Mpox Vaccination Willingness, Concern Profiles, and Associated Factors Among Men Who Have Sex with Men in Changsha, China

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The study is somewhat interesting.

In Materials and methods:

• Authors should define the following variables: education level, marital history, employment status, and local hukou status
• Authors should include information on how they calculated the required sample size for the study. This is essential.

In Table 1 Authors should specify the currency used to calculate income category.

In Discussion:

• Why didn’t the authors of this study base their classification of HIV on laboratory tests as well, in addition to the participants’ own reports?
• Authors should clarify whether the Mpox vaccine is offered free of charge to men who have sex with men (MSM) in the region under study, and how this might affect their willingness to get vaccinated against Mpox
• Authors should expand the discussion regarding the possibility that the study’s findings may be generalizable to MSM populations in other geographic or social contexts.
• Why didn’t the authors, instead of studying reported willingness to get vaccinated, analyze one group of MSM who had been vaccinated against Mpox and another group of MSM who had not been vaccinated??
• Authors should discuss the extent to which the compensation provided by the authors for the respondents of the study subjects (50 RMB for participation) may have influenced their participation

Author Response

Comments 1: Authors should define the following variables: education level, marital history, employment status, and local hukou status.

Response 1: Thank you for this helpful suggestion. We have now added Supplementary Table S2, which provides the original survey wording and coding rules for all descriptive variables listed in main text Table 1, including education level, marital history, employment status, and local hukou status. We have also added a cross-reference to this table in the Method section (Main text: Page 4 Line 170-175). 

 

Comments 2: Authors should include information on how they calculated the required sample size for the study. This is essential.

Response 2: Thank you for this important comment. We have now provided the full calculation process in Supplementary Material Section 1 (“RDS Sample Size Estimation”). Because MSM constitute a hard-to-reach population without a conventional sampling frame, a context in which RDS is commonly used (Heckathorn, 1997), sample size planning could not rely on assumptions of independent simple random sampling alone. Instead, a base sample size for proportion estimation was first obtained using the conventional single-proportion formula and then adjusted by a design effect to reflect the network-based dependence inherent in RDS, consistent with the RDS methodological literature on design effects and sample size planning (Salganik, 2006).

Specifically, we used a conservative prevalence assumption of 50% for the primary outcome (immediate willingness to receive mpox vaccination), which yields the largest base sample size for a given precision level. Using a 95% confidence level, a pragmatic margin of error of 7%, and a design effect of 2.0, the target sample size was estimated to be approximately 392 participants. The final study included 405 eligible participants who were verified, enrolled, and completed the survey, meeting this target.

We have also clarified in Supplementary Material Section 2 (“RDS Recruitment and Survey Implementation Process”) that questionnaires were completed in person under one-to-one guidance from trained study staff after eligibility verification, so that the final sample of 405 represents participants who were verified, enrolled, and completed the survey under supervised, in-person administration. In addition, we have previously clarified in Supplementary Material Section 4 (“RDS Equilibrium and Degree Diagnostics”) that recruitment diagnostics were examined for the primary outcome, and that the estimated distribution of immediate vaccination willingness became progressively stable across later recruitment waves, providing further support that the achieved sample was adequate within the sampled network structure.

 

Comment 3. In Table 1 Authors should specify the currency used to calculate income category.

Response 1: Thank you for this helpful suggestion. We have now specified the currency (RMB) in Table 1 (Main text: Page 9) for the income categories.

 

Comment 4. Why didn’t the authors of this study base their classification of HIV on laboratory tests as well, in addition to the participants’ own reports?

Response: Thank you for this important comment. We agree that laboratory-based HIV classification may provide a more objective measure in some study settings. However, the present study differed fundamentally from our previous survey and from clinic-based service settings in which laboratory confirmation was integrated into participant flow.

This RDS-based study recruited MSM through peer referral in the community, rather than enrolling individuals who were actively seeking HIV confirmatory testing services. In our previous 2023 survey, participants were drawn from our institution, where free HIV confirmatory testing and post-confirmation follow-up services were provided. Many participants in that study had already entered the HIV confirmation pathway, making laboratory-based classification both feasible and appropriate in that context. By contrast, in the present study, participants were recruited proactively through peer networks, and HIV testing was not a mandatory or standardized component of study enrollment. Although free screening was offered on a voluntary basis, requiring laboratory confirmation for analytic classification would have substantially increased participation barriers, altered the recruitment process, and changed the study population to one defined partly by willingness to undergo HIV serological testing or to authorize access to prior HIV testing or infection information at the time of peer referral. For this reason, HIV classification in the present study was based on self-report, which was more compatible with the community-based RDS design and allowed us to preserve recruitment feasibility and inclusiveness. Nevertheless, we acknowledge in the revised Limitations section (Main text: Page 20 Line 693-702) that self-reported HIV status may be subject to underreporting or nondisclosure in a sensitive social context.

