Trust, Access, and the Division of Responsibility for Vaccine Confidence in Central and Eastern Europe: Paired Own-Country and Regional Ratings from a Multi-Stakeholder Professional Panel

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript addresses vaccine confidence in Central and Eastern Europe (CEE), conducting a multi-stakeholder expert survey. Through an innovative design using "dual national/regional rating" and "impact/feasibility dual-dimension assessment," it identifies the cognitive asymmetry of "internalization of trust issues and externalization of structural issues," as well as two priority interventions: continuing education for clinicians and mobilization of trusted community figures. The research topic holds significant implications for regional public health policy, the design is sophisticated, the analysis meticulous, and the discussion in-depth. The following issues need to be addressed prior to publication:

1. Of the 62 respondents, 39 are from Romania, all of whom are professionals actively involved in vaccine-related issues. Does this highly engaged expert group systematically overestimate the severity of trust and information barriers while underestimating structural access barriers such as geographic and economic factors?
2. Why did the authors not design further tests (such as subgroup analysis of response variance for different types of barriers) to distinguish perceptual bias from the actual effects of institutional responsibility allocation?
3. The expected impact of the 16 interventions shows only a very weak and non-significant positive correlation with feasibility over 12–24 months (r = 0.21). To what extent does this near-zero correlation result from the conceptual heterogeneity of the intervention list itself, rather than a real-world decoupling of effectiveness and implementability?
4. The authors acknowledge that the study could only detect relatively large effect sizes (d = 0.37 requires approximately 120 per group). In this context, is the conclusion of “no significant difference in priorities between the two groups” at risk of Type II error (false negative)? Can it be concluded that the perspectives of different stakeholders are truly consistent, or is this due to insufficient sample size?
5. The study finds that "equity gap and coverage monitoring" is the most diffusely assigned and least consensual responsibility among the five system functions (normalized entropy 0.82). However, the questionnaire only asked "who should take the lead" without probing deeper causes for the ambiguity in responsibility allocation.
6. Does the tool design overlook key barriers within the healthcare supply side? Does this lead to the “continuing education for clinicians” solution focusing solely on communication skills, without addressing providers’ own confidence and legal risk concerns?
7. The study is a single time-point cross-sectional survey and cannot assess the temporal stability of expert perceptions or verify the actual effects of the recommended interventions.
8. How do the authors determine that these priority rankings are not instantaneous judgments driven by short-term events (e.g., specific outbreaks, local policy changes)? Without linkage to actual vaccination rate data, how can the policy value of expert perceptions be evaluated?
9. The study’s final policy insights focus mainly on the methodological level (e.g., adopting dual-level assessment, promoting peer learning), but do not propose specific, actionable policy paths tailored to the CEE context.
10. It is recommended to add a graphical abstract illustrating the study’s distinctive dual-perspective asymmetry model.
11. It is recommended to include a Key Points module after the abstract or before the conclusion, summarizing in 3–4 bullets the core scientific findings, methodological innovations, and policy implications to help readers grasp the article’s key contributions quickly.
12. Reference [23] is a companion paper from the same conference with substantial author overlap; [22] is a white paper released by the conference organizer and serves as the conceptual framework for this study. These multiple outputs from the same project form an internal citation cycle, all published concurrently in 2026, and cannot serve as independent validation evidence. The authors are advised to clearly specify the relationship between the two papers and emphasize in the limitations that their conclusions do not constitute independent validation.
13. The core conceptual framework is derived from two grey literature sources [8][22] (the ILC-UK report and COPAC white paper), neither of which underwent peer review. The authors are advised to state the role of these documents in the methods or limitations and supplement additional peer-reviewed literature to support the central framework.
14. The "spatial optimism" theory used to explain cognitive asymmetry cites only three early studies in environmental psychology [24-26], lacking direct empirical support from public health and vaccine domains. It is recommended to add related studies on similar cognitive biases in the health behavior field to strengthen the specificity of the theoretical foundation.
15. All literature on Roma vaccination [39-41] dates from 2022 or earlier.
16. Some reference DOI formats are inconsistent, with missing URL prefixes; some journal title abbreviations are inconsistent.
17. The Romanian language literature [2] does not provide an English title; it is recommended to supply an English translation to facilitate international reader comprehension.

Author Response

Dear Reviewer,

We are grateful for your judgement that the topic holds significant implications for regional public health policy and that the design, analysis and discussion are sound, and for seventeen comments that pushed the paper in two directions at once: towards more caution in reading professional perceptions, and towards more concrete relevance for practice. We have taken both. Where a comment overlapped with one from Reviewer 3 we say so, because the two together changed how we analysed and framed the central result. Our responses follow your comments in order; all changes are highlighted in the revised manuscript.

Comment 1.1
Does this highly engaged expert group systematically overestimate trust and information barriers while underestimating structural access barriers?

Response: This is, in our view, the single most important question raised about the paper. The submitted version already carried part of the answer, in the Limitations (a panel drawn from patients or from underserved communities could reasonably order the same fifteen items differently) and in the Conclusions (these are professional perceptions, not measured outcomes, and the low ranking of access records how those barriers look from the delivery side); both passages are retained. What is new is a direct answer in the Limitations: whether the panel over-weights trust and information barriers cannot be tested internally, because no independent measure of barrier severity exists for the same countries, and the one internal signal, that perceived severity of the culturally adapted communication barrier rose with depth of engagement with high-barrier populations (Section 3.6), is consistent with proximity raising the weight given to barriers a respondent does not personally face. The Abstract states the same caveat.

