I have completed a thorough peer review of the revised manuscript entitled "How Anterior Crossbite Severity Relates to Appearance-Based Bullying in School-Age Children: Evidence from the ROMA Index" (Version 2; DOI: 10.12688/f1000research.179073.2), originally submitted by M. Babakurd and colleagues.

I acknowledge that this is the second version of the manuscript, following an initial round of peer review in which the first reviewer recommended against acceptance due to fundamental statistical errors, particularly concerning the binary logistic regression analysis presented in Table 4 of Version 1. Having carefully examined Version 2, I am pleased to report that the authors have systematically addressed the most critical issues raised. Nevertheless, several remaining concerns require attention before the manuscript can be considered fully suitable for indexing.

I begin with the positive aspects of this work. The research question is clinically relevant and timely, as the relationship between malocclusion severity and psychosocial outcomes in children has gained increasing attention in orthodontic literature, yet data from Middle Eastern populations, particularly Syria, remain scarce. The authors should be commended for their large sample size of 2,080 children, which, despite some concerns regarding sample size justification discussed below, provides considerable statistical power and precision for most analyses. The use of the ROMA Index for clinical assessment represents a methodological strength, as this index offers standardized, reproducible, and internationally recognized grading of malocclusion severity, moving beyond simple binary classifications of presence or absence of crossbite. Furthermore, the modification and validation of the Olweus Bully/Victim Questionnaire for dental appearance-related bullying was conducted with appropriate rigor: the pilot study demonstrated good psychometric properties, including a Content Validity Index of 0.92, Cronbach's alpha of 0.85 for internal consistency, and an intraclass correlation coefficient of 0.88 for test-retest reliability.

The authors also deserve praise for their commitment to transparency and reproducibility, as they have deposited their underlying data, consent forms, and questionnaires in the Open Science Framework repository (DOI: 10.17605/ OSF.IO/HW8KM), which aligns with best practices in open science. Finally, the focus on severity-based analysis rather than simply comparing presence versus absence of anterior crossbite represents a conceptual advance that enriches our understanding of the dose-response relationship between clinical conditions and social victimization.

Turning to the critical issues that led to the initial rejection, I have carefully verified that the authors corrected the binary logistic regression error in Table 4 of Version 2. In Version 1, the odds ratios for children with malocclusion and anterior crossbite were reported as 0.000 (95% CI: 0.000 to 0.003) and for anterior crossbite only as 0.231 (95% CI: 0.176 to 0.303), which mathematically indicated a protective effect and directly contradicted the descriptive data showing 87.1% and 74.3% bullying victimization rates in those groups, respectively, compared to only 0.1% in the normal occlusion group. This error has been rectified in Version 2. The revised Table 4 now presents odds ratios that are consistent with the descriptive statistics: children with both malocclusion and anterior crossbite demonstrate an odds ratio of 4678.491 (95% CI: 504.983 to 43344.549, p < 0.001), those with anterior crossbite only show an odds ratio of 2160.414 (95% CI: 299.538 to 15581.948, p < 0.001), and those with other malocclusion only present an odds ratio of 497.615 (95% CI: 69.700 to 3552.681, p < 0.001), all compared to the normal occlusion reference group. The ordinal logistic regression analysis in Table 5, which demonstrated a dose-response relationship between ROMA severity grades and bullying intensity, remains conceptually sound and has been properly interpreted with attention to the inverse scoring of the modified Olweus scale.

The authors have also moderated causal and speculative language throughout the manuscript, adopting a purely associational framework as recommended by the first reviewer. Specifically, they no longer claim causal inference from their cross-sectional design, and they have acknowledged that unmeasured confounders such as baseline self-esteem, pre-existing psychological vulnerability, family support structures, and socioeconomic status may influence the observed associations. These corrections are substantial and demonstrate responsiveness to peer review.

