comparison without one of these or an explicit statement that the result is exploratory.
Verification
Verdict. Partially addressed. Auditors agreed the problem is open and no counterevidence surfaced. Reopened by the full-text reading: 1 finding from 1 paper survived both the reader and an independent second reviewer, so the verdict moves from confirmed-open to partially-addressed, on a single finding.
The source titled this. “Anchored versus unanchored is the decisive identification split” The title calls the anchored versus unanchored split decisive, but the audit found it is not the only axis that matters and that anchored adjustment is not modifier-only on a non-collapsible scale.
The source said. “The decisive methodological fault line is anchored versus unanchored, not MAIC versus STC versus ML-NMR: an anchored PAIC needs only effect modifiers of the within-study randomized contrast, whereas an unanchored PAIC needs every prognostic factor and every effect modifier that explains cross-study outcome differences. The asymmetry is widely stated but not operationalized, because no procedure exists for enumerating or defending the completeness of the unanchored adjustment set.”
What was wrong with it. The source framed the split as decisive against the choice among MAIC, STC and ML-NMR and said anchored comparisons need only effect modifiers; the source-code auditor rejected both, since the estimators differ in estimand and extrapolation and non-collapsible marginal estimands can require prognostic standardization. The completeness claim held up, and this previously unattributed entry is now sourced to Phillippo et al. 2018 and NICE DSU TSD 18, whose own text asserts the gap, with ML-UMR (arXiv:2606.20341) named as the nearest recent work.
Revised by the full-text reading. Applied 2026-07-26 full-text reading of the consolidated library (687 papers).
- Verdict changed
- confirmed-open -> partially-addressed 1 surviving finding from: A process to validate prognostic factors for unanchored matc. New citations: 10.57264/cer-2024-0235
- Literature and prior-art check
- open / supported. Retrieved the Phillippo et al. 2018 abstract verbatim; it carries the entry’s central asymmetry claim word for word: ‘We distinguish between anchored and unanchored comparisons according to whether a common comparator arm is used or not. Unanchored comparisons make much stronger assumptions, which are widely regarded as infeasible.’ NICE DSU TSD 18 states that an unanchored MAIC or STC ‘effectively assumes that absolute outcomes can be predicted from the covariates; that is, it assumes that all effect modifiers and prognostic factors are accounted for. This assumption is very strong, and largely considered impossible to meet’ (retrieved via search of the TSD text, not the PDF directly, hence confidence 0.8 not higher). That guidance sentence is itself the source assertion that licenses the entry’s ‘no accepted procedure for demonstrating completeness’ claim, so it is not an unverifiable absence claim. Nothing I found resolves the identification problem. The most recent relevant method, ML-UMR (arXiv 2606.20341, Chandler and Ishak, June 2026), extends unanchored comparison beyond the pairwise case and makes assumptions explicit for sensitivity testing; that is an estimation and reporting advance, not a completeness result, and the entry does not cite it.
- GPT-5.6 Sol, technical and source-code lens
- open / overstated. Anchoring is a major identification distinction, but C2 creates a false choice: MAIC, STC, and ML-NMR still differ in estimand, model dependence, extrapolation, and uncertainty. C3 is valid only for suitable conditional or collapsible relative-effect estimands under conditional constancy and support. On non-collapsible marginal scales, a purely prognostic covariate can matter without a conditional interaction. For example, a logistic model with constant conditional OR 3 and no treatment-covariate interaction gives marginal ORs 1.958 and 2.581 when Pr(X=1) changes from 0.1 to 0.9. C4 is therefore too reassuring: randomization identifies within-trial effects, not their transport, and network connectivity alone supplies neither overlap nor transitivity. Unanchored identification requires a sufficient set blocking study-selection paths for the relevant potential outcomes, not literally every prognostic variable. QBA characterizes sensitivity but does not itself license a causal interpretation. Claims about what NICE documents say are outside this technical audit.
- Decision path
-
R5:default-open → R9:substantiated-single-overstatement:codex
How the evidence falls in time
9 papers in the reviewed corpus bear on this problem, published 2019 to 2026. Partial progress is on the record, and later work still calls this open. Both belong in view.
| Year | Says | Paper |
|---|---|---|
| 2019 | confirms it is open | Population Adjustment Methods for Indirect Comparisons: A Review of National Institute for Health and Care Excellence Technology Appraisals |
| 2019 | confirms it is open | Using Matching-Adjusted Indirect Comparisons and Network Meta-analyses to Compare Efficacy of Brexanolone Injection with Selective Serotonin Reuptake Inhibitors for Treating Postpartum Depression |
| 2020 | confirms it is open | Performance of unanchored matching-adjusted indirect comparison (MAIC) for the evidence synthesis of single-arm trials with time-to-event outcomes |
| 2024 | confirms it is open | Feasibility of Indirect Treatment Comparisons Between Niraparib Plus Abiraterone Acetate and Other First-Line Poly ADP-Ribose Polymerase Inhibitor Treatment Regimens for Patients with BRCA1/2 Mutation-Positive Metastatic Castration-Resistant Prostate Cancer |
| 2024 | confirms it is open | Response to discussant comments on “NMA, the first 20 years” |
| 2025 | partly addresses it | A process to validate prognostic factors for unanchored matching-adjusted indirect comparison of single-arm trials in oncology: a proof-of-concept study |
| 2025 | confirms it is open | Comparison of Diagnostic Bronchoscopy Single-Arm Trials Using an Advanced Optimization Approach for Indirect Comparisons with Partial Individual Patient Data |
| 2025 | confirms it is open | Letter to the Editor Regarding “Dupilumab Versus Lebrikizumab Demonstrates Greater Likelihood of Achieving and Maintaining Improvements in Efficacy Outcomes Using a Placebo Adjusted Indirect Treatment Comparison” |
| 2026 | confirms it is open | Can We Trust PAICs in Rare Diseases? Methodological Challenges and Limitations |
Source
Derived from A:L7, A:L48, A:L80, C:L187-188, C:L196, C:L392, C:L2655-2656 of the reviewed corpus.