Comment 1.2
Why did the authors not design further tests to distinguish perceptual bias from the actual effects of institutional responsibility allocation?

Response: We agree that the two cannot be separated by this design. The structural reading, that access and life-course policy gaps may be assigned to the regional level because they require cross-border coordination, was already set out in the Discussion of the submitted version alongside the perceptual one, together with the statement that their relative contribution cannot be determined without independent measures of national and regional performance; that passage is retained. We have not added a further test, because any test built on these data would still rest on perceptions alone. We have added two things: the Discussion now presents spatial optimism, the division of responsibility between levels and coarser discrimination about the aggregate as three compatible explanations between which the design cannot choose, and the Limitations name the indicators a future test would need (WHO/UNICEF coverage estimates, out-of-pocket vaccine costs, the breadth of adult schedules).

Comment 1.3
To what extent does the near-zero impact–feasibility correlation (r = 0.21) result from the conceptual heterogeneity of the intervention list rather than a real decoupling?

Response: Reviewer 3 raised the same analysis from a different angle, and the two together led us to redo it. The r = 0.21 was computed across 16 intervention-level means, which discards the repeated-measures structure. On the 992 paired ratings, a linear mixed model with crossed random intercepts for respondent and intervention (feasibility regressed on expected impact, restricted maximum likelihood) gives a clear positive association (slope 0.40, SE 0.03, p < 0.001), and the repeated-measures correlation is r = 0.33 (95% CI 0.27–0.39). Section 3.4 now reports both, states the unit of analysis for each, and concludes that the dimensions are related but not interchangeable; the sentences in Section 3.4, the Discussion and the Conclusions that described the two axes as unrelated or independent have been reworded to say that the panel-level orderings only partly coincide. On the specific question: the heterogeneity of the list is real, but what the quadrant analysis rests on is the ordering of the panel’s average ratings, and that ordering diverges between the two dimensions whatever the within-respondent association. The Abstract no longer claims that expected impact did not predict feasibility.

Comment 1.4
Is the conclusion of “no significant difference between the two stakeholder groups” at risk of Type II error?

Response: Yes, and the manuscript now says so without hedging. The Limitations state that the stakeholder comparison used only the two largest categories, is powered to detect only large differences, and that its null results are inconclusive rather than confirmatory. The arithmetic is now given where the null result is claimed: with 25 and 21 respondents in the two largest groups the comparison has 80% power only for differences of about d = 0.85, and the largest difference actually observed (d = 0.37) would require roughly 116 respondents per group to detect. We have been careful throughout not to describe the groups as agreeing, only as not differing detectably at this sample size.

Comment 1.5
The questionnaire asked who should lead but did not probe the causes of the ambiguity in responsibility for equity monitoring.

Response: Accepted; this is a real limit of the instrument. The Limitations now record that respondents were asked which actor should lead but not why responsibility for equity monitoring is so diffusely assigned, so the reasons behind the widest dispersion in the set, whether contested mandate, absent capacity or missing data infrastructure, remain unexamined.

Comment 1.6
Does the instrument overlook barriers on the healthcare supply side, so that “continuing education for clinicians” addresses only communication skills and not providers’ own confidence and legal risk concerns?

Response: This is well observed and we have added it. The Limitations now state that the barrier list covers the supply side thinly: it captures weak clinician recommendation and workforce shortages but not clinicians’ own confidence in newer products, their perceived legal or professional exposure, or the time and reimbursement available for a vaccination conversation. The omission is visible in the study’s own data, which we now report: provider-side hesitancy, fear of adverse events and fear of legal responsibility appear in the first open-ended theme in Section 3.8, and calls for a no-fault compensation system in the third, none of them with a closed item to attach to. Their absence from the ranked results is therefore an artefact of the item list rather than evidence of low salience, and it follows that an education recommendation confined to communication skills would address only part of what the panel itself raised.

Comment 1.7
The study is a single time-point cross-sectional survey and cannot assess the temporal stability of expert perceptions.

Response: Agreed and now stated directly: ratings were collected at a single time point, so nothing here speaks to the stability of these judgements.

Comment 1.8
How do the authors know these rankings are not instantaneous judgements driven by short-term events? Without linkage to vaccination rate data, how can the policy value be evaluated?

Response: The Limitations now state that ratings were collected at a single time point and that priorities recorded at one moment may partly reflect what was salient at the time rather than a settled ordering. The submitted version already stated that no external validation against coverage data was attempted, so that the asymmetry tests establish that perceptions diverge and not which perception is closer to the truth; that statement is retained, and the Limitations now name the indicators such a validation would need.

Comment 1.9
The policy insights are mainly methodological and do not propose specific, actionable policy paths for the CEE context.

Response: The Discussion now contains a subsection on implications for practice (Section 4.1), before the Limitations. It names the two measures rated high on both expected impact and near-term feasibility as the place to start; argues that the high-impact, lower-feasibility group is a reason to begin those measures now rather than to defer them; and reports the concrete actions respondents themselves named in free text: paying community and Roma health mediators, vaccinating at the point of contact rather than referring onward, checking for missing doses at any clinical encounter, and following up families who do not return. We stop short of prescribing these as policy: none is established as effective by this survey, and the evidence on outreach cautions against assuming it. What we do recommend is that they be commissioned with evaluation attached, and that an owner be named for equity monitoring, a decision that can be taken without new evidence.

Comment 1.10
It is recommended to add a graphical abstract illustrating the dual-perspective asymmetry model.

Response: On reflection we have added one. We initially did not because Reviewer 3 asked us to stop presenting the asymmetry as an established mechanism, and we have accordingly reframed it as a difference in assessment between a national and an aggregate regional referent whose cause this design cannot identify. A graphical abstract depicting a “dual-perspective asymmetry model” would have reinstated, in the most prominent position in the article, the claim we had just been asked to withdraw. The version now supplied avoids that: three descriptive panels showing the paired own-country and regional ratings item by item, the divergence by domain, and the contrast between the most and least consensual lead-actor allocations, with a footer stating that the figure shows professional perceptions and not measured outcomes. Every value in it is taken from Table S1, Section 3.3 and Section 3.7, and the file is supplied as Graphical_Abstract.png (3300 × 1680 pixels).

Comment 1.11
It is recommended to include a Key Points module of 3–4 bullets.

Response: Added after the keywords: four bullets covering the barrier ordering, the dependence of assessments on the geographic referent, the value of rating impact and feasibility separately, and the unowned equity-monitoring function.

Comment 1.12
References [22] and [23] and the present paper form an internal citation cycle and cannot serve as independent validation.

Response: Accepted in full. The Limitations now state that the white paper [22], the qualitative synthesis [23] and this survey are three outputs of one project, published in the same year and drawing on overlapping groups of contributors, and that agreement among them is internal consistency within a single programme of work rather than independent corroboration from separate evidence streams.

Comment 1.13
The core conceptual framework rests on two grey-literature sources [8] and [22]; state their role and supplement with peer-reviewed literature.

Response: The Limitations now state explicitly that neither source was peer reviewed and that they are used as the origin of the item structure the panel was asked to rate, not as evidence for any claim made here. We add that the substantive claim they encode, that the determinants of vaccine confidence are multi-domain, is itself well supported in the peer-reviewed literature, and cite an account of the behavioural, social and contextual determinants of vaccine acceptance and demand written for the European region [54] and a synthesis of two decades of hesitancy research [55]. The grey-literature sources supplied the wording and grouping of the items, not the evidence for the constructs.

Comment 1.14
The spatial-optimism explanation cites only three early environmental-psychology studies; add evidence of similar biases in health behaviour.

Response: We take the point, and note that the framing itself has been softened following Reviewer 3: spatial optimism is now offered as one candidate explanation rather than as the mechanism. We have added the closest health-domain analogue we could find, which is not in vaccination: clinicians consistently judge antimicrobial resistance a more serious problem nationally than in their own institution or practice [56,57]. We say in the text that this analogue runs in one direction only, which matches the direction observed here for access barriers and is the opposite of that for trust and information barriers; it therefore shows that professionals’ assessments can depend on the referent without indicating which way they will run, it does not establish the mechanism, and we are not aware of an equivalent demonstration for vaccine confidence.

Comment 1.15
All literature on Roma vaccination [39–41] dates from 2022 or earlier.

Response: Correct, and now addressed. The Discussion cites recent coverage and attitude data on Roma children [47] and a 2026 realist review of health-system interventions for minority populations in middle- and high-income countries [48], which finds that outcomes depend on whether services are redesigned around the population rather than merely offered to it. Adding these two references, and removing former reference 50 (see the technical-check response), renumbered former references 47–49 as 49–51 and former 51–52 as 52–53; the four references added for comment 1.13 and comment 1.14 are 54–57.

Comment 1.16
Some DOI formats are inconsistent, with missing URL prefixes, and some journal abbreviations are inconsistent.

Response: Corrected. References 11 and 13 carried a malformed “doi:” prefix, reference 13 in the hybrid form “https://doi:10.1016/…”; both now use https://doi.org/. Reference 5 had a corrupted volume field (“ume 18”) and a lowercased DOI suffix. Journal abbreviations were corrected in references 2, 5, 6 and 9, and the lost diacritic in “Rémy” and a missing article number were restored in reference 1. In the present revision we also corrected the remaining inconsistencies of the same kind: journal abbreviations in references 17, 20, 29, 35, 37, 38 and 43 (for example “Front. Public Heal.” to “Front. Public Health”), misplaced author initials in references 34, 41, 42 and 45, the missing journal name in reference 42, missing article numbers in references 33 and 38, and a missing DOI in reference 12.

Comment 1.17
The Romanian-language reference [2] should carry an English translation of its title.

Response: Added in square brackets after the Romanian title, following MDPI practice for non-English sources. A typographic error in the second author’s name was also corrected.

We thank you again for the time and care you gave to the manuscript.

The authors

Reviewer 2 Report

Comments and Suggestions for Authors

The article written by Teodor Cristian Blidaru et al., entitled "Where Experts Place the Problem: Trust, Access and the Division of Responsibility for Vaccine Confidence in Central and Eastern Europe — A Multi-Stakeholder Expert Survey" is well presented and very easy to understand.

Authors in this work tried to answer many questions regarding the vaccine confidence in the region of the center and east of Europe. for this objective, authors used a structured survey among professionals working in different country of the region. 

The analysis and the presentation of results are interesting, discussion and conclusion are coherent to obtained results. However, some points should be improved before the publicayion of this study

• Abstract should be revized, Objectives should be clear and written before methodology section among the abstract.
• The article is very long. Many sections should be reduced
• Conclusion section should be reduced and should be more clear where authors demonstrated the main conclusion of the study
• Limitation section also should be short and included in discussion results
• Please reduce the number of figures and tables. Some of there should be incorporate in supplementary material

Theses remarks could improve the quality of the manuscript and make the paper more appropriate for publication...