Despite these important improvements, several methodological and reporting concerns persist and must be addressed before the manuscript can be accepted for indexing. The most significant remaining issue concerns the sample size justification. The authors initially calculated that approximately 150 children would be needed for prevalence estimation and approximately 236 children for the association between anterior crossbite and bullying, based on standard formulas for cross-sectional studies with 95% confidence level and 80% power. However, they ultimately recruited 2,080 children, nearly ten times the initial estimate. While a larger sample is not inherently problematic, the justification provided in Version 2 raises concerns about post-hoc rationalization. The authors invoke a design effect (Deff) of 2.5 due to cluster sampling, which is a reasonable adjustment, and they further argue that complex multivariable modeling, including ordinal logistic regression analyses with Mult categorical subgroups, necessitated a considerably larger sample to prevent model overfitting and ensure adequate cell frequencies. They also cite the need to detect small, expected effect sizes of f² = 0.02 with statistical power of 0.90. However, given that the observed associations are extraordinarily strong (Cramer's V = 0.578 for the main association, and odds ratios in the thousands), the justification of needing a sample size of 2,080 to detect a "small effect size" appears inconsistent. In fact, such strong associations could have been reliably detected with a much smaller sample. This discrepancy suggests that the sample size was likely determined by convenience or by the availability of schools and children rather than by a strict a priori power calculation based on expected effect sizes. While this does not invalidate the findings, it weakens the transparency of the study design and should be acknowledged as a limitation. The authors should explicitly state in the limitations section that the final sample size was larger than strictly necessary for the primary associations and that the initial sample size calculation was expanded primarily to enable stable estimation of subgroup effects and complex models, rather than for detecting the main effect.

A second concern relates to the extraordinarily high odds ratios and their extremely wide confidence intervals in Table 4. For example, the odds ratio for the malocclusion and anterior crossbite group is 4678 with a 95% confidence interval ranging from 505 to 43344, spanning nearly two orders of magnitude. Such wide intervals indicate low precision, which is a direct consequence of the small number of children in this subgroup (n = 31, as shown in Table 3) and the near-zero prevalence of bullying victimization in the normal occlusion group (only 1 out of 703 children, or 0.1%). This pattern raises the possibility of quasi-complete separation in the logistic regression model, a phenomenon that occurs when a predictor perfectly or almost perfectly predicts the outcome in one category. In such situations, maximum likelihood estimation can produce unstable odds ratios that are inflated and have poor coverage properties. While the direction of the association is undoubtedly correct—children with anterior crossbite have vastly higher bullying rates—the precise magnitude of the odds ratio should be interpreted with caution. The authors do not discuss this issue in the limitations section, nor do they report having conducted any sensitivity analyses, such as Firth's penalized likelihood logistic regression, which is specifically designed to handle separation problems. I recommend that the authors either perform such sensitivity analyses or, at minimum, add a paragraph in the discussion acknowledging that the extreme odds ratios and wide confidence intervals reflect both the strength of the association and the statistical instability arising from the sparse data in the reference group and the small subgroup with combined malocclusion and anterior crossbite.

Third, the description of bullying patterns and locations in Table 6, while informative, presents some interpretive challenges. The authors have now reintegrated the normal occlusion group as a baseline comparative anchor, which addresses the previous reviewer's concern. However, the statistical comparisons across four groups where one group (normal occlusion) has virtually no events (only one child reported teasing, and no child reported any other form of bullying) violate the assumption of the chi-square test that expected cell frequencies should be five or more. While the authors note that "no expected count less than five" in the footnote to Table 3, this claim is difficult to reconcile with the data in Table 6, where numerous cells have zero counts. For instance, in the analysis of unpleasant nicknames, the normal occlusion group has zero events, and the expected frequency under the null hypothesis would be substantially less than five. The p-values reported for these comparisons should therefore be interpreted with caution, as the chi-square approximation may be unreliable. The authors should consider using Fisher's exact test for these comparisons or explicitly acknowledge this limitation in the text.

Fourth, the age effect reported in Table 4 deserves more careful discussion. The authors found that each one-year increase in age elevated the likelihood of victimization by 23.9% (OR = 1.239, 95% CI: 1.111 to 1.383, p < 0.001). This finding is interesting and potentially important, as it suggests that older children within the 8-to-12-year range are at higher risk. However, the authors do not explore whether this age effect interacts with occlusal status. For example, is the association between anterior crossbite and bullying stronger in older children than in younger ones? Such interaction analysis would be valuable and could be performed by including an age-by-crossbite interaction term in the logistic regression model. If the authors have data to conduct this analysis, I encourage them to do so. If not, they should at least acknowledge that the age effect was examined only as a covariate and that potential moderation by occlusal status remains unexplored.

Fifth, while the authors have moderated causal language as requested, some phrasing remains overly assertive for a cross-sectional study. For example, in the abstract, they state that "anterior crossbite severity serves as a meaningful correlate" (which is acceptable) but in the conclusion they write that "early identification of anterior crossbite severity during mixed dentition enables timely recognition of children with an increased likelihood of experiencing psychosocial harm." This is acceptable as a statement of association, but the phrase "enables timely recognition" might be misread as implying that screening for crossbite will necessarily lead to reduced bullying, which is not demonstrated by this study. I suggest replacing "enables" with "may facilitate" to better reflect the exploratory nature of cross-sectional evidence.

Sixth, although the authors have provided detailed information about the ROMA Index and its application, I note that the reliability assessment (Cohen's kappa values ranging from 0.85 to 0.95) was performed on only 10% of participants with repeat examinations spaced 20 to 30 minutes apart. This short interval may artificially inflate kappa values because memory effects could influence the second examination. Ideally, repeat examinations should be separated by at least one week to minimize recall bias. The authors should acknowledge this as a minor limitation.

On balance, the manuscript has improved substantially from Version 1. The critical statistical error has been corrected, the language has been appropriately moderated, and the authors have demonstrated responsiveness to peer review. The research question remains important, the dataset is valuable, and the main finding—that anterior crossbite severity shows a strong, dose-response association with appearance-related bullying in Syrian schoolchildren—is both clinically relevant and methodologically supported after the corrections.

However, the remaining issues concerning sample size justification, instability of extreme odds ratios, potential quasi-complete separation, sparse data in chi-square comparisons, and lack of interaction analysis preclude acceptance in the current form. These issues are not fatal, but they require attention.

Therefore, my recommendation is:  Accept with minor revisions (Indexable after minor revisions). Specifically, I request that the authors:

(1) add a paragraph to the limitations section acknowledging that the sample size was substantially larger than necessary for the primary associations and providing a transparent justification for the expansion;

(2) conduct sensitivity analyses using Firth's penalized likelihood logistic regression to verify the stability of the odds ratios in Table 4, or, if such analyses are not feasible, explicitly acknowledge the potential for quasi-complete separation and the resulting imprecision;

(3) either recalculate the p-values in Table 6 using Fisher's exact test for comparisons involving zero cells or explicitly acknowledge the limitation of chi-square approximation under sparse data;

(4) consider testing an age-by-crossbite interaction or at least acknowledge its absence as a limitation; and

(5) make minor wording adjustments to avoid any remaining overstatements of causality.

These revisions are relatively modest and can be completed without new data collection or reanalysis of the entire dataset. Once these changes are implemented, I believe the manuscript will meet the standards for indexing in F1000Research.

Respectfully submitted.

Is the work clearly and accurately presented and does it cite the current literature?

Yes

If applicable, is the statistical analysis and its interpretation appropriate?

Partly

Are all the source data underlying the results available to ensure full reproducibility?

Partly

Is the study design appropriate and is the work technically sound?

No

Are the conclusions drawn adequately supported by the results?

No

Are sufficient details of methods and analysis provided to allow replication by others?

No

Reviewer Expertise:

Orthodontics

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

Thank you for the opportunity to review the manuscript entitled “How Anterior Crossbite Severity Relates to Appearance-Based Bullying in School-Age Children: Evidence from the ROMA Index” (version 1; DOI: 10.12688/f1000research.179073.1). I have completed a thorough critical appraisal of this work. While the research question is clinically relevant and the authors have attempted to address an important gap in the literature concerning anterior crossbite severity and psychosocial outcomes in Syrian schoolchildren, I regret to conclude that the manuscript is not acceptable for indexing in its current form due to fundamental statistical errors that compromise the validity and interpretability of the main findings.

The most serious flaw concerns the binary logistic regression analysis presented in Table 4. The authors report odds ratios for bullying victimization comparing different occlusal status groups to the reference category of normal occlusion. According to the table, children with both malocclusion and anterior crossbite have an Exp(B) of 0.000 with a 95% confidence interval of 0.000 to 0.003, and children with anterior crossbite only have an Exp(B) of 0.231 with a 95% confidence interval of 0.176 to 0.303. Both odds ratios are substantially below 1, which mathematically indicates a strong protective effect, meaning these groups would have a much lower risk of bullying compared to children with normal occlusion. However, this directly contradicts the descriptive data presented earlier in Table 3, where 87.1% of children in the malocclusion and anterior crossbite group and 74.3% of those with anterior crossbite only were classified as bullying victims, whereas only 0.1% of children with normal occlusion were victims. The odds ratios as reported are therefore impossible to reconcile with the observed proportions, indicating a critical error in either the coding of the outcome variable, the specification of the reference category, the direction of the calculated odds ratios, or the interpretation of the regression output. Until this error is corrected and the analysis is rerun with proper verification, no reliable conclusions can be drawn from the multivariable analysis, and consequently the paper’s central claim that anterior crossbite increases bullying risk lacks valid statistical support.