 

Comment 5. Authors should clarify whether the Mpox vaccine is offered free of charge to men who have sex with men (MSM) in the region under study, and how this might affect their willingness to get vaccinated against Mpox

Response: Thank you for this important comment. At the time of this study, no commercially available mpox vaccine was available on the Chinese mainland, and China’s mpox prevention and control strategy still relied primarily on non-vaccine measures, although vaccine development and clinical trials were ongoing. We have explicitly clarified this policy context in the revised Introduction section (Main text: Page 3 Lines 104-109). Accordingly, whether mpox vaccination would be offered free of charge to MSM in the study region had not yet been established in practice.

The present study therefore assessed willingness to receive mpox vaccination under a context of future potential vaccine availability. We agree that vaccine cost and access could influence future uptake once vaccines become available, but these factors could not be evaluated here as actual programmatic conditions because no routine vaccination program was yet in place. In addition, perceived burden, including cost-related concerns, was already captured within the concern-profile analysis rather than modeled as a separate implementation variable.

 

Comment 6. Authors should expand the discussion regarding the possibility that the study’s findings may be generalizable to MSM populations in other geographic or social contexts.

Response: Thank you for this important comment. We agree that the external validity of a single-city RDS study should be discussed more explicitly. In the Limitations section, we have acknowledged that the study was conducted in a single city and that transferability to MSM populations in other geographic, social, or service contexts may therefore be limited, and that socially isolated individuals or those weakly connected to peer networks may be less well captured in RDS-based studies. In the revised Limitation section (Main text: Page 20 Lines 676-681,687-692), we further note that regional variation in vaccination-related attitudes and psychosocial barriers across mainland China has not been well characterized, and that socially isolated subgroups may require additional community-based outreach, targeted field investigation, or dedicated studies specifically designed for their inclusion.

 

Comment 7. Why didn’t the authors, instead of studying reported willingness to get vaccinated, analyze one group of MSM who had been vaccinated against Mpox and another group of MSM who had not been vaccinated??

Response: Thank you for this important comment. At the time of this study, no commercially available mpox vaccine was available on the Chinese mainland, and no routine publicly accessible vaccination program for MSM was in place. Therefore, it was not feasible to define study groups based on actual mpox vaccination status in the local setting. In this context, assessing reported willingness to receive mpox vaccination was the more appropriate and policy-relevant research question, because it provides evidence on potential acceptance before vaccine implementation becomes available.

We also note that any very limited vaccination received through non-routine, unofficial, or overseas channels would represent a substantially different research question, reflecting access to exceptional vaccine pathways rather than uptake under a local public health program. For this reason, the present study focused on willingness rather than realized vaccination status.

 

Comment 8. Authors should discuss the extent to which the compensation provided by the authors for the respondents of the study subjects (50 RMB for participation) may have influenced their participation.

Response: Thank you for this helpful comment. We have clarified the compensation arrangement in the revised Supplementary Material Section 2 (“RDS Recruitment and Survey Implementation Process”). Participants received 50 RMB for completing the survey, and an additional 10 RMB recruitment incentive for each successfully referred peer who met eligibility criteria and completed the survey, for up to three peers. Thus, the maximum total compensation available to a participant was 80 RMB. We have updated the Informed Consent Statement section (Main text: Page 21 Line 746-751) accordingly.

In RDS, modest primary and secondary incentives are a standard component of study implementation to support participation and peer recruitment in hard-to-reach populations. We acknowledge that such compensation may influence willingness to participate. However, in this study the incentive structure was implemented as part of the standardized sampling and recruitment procedure rather than as an exposure or analytic variable of interest. As this arrangement has been clarified in the revised Supplementary Materials, with relevant references added, we did not discuss it further as a substantive explanatory factor in the main Discussion section.

 

Additional Clarifications

Following the reviewers’ suggestions, several new materials have been added. In the main text, Table 2 has been replaced with a heatmap (Figure 1). The two forest plots have also been revised and updated, and are now presented as Figure 2 and Figure 3. Accordingly, the main text now contains one table and three figures.

In the Supplementary Materials, three new sections have been added: (1) RDS Sample Size Estimation; (2) RDS Recruitment and Survey Implementation Process; and (3) Modelling Strategy Overview and Supplementary Technical Details. In addition, two new supplementary tables have been included: Supplementary Table S2 (Original Survey Questions and Coding Rules for Categorical and Continuous Variables) and Supplementary Table S5 (Mean Concern-item Scores across Candidate PAM Solutions and Ward Hierarchical Clustering in the Delayed/Refused Responses Subgroup). Four new supplementary figures have also been added: Supplementary Figure S1 (RDS participant flow and survey implementation process), Supplementary Figure S5 (Conceptual framework linking mpox awareness, psychosocial dimensions, prevention-related behaviors, and vaccination willingness), Supplementary Figure S6 (Average Silhouette Width across Candidate PAM Solutions [k = 2–4] for the Delayed/Refused Responses Subgroup), and Supplementary Figure S7 (Schematic overview of the analytic strategy).

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

Thanks for submitting this invaluable study.

The study covers an important side in vaccination against MPox.

Here are my concerns that need your attension:

1. Potential Sampling Bias: Respondent-driven sampling (RDS) enhances coverage relative to convenience sampling; however, it may inadequately represent socially isolated MSM or individuals not engaged in active peer networks.
2. Self-Reported Data: Key measures, such as sexual behavior, HIV status, mpox awareness, and vaccination willingness, were self-reported, which may introduce recall bias or underreporting, especially for sensitive topics like HIV status.
3. Limited Sample Size for Subgroup Analysis: The exploratory multinomial regression analysis within the non-immediate willingness subgroup was based on a modest sample size, reducing statistical power and precision for subgroup-specific estimates.