Author Response

Dear Reviewer,

We are grateful for your judgement that the article is well presented and easy to understand, and for the direction of your comments, which is towards concision. We note one tension: Reviewer 3 requires substantial additional methodological detail and the full instrument. We have tried to satisfy both by moving material out of the article rather than deleting it. Our responses follow your comments in order; all changes are highlighted in the revised manuscript.

Comment 2.1
The abstract should be revised; the objectives should be clear and stated before the methodology.

Response: The Abstract now opens with an explicit statement of the three objectives, how professionals weigh competing barriers, whether their assessments differ by geographic referent, and how they rank interventions on impact and feasibility, before the Methods section, within the Background/Objectives heading required by the journal template. It has also been shortened to about 340 words.

Comment 2.2
The article is very long; many sections should be reduced.

Response: You are right that the paper is long, and we should be candid about the net result. We have cut where cutting did not remove content the other reviewers asked for: the Limitations are now a subsection of the Discussion (comment 2.4); Table 1, Table 3 and Figure 6 have moved to the Supplementary Material (comment 2.5), leaving five figures and one table in the article; the Abstract is shorter; and a duplicated passage on rating dispersion has been removed. Against that, Reviewer 3 required substantive additions: eligibility and sampling detail in Methods 2.1, a fuller description of the analyses in Methods 2.3, the multilevel re-analysis in Section 3.4, the profile of the excluded respondents, and several further limitations. The main text (Introduction to Conclusions, excluding tables and captions) is therefore about 11,100 words in the submitted version and about 13,600 words now. We judged that meeting Reviewer 3’s requirements mattered more than the word count, but every addition is confined to what was asked for.

Comment 2.3
The Conclusions should be reduced and should state the main conclusion more clearly.

Response: Rewritten. The Conclusions now open with the substantive finding, follow with the geographic asymmetry stated as a difference in assessment, and close with three qualifications and the practical implication. The section is shorter and states the main conclusion in its first sentence.

Comment 2.4
The Limitations section should be shortened and incorporated into the Discussion.

Response: The Limitations are now Section 4.2, a subsection of the Discussion, and the Conclusions have been renumbered accordingly. The point about the reduced dispersion of regional ratings, previously made in two paragraphs, is now made once: the duplicate sentence has been removed and its one non-duplicated clause (that the standardised estimate should be taken as the size of the attributional component) has been moved to the paragraph where the point is first made. As explained under comment 2.2, the section has grown rather than shrunk because Reviewer 3 asked for several further limitations to be stated; it now sits inside the Discussion as requested.

Comment 2.5
Reduce the number of figures and tables; some should be moved to supplementary material.

Response: Done. Table 1 (barrier ratings by item), Table 3 (domain ratings by stakeholder group) and Figure 6 (barrier severity by domain for the two largest stakeholder groups) have moved to the Supplementary Material as Table S1, Table S4 and Figure S1, and the text refers to them there. The article now carries five figures and one table (the intervention ratings, formerly Table 2, now Table 1). Figure 1 already shows the item-level barrier ratings that Table 1 tabulated, and the stakeholder comparison in Section 3.9 is, as Reviewer 3 notes, under-powered, so neither move removes anything the argument depends on.

We thank you again for the time and care you gave to the manuscript.

The authors

Reviewer 3 Report

Comments and Suggestions for Authors

Manuscript vaccines-4538924

"Where Experts Place the Problem: Trust, Access and the Division of Responsibility for Vaccine Confidence in Central and Eastern Europe — A Multi-Stakeholder Expert Survey"

Dear authors,
This cross-sectional survey of 62 professionals rates barriers and interventions for vaccine confidence in Central and Eastern Europe (CEE). Rating each barrier at two geographic levels (own country and CEE region) and rating each intervention separately for expected impact and feasibility are strong design choices. The analysis is more transparent than most survey work in this field, and the Limitations section addresses many of the concerns a reviewer would ordinarily raise. Several problems remain, however, and some of them sit at the centre of the paper's argument.

Major comments

1. The central interpretive claim goes further than the design allows.

The paper reads the own-country versus CEE-region contrast as evidence that professionals locate trust problems “where the observer stands” and structural problems “elsewhere”, and frames this through the spatial-optimism literature. That interpretation runs through the title, Abstract, Results, Discussion and Conclusions.

The comparison is not conceptually identical for all respondents. The 47 CEE-based professionals compare a specific national setting with an aggregate region that contains it. The 15 non-CEE respondents compare two geographically separate settings. Pooling both groups in the headline within-respondent analysis therefore treats two potentially different contrasts as estimates of a single construct.

More importantly, the data show that national-level and aggregate-regional ratings diverge by barrier type. They do not directly show that respondents attribute structural problems to “elsewhere”. A Romanian respondent, for example, compares Romania with CEE as a whole, not Romania with the rest of CEE.

The manuscript's own analyses reinforce this distinction. Regional ratings are substantially less dispersed than own-country ratings (within-respondent SD 0.67 vs. 0.99), consistent with coarser discrimination about a broader and potentially less familiar object. Standardisation attenuates the interaction to +0.34 SD units and excludes 12 respondents with zero variance at one level. Persistence of the effect is informative, but neither standardisation nor the subgroup sensitivity analyses eliminate the conceptual difference between a specific national referent and an aggregate regional referent.

The between-group comparison is the cleaner of the two contrasts because CEE-based and non-CEE-based respondents rate the same regional object. The non-CEE arm is small (n = 15), and the effect estimate (d = 1.02) should therefore be interpreted cautiously, but this analysis is less vulnerable to the aggregation issue.

The central finding should be reframed as differential attribution between own-country and aggregate regional assessments. Spatial optimism may be discussed as one possible explanatory framework, but the present design does not establish it as the underlying mechanism. The title-level framing, Abstract conclusions, Discussion and Conclusions should be revised accordingly.

2. The dispersion control analysis contains mathematical coupling.

Section 3.3 correlates each barrier's own-country mean (X) with its own-country-minus-region difference (X − Y), reporting r = 0.91 and R² = 0.83. X appears in both variables being correlated, creating mathematical coupling and potentially producing a substantial association by construction. This problem is well established in repeated-measurement analyses (Oldham, J Chronic Dis 1962;15:969–977; Bland and Altman, Lancet 1995;346:1085–1087).

Accordingly, corr(X, X − Y) cannot be interpreted as independent evidence that the magnitude of the own-country rating explains the country-region difference or supports the proposed compression mechanism.

The manuscript already contains methodologically cleaner evidence that regional ratings are less dispersed: the paired comparison of within-respondent SDs (0.99 vs. 0.67, p < 0.001, d_z = 0.91). That analysis directly addresses dispersion and is sufficient for this part of the argument.

The r = 0.91 statistic should therefore be removed or replaced. If the authors wish to examine whether rating magnitude is associated with the country-region difference, analysing or plotting the mean of the paired measurements against their difference, rather than either individual component against the difference, would avoid this specific coupling problem.

3. The impact-feasibility analysis aggregates away the repeated-measures structure.

The paper reports r = 0.21 (95% CI −0.32 to 0.64, p = 0.44) between expected impact and feasibility and concludes in the Abstract that “expected impact did not predict feasibility”. This correlation is calculated across 16 intervention-level means. The statistical unit for this analysis is therefore the intervention (N = 16), not the respondent.

The confidence interval extends from a moderately negative to a substantially positive association. The result is therefore compatible with no association but also with associations of potentially meaningful magnitude; it does not establish absence of association. In addition, the 16 interventions are a fixed list defined by the investigators rather than a random sample from a broader population of interventions, which further limits interpretation of the conventional Pearson inferential test.

Each of the 62 respondents rated both impact and feasibility for all 16 interventions, producing 62 × 16 paired ratings with a crossed respondent × intervention structure. A multilevel model accounting for respondent and intervention effects would provide a more appropriate and defensible estimate of the association. A repeated-measures correlation could also provide complementary information on the within-respondent relationship.

The authors should either provide an analysis that respects this repeated-measures structure or restrict the conclusion to what the present analysis demonstrates: across the 16 intervention-level mean ratings, the observed association between expected impact and feasibility was weak and statistically imprecise. The Abstract statement that “expected impact did not predict feasibility” should be revised regardless.

4. Sample composition and the operational definition of “expert” require clarification.

The panel is a convenience sample circulated through meeting participants and their professional networks, without a fixed sampling frame or calculable response rate. Romania accounts for 39 of 62 respondents (63%); only 15 respondents are based outside CEE; and excluding Romania leaves eight non-Romanian CEE respondents. These limitations are disclosed, and the subgroup sensitivity analyses are useful.

However, the manuscript repeatedly describes the sample as an “expert panel” without defining eligibility criteria or specifying what qualified respondents as experts for this exercise. Self-rated familiarity with formal vaccine-confidence frameworks was modest (median 2/5; 21 of 62 selected the lowest point). This does not imply that respondents lacked relevant practical expertise, but it does mean that the terminology requires justification. Please either specify the criteria used to define expertise or use a more neutral descriptor such as “stakeholder panel” or “professional panel”.

Stakeholder composition is also markedly uneven: civil-society and patient-organisation representatives account for 25 respondents, clinical HCPs for 21, academics for 6, public-health specialists for 5, public-authority staff for 2, and other roles for 3. Aggregate results may therefore disproportionately reflect the perspectives of the two largest constituencies. Stakeholder comparisons were feasible only for these two groups and were powered only to detect relatively large effects. This compositional imbalance should be discussed more explicitly.

5. The instrument was not externally piloted and its domain structure requires stronger justification.

The questionnaire underwent internal review only and was not externally piloted. Whether the 15 barriers, 16 interventions and geographic-level instructions were interpreted consistently across respondents from different countries and professional backgrounds is therefore unknown.

The four barrier domains were defined conceptually before data collection and were not empirically derived; no factor analysis was performed. If these domains function as formative indices, with items grouped conceptually rather than assumed to reflect a common latent construct, Cronbach's alpha does not validate their structure, and averaging items within domains for inferential testing requires a separate conceptual justification.

This is particularly relevant for the three-item Information & Literacy domain, which has an alpha of 0.53 and mean inter-item correlation of 0.29 at the own-country level. These values do not necessarily invalidate a conceptually constructed domain, but they reinforce the need to clarify what the domain scores are intended to represent and why averaging the constituent items is appropriate.

The complete survey instrument could not be identified in the material provided for review. For a study in which measurement validity depends heavily on the exact wording of items, geographic instructions, ranking prompts, actor categories and open-ended questions, the full instrument must be available for methodological appraisal. If it has not already been supplied separately, please include the complete questionnaire and the completed CROSS checklist as Supplementary Material.

6. Meeting exposure and the sister paper require careful separation.

Responses were collected between 8 May and 30 June 2026, straddling the 14 May meeting. The principal within-respondent contrast was similar in the earlier and later response windows, and the difference between windows was not statistically detectable (p = 0.83). However, response date is not equivalent to meeting exposure.

If actual attendance data are available, comparison of attendees and non-attendees would be more informative than a temporal split. If attendance data are unavailable, this should be stated as an additional limitation. A non-significant comparison between 37 and 25 respondents should not be interpreted as demonstrating that the meeting had no influence.

The relationship with the sister paper (Ref. 23) is otherwise handled appropriately: overlapping participant pools, common framing and the absence of independent replication are explicitly acknowledged. Findings originating from Ref. 23 should nevertheless remain clearly distinguishable from results generated by the present survey throughout the Discussion.

7. The ethics justification should identify the applicable institutional framework.

The manuscript argues that EU Regulation 536/2014 and the cited Romanian legislation governing medicinal products and patient care do not apply to this study. This may be correct, but demonstrating that clinical-trial legislation is inapplicable does not by itself establish that no institutional or national ethics framework governing research involving human participants required review or formal exemption.

Please identify the institutional policy or competent ethics body under which ethical review was considered unnecessary. If a waiver or exemption determination exists, it should be cited. Alternatively, if the relevant institution formally excludes anonymous professional-opinion surveys of this type from ethics review, that institutional policy should be stated explicitly. This would provide a stronger and more directly relevant justification than relying principally on legislation that does not apply to the study.

Minor comments

1. For the normalised Shannon entropy, , please specify the logarithm convention used for reproducibility. Provided the same base is used in numerator and denominator, the normalised value is invariant to the logarithm base; this is therefore a documentation issue rather than a mathematical error.
2. The standardised analysis in Section 3.3 excludes 12 respondents with zero variance at one geographic level. Please report their distribution by CEE/non-CEE status and, if feasible, stakeholder group, and briefly discuss whether their exclusion could influence the resulting estimate.
3. Figure 2 contains a residual “Chart Area” label on the left side. Please remove this artefact before publication.
4. The forced-choice ranking assigns five points for first place down to one for fifth place. Please state explicitly that this weighting is conventional and descriptive and that alternative weighting schemes could alter the precise ordering. Reporting both weighted score and top-five frequency, as currently done, is useful.
5. Table 2 is ordered by impact-feasibility gap. A supplementary presentation ordered separately by expected impact and by feasibility would facilitate comparison with Figure 4.
6. Please review the reference list carefully. Ref. 13 contains malformed DOI formatting, and journal abbreviations and other bibliographic elements are inconsistent across several entries.

The sentence around line 95 describing the sister paper is difficult to parse and should be simplified. 

Author Response

Dear Reviewer,

We are grateful for your judgement that rating each barrier at two geographic levels and each intervention on two dimensions are strong design choices, and for a review that identified two problems at the centre of the paper’s argument. We have accepted both; the changes, including to the title, are set out under comments 3.1 to 3.3. The complete instrument and the CROSS checklist are now supplied, and the ethics question has been settled by a formal determination, described under comment 3.7. Our responses follow your comments in order, major comments first; all changes are highlighted in the revised manuscript.

Comment 3.1 (major)
The central interpretive claim goes further than the design allows. The own-country/region contrast is not conceptually identical for all respondents, and the data do not show that respondents attribute structural problems to “elsewhere”. The finding should be reframed, and the title, Abstract, Discussion and Conclusions revised.

Response: Accepted in full, including at title level. The title is now “Trust, Access and the Division of Responsibility for Vaccine Confidence in Central and Eastern Europe: Paired Own-Country and Regional Ratings from a Multi-Stakeholder Professional Panel”, which names the design rather than an interpretation and no longer asserts where observers place the problem. The sentence that closed Section 3.3 (“trust problems are placed where the observer stands, and structural and access problems elsewhere”) has been replaced by a statement that assessments differ by barrier type according to the geographic referent and that the data do not show where respondents locate each problem. The Discussion paragraph on the asymmetry has been rewritten: it no longer presents comparative optimism as the mechanism but as one of three compatible explanations, alongside a division of responsibility between levels and coarser discrimination about the aggregate, between which the design cannot choose; the Introduction, the Abstract and the Conclusions have been aligned with it. The Limitations state the aggregation problem in your own terms: for the 47 CEE-based respondents the regional referent contains the national one, for the 15 based outside it the two are separate places, and pooling them treats two different contrasts as estimates of one quantity; we add that a Romanian respondent comparing Romania with CEE as a whole is not comparing Romania with the rest of CEE, and that the between-group contrast is the cleaner of the two on this point. Three phrases elsewhere that still spoke of where respondents place or locate each kind of problem (Introduction, Section 3.3 and Section 4.2) have been reworded to the same effect.

Comment 3.2 (major)
The dispersion control analysis contains mathematical coupling: corr(X, X − Y) with r = 0.91 cannot be interpreted as independent evidence. The statistic should be removed or replaced.

Response: Accepted, and we are grateful for it. We tested the point before acting on it: under a null model in which the own-country and regional item means are independent with the observed variances, the expected value of corr(X, X − Y) is +0.87, and about one random pairing in ten reaches or exceeds the published 0.909. The statistic therefore carried almost no information beyond its own construction. It has been deleted from Section 3.3. As you note, the paired comparison of within-respondent standard deviations (0.99 versus 0.67; paired t(61) = 7.20, p < 0.001, d_z = 0.91) addresses dispersion directly and is sufficient for this part of the argument; it is retained. For completeness, the Bland–Altman formulation you suggest, corr((X + Y)/2, X − Y), gives r = 0.85 on these data; we have added it in the text, since it is the formulation you named, without an accompanying figure, because Reviewer 2 asks for fewer display items. We also say in the text what this statistic can and cannot show: because the own-country item means are more dispersed than the regional ones, the mean–difference correlation has an expected value of about 0.5 under independence, so the observed 0.85 restates the compression rather than providing separate evidence about attribution; the Methods now describe this formulation in place of the withdrawn one.

