Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a relevant and potentially useful review. However, the manuscript presents significant conceptual and methodological problems that limit its current suitability for publication. My comments follow section by section.

 

Abstract:

- The statement that these biomarkers may be relevant to “anti-tumoral response, treatment stratification, and clinically relevant decisions” feels too ambitious given that the included studies are mainly prognostic, heterogeneous, and not specifically designed to evaluate treatment response or therapeutic selection.

 

Introduction:

- The manuscript currently states that IHC-based studies were excluded because IHC is “inherently prone to bias due to its hypothesis-driven nature,” which is too categorical and methodologically debatable. The justification should be reframed more rigorously around review scope rather than implied inferiority of non-omics evidence. In addition, the Introduction repeatedly links prognostic biomarkers to cN0 neck management, adjuvant therapy selection, and surveillance intensity, but this translational framing is more aspirational than evidence-based for the included dataset.

 

Materials and Methods:

- The eligibility criteria need clarification. The population is described as “primarily surgically treated TSCC,” yet it is not clear how strictly this was applied across included studies or whether mixed-treatment cohorts were excluded if prognostic analyses remained eligible.

- The exclusion of non-English studies introduces language bias and should be acknowledged more clearly as a limitation rather than a neutral procedural choice.

- The decision not to search grey literature is understandable, but it should likewise be discussed as a possible source of publication bias.

- The use of QUIPS for single prognostic factor studies and PROBAST for prognostic models is reasonable, but the manuscript should explain more explicitly how hybrid studies were handled, especially papers that mix single-marker findings, signatures, and nomograms.

- The exploratory meta-analysis of CA9 needs more caution. The manuscript itself acknowledges that most biomarkers were too heterogeneous for pooling, and only CA9 met the criteria. However, with only two studies, moderate heterogeneity, different platforms, and likely substantial clinical and analytical differences, the practical value of pooled estimates is very limited.

- Include as supplementary material a list of the pre-selected studies that were excluded in the final screening and state the reason for the exclusion of each study.

 

Results:

- The study selection reporting appears internally inconsistent. In the narrative, the title/abstract screening “reduced the set to 47 articles,” and “30 were excluded,” leading to 17 included studies. However, the PRISMA figure on page 9 shows a different flow with 1006 records screened, 115 reports sought, 114 reports assessed, and exclusions of 67 non-omics studies plus 31 wrong-study-design reports before reaching 17 included studies. These numbers do not reconcile cleanly and must be corrected.

- There is also a table-numbering inconsistency. Table 1 is introduced as the PICO framework, yet in the Results the authors write that “Details of the 17 eligible studies are summarized in Table 1,” whereas the eligible studies are actually listed later in Table 2. This should be corrected throughout.

- The biomarker synthesis is detailed, but the Results would benefit from stricter separation between descriptive findings and higher-confidence findings. At present, single-study associations, internally validated signatures, and replicated markers are discussed in a fairly continuous narrative, which may blur evidentiary strength.

- The statement that Figure 4 shows “the most reliable prognostic biomarkers” is too strong. Reliability is not demonstrated here in a formal way; at best, the figure shows the most statistically robust biomarkers among those with extractable multivariable OS estimates. The wording should be revised. Similarly, the CA9 section should not imply that exploratory pooling meaningfully resolves the broader reproducibility problem.

- The REMARK assessment is also presented somewhat optimistically. The manuscript says reporting quality was generally acceptable and that most studies scored above 80%, yet nearly all studies remained moderate- or high-risk by QUIPS/PROBAST. That contrast is actually important and should be made more explicit in the Results rather than only later in the Discussion.

 

Discussion:

- The manuscript includes what appears to be an editorial/template instruction embedded in the body text: “Authors should discuss the results and how they can be interpreted…” This must be removed.

- The discussion of biomarkers as possible criteria for anti-tumoral response, treatment sensitivity, adjuvant decision-making, and cN0 management is speculative. The authors do acknowledge this later, but the framing remains broader than the actual evidence base, which is mostly prognostic and rarely treatment-annotated. That interpretation should be more explicitly labeled as a future hypothesis.

- The CA9 discussion is biologically plausible and probably the most convincing part of the manuscript, but it still needs tighter calibration.

 

Conclusion:

- The Conclusion is better balanced than the abstract and appropriately states that the evidence is hypothesis-generating rather than practice-changing. That said, it would be improved by one final sentence explicitly stating that no biomarker is currently ready for routine prognostic use in TSCC. This would align the conclusion with the risk-of-bias findings and avoid any residual ambiguity.

Comments on the Quality of English Language

 Language, structure, and editorial quality:

- The manuscript needs another careful round of English and technical editing. There are multiple small but noticeable issues.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1. Risk of double counting in meta-analysis
Both Zhu et al. and Wang et al. utilize TCGA and GEO datasets. Pooling these studies without explicitly addressing dataset overlap introduces a significant risk of double counting patients and may lead to biased pooled HR estimates.

2. Inappropriate meta-analysis with k = 2
Performing a random-effects model and calculating heterogeneity statistics (I², τ²) with only two studies is statistically unstable. These estimates are unreliable and should not be interpreted as robust findings.