Thanks

Author Response

Comment 1. Potential Sampling Bias: Respondent-driven sampling (RDS) enhances coverage relative to convenience sampling; however, it may inadequately represent socially isolated MSM or individuals not engaged in active peer networks.

Response: Thank you for this important observation. We agree that RDS does not eliminate all sampling bias, and that socially isolated MSM or individuals weakly connected to peer networks may remain underrepresented. We have therefore acknowledged in the Limitations section that, although RDS may improve coverage relative to conventional convenience-based approaches, it remains dependent on underlying peer-network connectivity and may underrepresent MSM who are socially isolated, weakly connected to recruitment chains, or otherwise outside the active peer networks through which recruitment occurred. We now also clarify more explicitly in the revised Limitations section (Main text: Page 20 Lines 682–693) that, in the absence of a conventional sampling frame, these subgroups may be difficult to capture through alternative recruitment strategies as well, and may require additional community-based outreach, targeted field investigation, or dedicated studies focusing specifically on more weakly connected or marginalized subgroups.

 

Comment 2. Self-Reported Data: Key measures, such as sexual behavior, HIV status, mpox awareness, and vaccination willingness, were self-reported, which may introduce recall bias or underreporting, especially for sensitive topics like HIV status.

Response: Thank you for this important comment. We agree that self-reported data may still be subject to reporting bias, including recall error for some measures and social desirability bias or underreporting for more sensitive topics. However, the present survey was not administered as an unsupervised self-completed questionnaire. Instead, all eligible participants were verified and completed the survey in person under one-to-one guidance from trained study staff. We have clarified this in Supplementary Material Section 2 (“RDS Recruitment and Survey Implementation Process”) to improve transparency. During survey administration, when respondents had difficulty understanding an item, clarification could be provided in real time, and sufficient time was allowed for survey completion. These procedures were intended to improve comprehension, reduce item-level misunderstanding, and strengthen data quality under supervised, in-person administration.

With respect to HIV-related variables in particular, we agree that laboratory-based HIV classification may provide a more objective measure in some study settings. However, the present study differed fundamentally from clinic-, service-, or follow-up-based settings in which HIV status is already embedded in routine testing or management pathways. This RDS-based study recruited MSM through peer referral in the community, rather than enrolling individuals who were actively seeking HIV confirmatory testing or who had already entered HIV care or follow-up pathways in which infection status was already known to both participants and service providers. HIV testing was therefore not a mandatory or standardized component of enrollment in the present study. Importantly, in this context, the key concern for HIV status is less recall error than possible underreporting or nondisclosure in a sensitive social setting. Any discrepancy between laboratory-confirmed infection and self-reported HIV status in this study would therefore be more likely to reflect disclosure behavior than memory failure. Although free HIV screening was offered on a voluntary basis, requiring laboratory confirmation for analytic classification would have increased participation barriers, altered the recruitment process, and shifted the study population toward individuals willing to undergo HIV serological testing or to authorize access to prior HIV testing or infection information at the time of peer referral. In other words, HIV disclosure or HIV ascertainment would effectively have become part of study participation itself, likely introducing additional disclosure-related selection bias on top of the peer-driven recruitment process.For this reason, HIV classification in the present study was based on self-report and recent testing behavior, which was more compatible with the community-based RDS design.

At the same time, we acknowledge in the revised Limitations section (Main text: Page 20 Line 693-702) that, despite these quality-control measures, self-report bias cannot be fully excluded.

 

Comment 3.Limited Sample Size for Subgroup Analysis: The exploratory multinomial regression analysis within the non-immediate willingness subgroup was based on a modest sample size, reducing statistical power and precision for subgroup-specific estimates.

Response: Thank you for this important comment. We agree that the multinomial regression within the non-immediate willingness subgroup should be interpreted cautiously. Although this subgroup included approximately 200 participants, the effective information available for estimation was more limited after division into multiple concern-profile categories and inclusion of predictors with several categorical levels. We therefore examined sparsity prior to modeling and recoded selected variables with sparse categories to reduce instability and the risk of quasi-complete separation. In particular, sexual orientation was collapsed to homosexual versus non-homosexual, and HIV-related behavioral status was recoded into three categories, including a combined sparse category for HIV-positive participants and those who preferred not to disclose HIV status. These recoding steps have already described in the Supplementary Material Section 9 (“Recoded Variables in Multinomial Regression for Concern-Profile Membership). Even with these precautions, some subgroup-specific estimates remained imprecise, as reflected in wider confidence intervals. We have therefore clarified more explicitly in the revised Limiations section (Main text: Page 20 Line 702-713) that this multinomial analysis was exploratory and hypothesis-generating rather than definitive.

We have further revised the relevant text in the Introduction (Main text: Page 3 Line 115-116),  Methods (Main text: Page 5 Line 195), Results (Main text: Page 10 Line 398, 403) and Discussion (Main text: Page 19 Line 632, 637-638) sections to explicitly clarify that the clustering analysis was exploratory and hypothesis-generating rather than confirmatory.

 

Additional Clarifications

Following the reviewers’ suggestions, several new materials have been added. In the main text, Table 2 has been replaced with a heatmap (Figure 1). The two forest plots have also been revised and updated, and are now presented as Figure 2 and Figure 3. Accordingly, the main text now contains one table and three figures.

In the Supplementary Materials, three new sections have been added: (1) RDS Sample Size Estimation; (2) RDS Recruitment and Survey Implementation Process; and (3) Modelling Strategy Overview and Supplementary Technical Details. In addition, two new supplementary tables have been included: Supplementary Table S2 (Original Survey Questions and Coding Rules for Categorical and Continuous Variables) and Supplementary Table S5 (Mean Concern-item Scores across Candidate PAM Solutions and Ward Hierarchical Clustering in the Delayed/Refused Responses Subgroup). Four new supplementary figures have also been added: Supplementary Figure S1 (RDS participant flow and survey implementation process), Supplementary Figure S5 (Conceptual framework linking mpox awareness, psychosocial dimensions, prevention-related behaviors, and vaccination willingness), Supplementary Figure S6 (Average Silhouette Width across Candidate PAM Solutions [k = 2–4] for the Delayed/Refused Responses Subgroup), and Supplementary Figure S7 (Schematic overview of the analytic strategy).

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript (“Mpox vaccination willingness, concern profiles, and associated factors among men who have sex with men in Changsha, China”, Manuscript ID: vaccines-4241684) by Zhou et al. presents a well-designed cross-sectional study using respondent-driven sampling (RDS) to examine mpox vaccination willingness among MSM in Changsha, China. The study goes beyond a binary framing of vaccine acceptance by distinguishing immediate vs. non-immediate willingness and further characterizing heterogeneity within the non-immediate subgroup via clustering analysis. The integration of behavioral, psychosocial, and network-informed modeling approaches is a notable strength. Overall, the study is methodologically sophisticated, clearly written, and potentially impactful, particularly for informing targeted vaccination strategies in key populations. However, several issues related to clarity, justification of analytic choices, interpretation, and generalizability should be addressed before publication.

1. The decision to combine delayed and refused vaccination into a single “non-immediate” category is understandable but conceptually debatable. These groups may differ meaningfully in psychology, access barriers, and intervention needs. Please provide a stronger justification for this categorization. Additionally, please consider (1) sensitivity analyses separating delay vs. refusal, OR (2) explicitly framing this as a pragmatic rather than conceptual grouping.

 

2. While RDS diagnostics are reported, the manuscript may overstate representativeness. RDS assumptions (e.g., random recruitment within networks, accurate degree reporting) are rarely fully met. However, there is a limited discussion of potential bias from seed selection, homophily beyond the tested variables, and undercoverage of isolated MSM. Please temper claims about representativeness. Expand limitations to explicitly discuss RDS assumptions and violations, and implications for external validity.

 

3. The multi-model approach (logistic regression, RDS-weighted, GEE, network-lag, elastic net) is impressive but risks overcomplexity relative to sample size (n = 405). Potential overfitting, especially with (1) multiple interaction terms, (2) penalized selection and forced variables, and (3) limited discussion of model stability. Please clarify the events-per-variable considerations, whether shrinkage/validation diagnostics were assessed, and consider simplifying or clearly justifying why all models are necessary.

 

4. The use of interaction terms between awareness and PCA-derived scores is innovative but difficult to interpret. Results are described as “informative” but not always clearly translated into real-world meaning. Also, the inverse association of perceived transmission likelihood is counterintuitive and underexplained. Please, provide a clearer interpretation of (1) What does a unit increase in these interaction terms represent in practice? (2) Why might higher perceived transmission likelihood reduce immediate willingness?

 

5. The clustering of non-immediate respondents is a key contribution but requires stronger methodological transparency. There is limited justification for (1) the choice of the PAM algorithm, (2) the selection of 3 clusters, and (3) stability and reproducibility are not discussed. Please consider including (1) cluster validation metrics (e.g., silhouette width), (2) sensitivity analyses (if conducted), and (3) clarify whether clustering is exploratory vs. confirmatory.

 

6. Although the authors acknowledge the cross-sectional design, some interpretations imply causal pathways (e.g., “PrEP use reflects prevention orientation influencing vaccination”). Please consistently use non-causal language such as “associated with” instead of “reflects” or “indicates”. Additionally, please strengthen caution in the ‘Discussion’ section.

 

7. The study is conducted in a single city (Changsha). Findings may not generalize to (1) other regions of China, (2) rural MSM populations, or (3) different healthcare access contexts. Please expand the discussion on (1) regional variability and (2) cultural and healthcare system differences.

 

8. Some sections (especially Methods 2.3) are overly dense and technical. Please consider (1) adding schematic diagrams for the modeling strategy, (2) moving technical details to the Supplementary Material.

 

9. “Non-immediate willingness” is somewhat awkward. Consider alternatives for “Delayed or unwilling”, “Non-urgent willingness”. Please define clearly at first use and maintain consistency.

 

10. Table 2 is informative but dense. Please highlight key distinguishing features across clusters. Additionally, consider a visual heatmap for concern profiles.

 

11. STROBE compliance is mentioned but could be improved by (1) explicitly reporting missing data handling and (2) providing a participant flow diagram.

 

12. Minor grammatical improvements are needed; for example, “may have provided broader coverage” should be changed to “may provide broader coverage”. Occasional repetition in the ‘Discussion’ section could be reduced.

 

13. Add a conceptual framework linking awareness, psychosocial factors, behaviors, and willingness.

 

14. Please consider discussing implications for (1) digital health campaigns, (2) MSM community-based interventions, and (3) a brief comparison with global mpox vaccination trends could strengthen context.

 

The manuscript is strong and potentially publishable, but requires clarification of key methodological choices, improved interpretability, and more cautious framing of findings before acceptance.

Author Response

Comment 1.The decision to combine delayed and refused vaccination into a single “non-immediate” category is understandable but conceptually debatable. These groups may differ meaningfully in psychology, access barriers, and intervention needs. Please provide a stronger justification for this categorization. Additionally, please consider (1) sensitivity analyses separating delay vs. refusal, OR (2) explicitly framing this as a pragmatic rather than conceptual grouping.

Response: Thank you for this important comment. We agree that delayed and refused vaccination are not necessarily identical in their underlying psychology or intervention implications. We have therefore revised the manuscript to make it clearer that this was a pragmatic rather than a purely conceptual grouping in the primary analysis.

We have added clarifying text in the Methods section (Main text: Page 4 Lines 170-175) and Supplementary Table S2, which provides the original survey wording and coding rules for all descriptive variables listed in main text Table 1, including the exact wording of the vaccination willingness item. The outcome item asked: “If an mpox vaccine were available, would you be willing to receive it?” with response options of (1) immediate vaccination, (2) wait and see before deciding (delay vaccination), and (3) refuse vaccination. In this wording, option (2) was included to capture respondents whose first reaction was neither immediate acceptance nor clear refusal, rather than to define a fully distinct conceptual endpoint. It indicates deferred or withheld acceptance at the time of response rather than confirmed future uptake, while option (3) indicates refusal at the time of response, but not necessarily an immutable future position. Thus, both options (2) and (3) represented lack of immediate willingness at the time of the survey, whereas option (1) captured immediate readiness to vaccinate.

We have also clarified that this pragmatic categorization is broadly consistent with established vaccine hesitancy frameworks, including the WHO SAGE definition and the 5C model, in which delayed acceptance and refusal are both recognized within the scope of hesitancy, while still potentially reflecting overlapping but non-identical psychological antecedents. As discussed in the Introduction (Main text: lines 77-81) and Discussion (Main text: Page 19 Line 668-671), we therefore did not treat the delayed/refused group as fully homogeneous. Instead, we further examined variation within this group through concern-profile analysis and exploratory multinomial regression.

 

Comment 2.While RDS diagnostics are reported, the manuscript may overstate representativeness. RDS assumptions (e.g., random recruitment within networks, accurate degree reporting) are rarely fully met. However, there is a limited discussion of potential bias from seed selection, homophily beyond the tested variables, and undercoverage of isolated MSM. Please temper claims about representativeness. Expand limitations to explicitly discuss RDS assumptions and violations, and implications for external validity.

Response: Thank you for this important comment. We agree that RDS diagnostics should not be interpreted as guaranteeing representativeness. We have therefore tempered the relevant wording throughout the revised manuscript, including not only statements related to coverage and representativeness, but also broader interpretive language in the Discussion and Conclusion (Main text: Page 17-21 Line 542, 575, 601, 602, 632, 638, 657-659, 682-692, 702-714, 719, 723, 727-728) where the original wording might have been read as implying stronger inference or generalizability than warranted. In the revised Limitations section (Main text: line 682-692), we now state explicitly that standard RDS diagnostics and our additional assessments can provide useful information regarding recruitment dynamics for the measured variables, but cannot verify all RDS assumptions. And residual bias due to self-reported network size, seed dependence, unmeasured homophily, and undercoverage of socially isolated or weakly connected MSM cannot be excluded. These limitations should be considered when interpreting external validity.

 

Comment 3.The multi-model approach (logistic regression, RDS-weighted, GEE, network-lag, elastic net) is impressive but risks overcomplexity relative to sample size (n = 405). Potential overfitting, especially with (1) multiple interaction terms, (2) penalized selection and forced variables, and (3) limited discussion of model stability. Please clarify the events-per-variable considerations, whether shrinkage/validation diagnostics were assessed, and consider simplifying or clearly justifying why all models are necessary.

Response: Thank you for this important comment. We agree that the multi-model strategy should be more clearly justified in relation to sample size and study aims. We have therefore revised the manuscript to distinguish more explicitly between the primary inferential model and the complementary robustness analyses.

In the revised Methods section (Main text: Page 7, Line 310-312) and the added Supplementary Material Section 8 (“Modelling Strategy Overview and Supplementary Technical Details”), cluster-robust logistic regression is now identified more clearly as the primary inferential model. The RDS-weighted model, GEE model, and recruiter-linked network-lag model are presented more explicitly as complementary robustness analyses, each addressing a different methodological concern relevant to RDS-based data rather than serving as interchangeable parallel models for maximal inference. The RDS-weighted model addresses unequal sampling probabilities under RDS, the GEE model addresses within-chain correlation from a population-averaged perspective, and the recruiter-linked network-lag model evaluates potential dependence along recruitment links. In this sense, these additional models function as sensitivity or robustness analyses under different assumptions rather than as multiple competing primary models.

Importantly, the primary inferential model itself was not highly parameterized. The binary outcome was approximately balanced, and the number of free parameters in the main model remained modest relative to the available sample size. We have added events-per-variable (EPV) information for the primary inferential model in the Supplementary Material Section 8 (“Modelling Strategy Overview and Supplementary Technical Details”). For Model 1 (standard logistic regression with chain-clustered robust standard errors), the smaller outcome category contained 201 participants and the model included 15 free parameters, yielding an EPV of 13.4. This suggests that the primary model was not highly parameterized relative to the available sample size. Regarding penalized selection and model stability, elastic net was used only at the screening stage for extended predictors, whereas a priori core covariates were retained in all models based on theoretical relevance and prior literature. As specified in the original Methods, penalized screening was implemented using elastic net logistic regression with 10-fold cross-validation and lambda.min. This step was introduced to reduce overfitting risk and avoid unnecessary model complexity in the extended predictor set, rather than to maximize flexibility or to serve as the sole basis for inference.

We also clarify that the psychosocial terms were not exploratory interaction terms introduced to increase model flexibility. As described in the Methods section (Main text: Page 6 Line 246-259), PCA-derived psychosocial scores were defined only among participants with prior awareness of mpox. To avoid structural missingness and unnecessary loss of the full analytical sample, prespecified awareness-conditioned psychosocial terms were constructed by combining mpox awareness status with each psychosocial component score. These terms were therefore intended to retain sample inclusion while reducing redundancy, rather than to increase overfitting risk. To avoid misunderstanding, we have renamed the previous “interaction terms” as “awareness-conditioned psychosocial terms” throughout the manuscript.

 

Comment 4.The use of interaction terms between awareness and PCA-derived scores is innovative but difficult to interpret. Results are described as “informative” but not always clearly translated into real-world meaning. Also, the inverse association of perceived transmission likelihood is counterintuitive and underexplained. Please, provide a clearer interpretation of (1) What does a unit increase in these interaction terms represent in practice? (2) Why might higher perceived transmission likelihood reduce immediate willingness?

Response: Thank you for this important comment. We agree that the awareness-conditioned psychosocial terms require clearer interpretation. We have revised the manuscript to clarify that these were not exploratory interaction terms introduced to increase model complexity (detailed in our response to Comment 3), but prespecified conditional terms used to retain the full analytical sample while respecting structural missingness by design.

In terms of the inverse association for perceived transmission likelihood, we have clarified that this construct should not be interpreted as perceived personal susceptibility. Instead, it reflects stronger endorsement that mpox may be transmitted through multiple routes. In this setting, higher scores may capture broader route endorsement or diffuse uncertainty rather than greater felt personal risk. Such perceptions may not translate into stronger immediate willingness to vaccinate and may instead coexist with beliefs that risk can be managed behaviorally. We have therefore revised the Discussion section (Main text: Page 18 Line 587-594) to interpret this finding more cautiously as hypothesis-generating rather than definitive.

 

Comment 5. The clustering of non-immediate respondents is a key contribution but requires stronger methodological transparency. There is limited justification for (1) the choice of the PAM algorithm, (2) the selection of 3 clusters, and (3) stability and reproducibility are not discussed. Please consider including (1) cluster validation metrics (e.g., silhouette width), (2) sensitivity analyses (if conducted), and (3) clarify whether clustering is exploratory vs. confirmatory.

Response: Thank you for this important comment. We agree that the concern-profile clustering should be described with greater methodological transparency. The original Supplementary Material Section 7 (“Concern Profiles within the Delayed/Refused Responses Subgroup”) already specified the use of PAM and the general rationale for this choice. In the revised version, we have expanded this section and revised the relevant text in the Introduction (Main text: Page 3 Line 115-116),  Methods (Main text: Page 5 Line 195), Results (Main text: Page 10 Line 398-399, 403) and Limitations (Main text: Page 20 Line 702-713) sections to clarify more explicitly that the clustering analysis was exploratory and hypothesis-generating rather than confirmatory.

PAM was chosen because it is a medoid-based partitioning method that is relatively robust to outlying or extreme response patterns and yields cluster representatives that are easier to interpret in respondent-profile analysis. This was considered advantageous for clustering standardized concern-item responses within the modest-sized delayed/refused subgroup. To improve transparency regarding cluster selection, we now report average silhouette width across candidate solutions (k = 2-4) in the revised Supplementary Material Section 7 (“Concern Profiles within the Delayed/Refused Responses Subgroup”). The silhouette values were 0.177 for k = 2, 0.162 for k = 3, and 0.156 for k = 4. Although the 2-cluster solution showed the highest internal separation, the 3-cluster solution was retained because it provided a more interpretable representation of heterogeneity within the delayed/refused subgroup. In particular, the 2-cluster solution collapsed substantively distinct higher-concern patterns, whereas the 4-cluster solution introduced additional fragmentation without clear interpretive gain.

We also added sensitivity information in the Supplementary Material Section 7 (“Concern Profiles within the Delayed/Refused Responses Subgroup”) and Supplementary Table S5. The Ward hierarchical clustering solution showed broadly similar profile patterns to the retained 3-cluster PAM solution. In particular, both approaches identified (i) a broadly elevated multi-concern profile, (ii) a lower-concern / low-urgency profile, and (iii) a more selective safety- and burden-related concern profile, although the allocation of observations across clusters differed somewhat between algorithms. Bootstrap-based Jaccard summaries further suggested moderate stability overall (mean Jaccard = 0.785; median = 0.828; mean minimum cluster Jaccard = 0.712). Accordingly, we now present the retained 3-cluster solution as an interpretable exploratory profile structure rather than as a definitive latent class solution.

 

Comment 6.Although the authors acknowledge the cross-sectional design, some interpretations imply causal pathways (e.g., “PrEP use reflects prevention orientation influencing vaccination”). Please consistently use non-causal language such as “associated with” instead of “reflects” or “indicates”. Additionally, please strengthen caution in the ‘Discussion’ section.

Response: Thank you for this important comment. We agree that, given the cross-sectional design, causal interpretations should be avoided. We will therefore revise the manuscript to use more consistently non-causal language throughout the Discussion and Conclusion sections and to strengthen caution in interpreting the observed relationships. In particular, wording that could be read as implying causal pathways (e.g., “reflects,” “indicates,” or “influences”) have been replaced with more neutral formulations such as “was associated with,” “was related to,” “was linked to,” or “may be consistent with.”  We will also revise the relevant interpretive sentences to make clearer that the findings are observational, hypothesis-generating, and should not be interpreted as evidence of temporal or causal effects. We marked the revised phrases clearly in the manuscript (Main text: Page 10 Line 403, Page 17-21 Line 542, 575, 600, 601, 602-603, 632, 637, 638, 657-659, 711, 723-724, 727-728).

 

Comment 7. The study is conducted in a single city (Changsha). Findings may not generalize to (1) other regions of China, (2) rural MSM populations, or (3) different healthcare access contexts. Please expand the discussion on (1) regional variability and (2) cultural and healthcare system differences.

Response: Thank you for this important comment. We agree that the external validity of a single-city study should be discussed more explicitly. In the revised manuscript, we have expanded the Limitations section (Main text: Page 20 Line 676-682) to note that transferability may be limited not only across different geographic regions of China, but also for rural MSM populations and for settings that differ in healthcare accessibility, community connectedness, and local sexual-health service infrastructure. We have therefore tempered the generalizability of the findings accordingly.

 

Comment 8. Some sections (especially Methods 2.3) are overly dense and technical. Please consider (1) adding schematic diagrams for the modeling strategy, (2) moving technical details to the Supplementary Material.

Response: Response: Thank you for this helpful suggestion. We agree that some parts of the Methods section (especially Section 2.3) were dense, and we will therefore add a schematic diagram of the modeling strategy (Supplementary Figure S7) in Supplementary Material Section 8 (“Modelling Strategy Overview and Supplementary Technical Details”)  to improve readability and help readers follow the analytic workflow more easily.

At the same time, we have retained the core methodological details in the main Methods section because several of these design choices are central to interpretation of the findings. In the present review round, multiple comments concerned model specification, variable construction, and clustering strategy, suggesting that these technical details are important for transparency rather than merely supplementary. Accordingly, we have not moved the reviewer-raised core methodological rationale out of the main text. Instead, we have aimed to improve accessibility by adding the schematic overview and consolidating selected implementation-level technical details, including moving covariate preparation and elastic net tuning/selection procedures into the same Supplementary Material Section 8 (“Modelling Strategy Overview and Supplementary Technical Details”) that accompanies the schematic diagram. In this way, the main Methods remain interpretable while additional technical detail and sensitivity information are available in the Supplementary Material.

 

Comment 9.“Non-immediate willingness” is somewhat awkward. Consider alternatives for “Delayed or unwilling”, “Non-urgent willingness”. Please define clearly at first use and maintain consistency.

Response: Thank you for this helpful suggestion. We agree that the term “non-immediate willingness” was awkward. To improve clarity, we have retained “mpox vaccination willingness” as the overall construct, consistent with the wording of the survey item, but revised the grouped response label in the primary analysis. We now distinguish “immediate willingness” from “delayed/refused responses,” which more directly reflects the original response options and the pragmatic intent of the grouping. We have now defined this terminology clearly at first use in the Methods section (Main text: Page 4 Line 170-175), applied it consistently throughout the manuscript, and provided the original wording and coding rules in Supplementary Table S2.

 

Comment 10. Table 2 is informative but dense. Please highlight key distinguishing features across clusters. Additionally, consider a visual heatmap for concern profiles.