Comment 3.3 (major)
The impact–feasibility analysis aggregates away the repeated-measures structure. A multilevel model would be more defensible, and the Abstract statement that expected impact did not predict feasibility should be revised regardless.

Response: Accepted. Section 3.4 now reports the intervention-level correlation with its confidence interval and states explicitly that the unit of analysis is the intervention, that the interval spans values from a moderate negative to a substantial positive association, and that the 16 interventions are a fixed list rather than a sample. It then reports the analysis on the 992 paired ratings: a linear mixed model with crossed random intercepts for respondent and intervention, feasibility regressed on expected impact and fitted by restricted maximum likelihood (slope 0.40, SE 0.03, 95% CI 0.33–0.46, p < 0.001, with the variance components reported), and a repeated-measures correlation of r = 0.33 (95% CI 0.27–0.39). The two dimensions share about a tenth of their variance, so they are related without being interchangeable. Methods 2.3 describes both analyses and the software used. The Abstract claim has been removed, and the sentences in Section 3.4, the Discussion and the Conclusions that described the two axes as unrelated or independent have been reworded so that the text claims only that the panel-level orderings partly diverge, which is what the quadrant analysis rests on.

Comment 3.4 (major)
The sample composition and the operational definition of “expert” require clarification; either specify eligibility criteria or use a more neutral descriptor.

Response: We have taken the second option throughout, including in the title. Methods 2.1 now states that eligibility was defined by professional role rather than academic credential, that no threshold of seniority, publication record or formal training was applied and no respondent was screened out, that there was no sampling frame so no response rate can be calculated, and that the panel is described as a professional or stakeholder panel rather than an expert panel in any credentialled sense. The uneven composition is now flagged where the numbers are given, and the Limitations discuss it.

Comment 3.5 (major)
The instrument was not externally piloted, the domain structure needs stronger justification, and the complete instrument and CROSS checklist should be supplied as Supplementary Material.

Response: The complete instrument is now Supplementary File S1, reproduced item by item with every response option and scale anchor, every grid row, the section introductions and the geographic instructions as administered; the completed CROSS checklist is Supplementary File S2. On the domains, we now state that they are conceptual groupings fixed before data collection, that their items were chosen because they name related policy problems rather than because they were expected to load on a common latent variable, and that they are therefore closer to formative indices than to reflective scales, which is why no factor analysis was undertaken and why alpha describes how closely the items happened to move together rather than validating the grouping. We add that averaging within a domain is justified as a summary of its items alone, and that this is most consequential for the three-item information and literacy domain, whose alpha of 0.53 indicates its items behave fairly independently.

Comment 3.6 (major)
Response date is not equivalent to meeting exposure; a non-significant comparison between 37 and 25 respondents should not be read as showing the meeting had no influence.

Response: Agreed. Attendance was not recorded, and responses were collected without identifiers, so the comparison cannot be reconstructed. Methods 2.1 now states that response date is a proxy for exposure and not a record of it: respondents who answered after 14 May cannot be assumed to have attended, and those who answered before cannot be assumed not to have been exposed to the meeting agenda. The sentences that described respondents as having attended or not attended, in Methods 2.1 and in the Limitations, have been reworded to refer only to the date of response, and the Limitations retain the statement that the meeting may have shaped later answers in ways this design cannot separate.

Comment 3.7 (major)
The ethics justification should identify the institutional framework under which review was considered unnecessary, and cite any waiver or exemption determination.

Response: Done. The Research Ethics Committee of the National Institute for Health Services Management (INMSS, Bucharest), with which one of the authors is affiliated, reviewed the study, the full instrument with its consent text and an extended summary, and issued a determination of exemption (No. 3120 of 11 September 2026, meeting of 10 September 2026). It confirms that the study falls within the category of anonymous, voluntary, non-interventional surveys of professional opinion that are not subject to prior review under the applicable legal framework (Law No. 206/2004 on good conduct in scientific research, Regulation (EU) 2016/679 and Law No. 190/2018), finds no ethical objection to the consent procedure or to the handling of the free-text role field, and states expressly that the assessment was requested after data collection and does not constitute prior approval; we report it in those terms. Section 2.4 now names the institutional framework and cites the determination, the Institutional Review Board Statement has been rewritten accordingly, and the determination is available to the editorial office and to yous on request.

Minor comment 1
Specify the logarithm convention used for the normalised Shannon entropy.

Response: Methods 2.3 now states that the entropy is calculated with natural logarithms and divided by log 5 so that the result runs from 0 to 1.

Minor comment 2
Report the distribution of the 12 respondents excluded from the standardised analysis and discuss whether their exclusion could influence the estimate.

Response: Section 3.3 now reports it. The twelve comprise nine CEE-based respondents, all in Romania, and three based outside the region (19% of the CEE-based group and 20% of the non-CEE group), so the exclusion is close to proportionate and does not favour either side of the between-group contrast. They span five stakeholder categories. In nine of the twelve only the regional ratings carried no variance; the other three rated every item identically at both levels and therefore contribute no own-country-minus-region difference to the unstandardised analysis either. No respondent had flat own-country ratings with varying regional ratings, which is itself consistent with reduced discrimination about the aggregate object. On whether the exclusion could influence the estimate: the twelve are also the respondents with the weakest raw asymmetry (mean interaction +0.15, against +0.50 among the fifty retained), so the standardised estimate describes the fifty respondents whose ratings varied at both levels; because standardisation is undefined for them no re-estimate including them is possible, and the text now says that the raw and standardised results should be read together.