Beyond this critical error, there are additional methodological concerns that require attention. The sample size justification is problematic. The authors initially calculate that approximately 150 children would be needed for prevalence estimation and approximately 236 children for the association between crossbite and bullying, yet they ultimately recruit 2,080 children without providing a clear, statistically grounded rationale for this tenfold expansion. While a larger sample is not inherently problematic, the absence of a priori justification based on expected effect sizes, design effects from cluster sampling, or required power for subgroup and ordinal regression analyses weakens the transparency of the study design. Furthermore, the manuscript claims a dose-response relationship between crossbite severity and bullying severity based on ordinal logistic regression (Table 5), which is conceptually sound, but this finding is undermined by the unresolved inconsistency in the binary regression results. Readers and reviewers cannot fully trust the ordinal model when the more basic binary model contains an uncorrected error.

Additionally, the authors invoke Bradford Hill criteria for causality, particularly the dose-response principle, yet the cross-sectional design of the study precludes any causal inference. The observed associations, even if correctly analyzed, could be explained by unmeasured confounders such as baseline self-esteem, pre-existing psychological vulnerability, family support structures, or socioeconomic status, none of which were included in the regression models. The language throughout the discussion, including phrases such as “causal inference” and “risk indicator” being interpreted as more than associative, overreaches the limitations of the study design. The authors should reframe their conclusions to reflect only associational findings.

There are also several reporting inconsistencies that detract from the manuscript’s clarity. On page 4, the text refers to “anterior cross-profile prevalence” and “anterior cross-site” instead of anterior crossbite, suggesting typographical errors that should be corrected. The description of the ROMA grading system indicates that higher grades represent greater severity, yet the interpretation of the ordinal regression coefficients requires careful attention to the coding of bullying severity scores, where lower scores paradoxically indicate greater bullying exposure. While the authors acknowledge this coding scheme, the presentation remains confusing for readers not intimately familiar with the modified Olweus scoring method. Finally, the exclusion of children with normal occlusion from the analysis of bullying patterns and locations (Table 6) is justified on the grounds that bullying was negligible in this group, but this decision removes an important comparative anchor and limits the interpretability of the location data.

On a positive note, the study has several strengths that should be acknowledged. The use of the ROMA Index provides a standardized and reproducible clinical assessment, and the pilot testing of the modified Olweus questionnaire demonstrated good psychometric properties with a content validity index of 0.92, Cronbach’s alpha of 0.85, and an intraclass correlation coefficient of 0.88. The sample size of 2,080 children is impressive and, if properly analyzed, could provide precise estimates. The authors should also be commended for making their data and materials available through the Open Science Framework, which aligns with F1000Research’s commitment to transparency and reproducibility. The study addresses a genuinely under-researched population, namely children in Syria, and the focus on severity-based analysis rather than simple presence or absence of malocclusion is conceptually valuable.

I believe the research question remains important and the dataset has potential value. Therefore, I encourage the authors to thoroughly revise the statistical analysis, specifically correcting the binary logistic regression error and ensuring consistency across all tables and interpretations. They should also provide a clearer and more justified sample size calculation, tone down causal language, and improve reporting consistency. Once these revisions are made, the manuscript could be resubmitted as a new version for further review. Given the severity of the statistical error, I cannot recommend acceptance at this stage, but I remain open to evaluating a corrected version in the future.

Is the work clearly and accurately presented and does it cite the current literature?

Yes

If applicable, is the statistical analysis and its interpretation appropriate?

Partly

Are all the source data underlying the results available to ensure full reproducibility?

Partly

Is the study design appropriate and is the work technically sound?

No

Are the conclusions drawn adequately supported by the results?

No

Are sufficient details of methods and analysis provided to allow replication by others?

No

Reviewer Expertise:

Orthodontics

I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.