Response: Thank you for this helpful suggestion. We agree that the original Table 2 contained dense information and that the key distinguishing features across clusters should be more visually accessible. In the original Results text, we already described the main distinguishing features of the three concern profiles in words. To further improve readability and facilitate visual comparison, we have now replaced the original Table 2 with a heatmap of the mean concern-item scores across the three retained concern profiles (Figure 1), which preserves the key information while presenting the profile patterns more intuitively.

 

Comment 11.STROBE compliance is mentioned but could be improved by (1) explicitly reporting missing data handling and (2) providing a participant flow diagram.

Response: Thank you for this helpful suggestion. We agree that STROBE reporting can be made more explicit in these respects. In the revised manuscript, we have clarified missing-data handling more directly in the Statistical Analysis and Reporting section (Main text: Page 7-8 Line 329-334), Because the survey was administered in person under one-to-one guidance after eligibility verification, the final analytic dataset did not contain item-level missing data for the primary regression analyses. Accordingly, Models 1-3 were estimated on the full analytic sample. For sensitive questions, response options such as “prefer not to say” were retained as substantive response categories rather than treated as missing values. The only record excluded from Model 2A (RDS-weighted analysis) was one participant with non-positive reported network degree, for whom an RDS weight could not be computed.

In addition, because the analytic sample structure is already summarized in the schematic modelling strategy diagram (Supplementary Figure S7), we further added a separate participant flow diagram (Supplementary Figure S1) in the Supplementary Material Section 2 (“RDS Recruitment and Survey Implementation Process”) that focuses on the RDS recruitment and survey implementation process, including peer referral, eligibility verification, and completion of the in-person survey. This complements the modelling workflow without duplicating it.

 

Comment 12.Minor grammatical improvements are needed; for example, “may have provided broader coverage” should be changed to “may provide broader coverage”. Occasional repetition in the ‘Discussion’ section could be reduced.

Response: Thank you for this helpful suggestion. We have revised the sentence accordingly and replaced “may have provided broader coverage” with “was intended to improve community-based access to” in the revised manuscript (Main text: Page 19 Line 657). We have also reviewed the Discussion section to reduce occasional repetition and improve overall clarity.

 

Comment 13. Add a conceptual framework linking awareness, psychosocial factors, behaviors, and willingness.

Response: Thank you for this helpful suggestion. We agree that the conceptual links among awareness, psychosocial factors, prevention-related behaviors, and vaccination willingness can be made more explicit. We have therefore added a conceptual framework figure (Supplementary Figure S5) in Supplementary Material Section 6 (“Mpox Awareness and Psychosocial Constructs”) to illustrate the hypothesized relationships among these domains and to clarify how the primary and subgroup analyses relate to the overall study logic. As shown in the framework, mpox awareness was conceptualized as shaping awareness-conditioned psychosocial dimensions, which may in turn relate to prevention-related behaviors and vaccination willingness, while heterogeneity within the delayed/refused response group was further examined through exploratory concern-profile analysis. Because the study was cross-sectional, the figure is intended to summarize conceptual pathways rather than imply confirmed causal direction.

 

Comment 14.Please consider discussing implications for (1) digital health campaigns, (2) MSM community-based interventions, and (3) a brief comparison with global mpox vaccination trends could strengthen context.

Response: Thank you for this helpful suggestion. We agree that the broader public health implications of the findings can be discussed more explicitly. In the revised Discussion (Main text: Page 19 Line 640-654), we have added a short implications-oriented paragraph addressing the relevance of the findings for digital health communication and MSM community-based interventions. We have also strengthened the existing discussion of broader international mpox vaccination trends by clarifying that such comparisons should be interpreted cautiously because survey timing, vaccine availability, rollout context, and healthcare access differed across settings (Main text: Page 17-18 Line 570–572).

 

Additional Clarifications

Following the reviewers’ suggestions, several new materials have been added. In the main text, Table 2 has been replaced with a heatmap (Figure 1). The two forest plots have also been revised and updated, and are now presented as Figure 2 and Figure 3. Accordingly, the main text now contains one table and three figures.

In the Supplementary Materials, three new sections have been added: (1) RDS Sample Size Estimation; (2) RDS Recruitment and Survey Implementation Process; and (3) Modelling Strategy Overview and Supplementary Technical Details. In addition, two new supplementary tables have been included: Supplementary Table S2 (Original Survey Questions and Coding Rules for Categorical and Continuous Variables) and Supplementary Table S5 (Mean Concern-item Scores across Candidate PAM Solutions and Ward Hierarchical Clustering in the Delayed/Refused Responses Subgroup). Four new supplementary figures have also been added: Supplementary Figure S1 (RDS participant flow and survey implementation process), Supplementary Figure S5 (Conceptual framework linking mpox awareness, psychosocial dimensions, prevention-related behaviors, and vaccination willingness), Supplementary Figure S6 (Average Silhouette Width across Candidate PAM Solutions [k = 2–4] for the Delayed/Refused Responses Subgroup), and Supplementary Figure S7 (Schematic overview of the analytic strategy).

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript is correct.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have made a sincere effort to incorporate most of my suggestions/comments in the modified version of the manuscript. The quality of the manuscript has improved significantly. I, therefore, recommend the article for publication in the Vaccines journal.