Minor comment 3
Figure 2 contains a residual “Chart Area” label on the left side.

Response: The label was an on-screen tooltip captured when the chart was exported from Excel. It has been removed from the image; the figure is otherwise unchanged.

Minor comment 4
State that the 5-to-1 forced-choice weighting is conventional and that alternative schemes could alter the ordering.

Response: Added to Methods 2.3: the weighting is described as conventional and descriptive rather than derived from any model of preference intensity, a different scheme could alter the precise ordering, and both the weighted score and the top-five frequency are reported so that the ranking can be read without relying on the weights.

Minor comment 5
A supplementary presentation of Table 2 ordered by expected impact and by feasibility would help comparison with Figure 4.

Response: Supplied as Supplementary Tables S2 and S3, ordering the sixteen interventions by expected impact and by feasibility respectively, each with the impact–feasibility gap and the quadrant classification against the two medians used in Figure 4; both are cited in Section 3.4.

Minor comment 6
Reference 13 contains malformed DOI formatting and bibliographic elements are inconsistent across entries.

Response: Reference 13 read “https://doi:10.1016/j.vaccine.2015.04.036”, a hybrid of the two DOI forms; it and reference 11 now use https://doi.org/. Further corrections are listed under comment 1.16. The Crossref check of all 57 entries is described under checklist item (I); the remaining inconsistencies in journal abbreviations and author initials are corrected.

Minor comment 7
The sentence around line 95 describing the sister paper is difficult to parse and should be simplified.

Response: Simplified. The sentence now reads: “The two studies share a framing event and an overlapping pool of participants, but use different instruments, data and analytical questions. Section 4.2 explains why neither can be treated as independent corroboration of the other.”

We thank you again for the time and care you gave to the manuscript.

The authors

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author resolved my confusion and suggested publishing the manuscript.

Author Response

Thank you for your time and contribution for making the manuscript more clear and informative!

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for the revision and the detailed response to my comments. The changes substantially improve the methodological rigor and the interpretation of the results, and they address my main concerns satisfactorily. Two points remain before I can recommend acceptance:

1. Clarify in Methods the distinction between the primary interaction test and the subgroup analyses (CEE-based, non-CEE-based, Romanian, and Romania-excluded), stating which were pre-specified and how multiplicity was handled. If these are sensitivity analyses, the text should avoid implying they constitute independent confirmation of the finding.
2. Add the Benjamini–Hochberg adjusted values to Table S1, alongside the unadjusted p-values, for the fifteen item-level comparisons. This makes the minimum value reported in the text (q = 0.051) verifiable. The table note should identify the set of comparisons to which the correction was applied.

Once these two points are addressed, the manuscript is suitable for publication.

Thank you for your work and for the attention given to my observations.

Author Response

Dear Reviewer,

Thank you for the second reading. Both points have been addressed; the new text is marked as tracked changes in the revised manuscript, the yellow of the first round being retained.

Comment 1 (primary test versus subgroup analyses, pre-specification, multiplicity)

Response: The subgroup analyses are sensitivity analyses, and the manuscript now says so. The last paragraph of Section 2.3 has been rewritten to distinguish three levels of inference: the two pre-specified primary tests (the within-respondent difference-of-differences interaction and the between-group contrast on the regional ratings), each a single planned comparison reported once without adjustment; the re-estimation of these contrasts among CEE-based, non-CEE-based and Romanian respondents and with Romania excluded, which was not pre-specified but planned as a sensitivity analysis once the composition of the panel was known (63% Romanian), on overlapping and nested subsets of the same sample, without multiplicity adjustment and not treated as confirmation of the finding; and the exploratory level, the fifteen per-item comparisons assessed as one family under the Benjamini–Hochberg correction. The wording elsewhere has been aligned: Section 2.1 now speaks of sensitivity analyses rather than replication; in Section 3.3 “It held in every subgroup examined” now reads that the interaction kept its direction and a similar size in the sensitivity analyses, with the four estimates and the statement that these are overlapping subsets of the same panel, not independent replications; the two Discussion sentences that described the result as the one that held “most consistently across subgroups” no longer offer subgroup consistency as evidence; and the Limitations now say that the re-estimated results kept their direction in overlapping subsets of one panel.

Comment 2 (Benjamini–Hochberg adjusted values in Table S1, with the family identified)

Response: Done, in a new Table S2 placed immediately after Table S1, which we have kept as the table of ratings. Table S2 lists the fifteen items in the order of Table S1 and gives, for each, the mean own-country and regional ratings, the mean within-respondent difference, the two-sided Wilcoxon signed-rank p-value, the rank-biserial correlation and the Benjamini–Hochberg adjusted value (q). The table note states that the correction was applied across these fifteen comparisons only, as one family, that no comparison reaches q < 0.05 (smallest q = 0.051, for C1.2 and C3.2), and that the primary difference-of-differences test is a single pre-specified comparison outside this family. Section 3.3 refers to Table S2 where the smallest q is quoted, and the Methods state that the adjusted values are given there alongside the unadjusted ones. The former Tables S2 to S4 are now S3 to S5, and the references in Sections 3.4 and 3.9 and in the Supplementary Materials statement have been renumbered.

Thank you again for the care given to the manuscript in both rounds.