3. Incorrect hazard ratio interpretation
The manuscript states that TNFAIP3 and NRAS are associated with poorer survival. However, the reported HR values (<1) indicate a lower risk, not poorer survival. This is a fundamental interpretative error.

4. Biomarker count inconsistency
At the beginning of Section 3.3, the text states that there are “90 distinct molecular biomarkers.” However, the breakdown lists 59 mRNAs (62%), 19 lncRNAs (21%), 9 circRNAs (10%), and 6 miRNAs (7%). These categories sum to 93 rather than 90. These numbers are internally inconsistent and should be corrected.

5. Methodological inconsistency
Lee et al. [20] is described as microarray-based in the main text but as RNA-seq in the table. This inconsistency raises concerns about the accuracy of data extraction and reporting.

6. Gene naming error
AC139530.1 is incorrectly labeled as AAC139530.1 in Figure 4. This affects reproducibility and database traceability.

7. Lack of independent validation and replication
Most biomarkers are reported in single studies, with CA9 being the only biomarker appearing in more than one study. However, even for CA9, the included studies rely on overlapping public datasets (TCGA/GEO), limiting true independent validation. This significantly weakens the robustness and translational value of the findings.

8. Confounding factors not clearly addressed
Although multivariable analysis is mentioned, the manuscript does not clearly specify which covariates were adjusted for across studies.

9. Template text not removed
At the end of Section 3.3.1, residual journal template text remains (“Authors should discuss the results and how they can be interpreted from the perspective of previous studies... Future research directions may also be highlighted.”). This should have been removed before submission and reflects careless manuscript preparation.

10. Reference inconsistencies
The reference list appears to contain numbering irregularities, such as missing or out-of-order citations. The authors should carefully verify the reference sequence and ensure consistency throughout the manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have made substantial improvements to the manuscript. The revised version is methodologically stronger, more transparent, and significantly better calibrated scientifically. Most of the major concerns raised in the first round were adequately addressed. However, a few residual issues remain:

 

- CA9 Meta-analysis: Although improved, the CA9 section still risks implicit overemphasis, given: Only two studies included; Platform and cohort heterogeneity. Add one final explicit sentence in Results and/or Discussion such as: “This analysis should be interpreted strictly as exploratory and not as evidence of reproducible clinical effect.”

- Some minor inconsistencies remain: “biomarkers” vs “signatures” vs “features”; “prognostic” vs “predictive” occasionally appear close together. Ensure strict and consistent terminology throughout.

Final Recommendation: Accept after minor revisions.

Author Response

"Please see the attachment." 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have made substantial and meaningful revisions in response to the previous review. Several critical methodological and interpretative issues have been appropriately addressed, particularly regarding statistical transparency, data consistency, and the overinterpretation of findings.
Despite these improvements, several issues still require clarification or refinement before the manuscript can be considered for acceptance:

1. Residual Overinterpretation of Translational Relevance
Although the authors have appropriately downgraded their conclusions, the manuscript still occasionally implies potential clinical applicability (e.g., treatment stratification or surveillance decisions) without sufficient supporting evidence.
Further align all translational statements with the clearly stated “hypothesis-generating” nature of the current evidence to avoid overinterpretation.

2. Evidence Tier Framework – Conceptually Strong but Requires Careful Framing
The introduction of the evidence tier system is a strength of the manuscript. However:
• Most biomarkers are supported by single studies or analyses of overlapping public datasets 
• True independent validation remains extremely limited 
Explicitly clarify that the tier classification represents relative prioritization within a limited evidence base, rather than an indication of clinical readiness or applicability.

3. Dataset Overlap – Acknowledged but Still a Structural Limitation
While the issue of TCGA/GEO dataset overlap has been acknowledged, the manuscript still relies heavily on findings derived from these datasets.

More explicitly emphasize that:
• Computational replication does not equate to biological or clinical validation 
• This limitation reflects a broader constraint of the field, not only this review 

4. Integration of Risk of Bias into Interpretation
The application of QUIPS, PROBAST, and REMARK is appropriate and well executed, particularly in the Supplementary Materials. However, their implications are not fully integrated into the main narrative.
Strengthen the connection between:
• risk-of-bias assessments 
• and the interpretation of biomarker robustness and credibility 
This will improve the interpretability and translational relevance of the review.

5. Integration of Clinicopathologic Covariates and Sub-site Heterogeneity
While the authors have expanded their discussion of multivariable models and covariates, the manuscript still lacks a sufficiently critical analysis of biological heterogeneity within tongue cancer.
(a) Anatomic Sub-site Specificity
There are well-recognized biological and immunological differences between:
• base of tongue (oropharyngeal) 
• oral/mobile tongue 
However, the current synthesis appears to treat “tongue cancer” as a relatively homogeneous entity.

Clarify whether the included studies distinguished between these sub-sites, and discuss how such heterogeneity may influence biomarker performance and effect size estimation.

(b) Independent Prognostic Value Beyond Clinical Factors
It remains unclear whether the identified high-tier biomarkers provide meaningful incremental prognostic value beyond established clinicopathologic parameters, such as:
• nodal status 
• extracapsular extension 
• perineural invasion (PNI) 

Explicitly discuss whether:
• these biomarkers were validated in multivariable models including key clinical covariates 
• they retain independent prognostic significance in the context of aggressive disease features

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf