Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents a systematic review and exploratory quantitative synthesis evaluating circulating microRNAs as potential biomarkers of atrial fibrillation recurrence after catheter ablation. The topic is clinically relevant, particularly given the increasing interest in molecular biomarkers for risk stratification after AF ablation. The manuscript has several strengths, including a candidate-centered evidence-mapping framework, explicit handling of overlapping cohorts, separate analyses of odds ratios and hazard ratios, assessment of risk of bias using QUIPS, and evaluation of certainty using GRADE.

The authors are also appropriately cautious regarding the clinical applicability of circulating microRNAs and clearly acknowledge the substantial biological and analytical heterogeneity across studies. The distinction between association, mechanistic plausibility, and clinical utility is particularly appropriate.

Overall, I consider the manuscript potentially suitable for publication after minor revision. The following points should be addressed to improve clarity, reproducibility, and consistency.

Major comments

1. Clarify the discrepancy in the number of participants reported in different sections.

The Abstract states that the 25 independent recurrence cohorts included 3205 analyzed participants, whereas the Results section reports 3309 enrolled and 3205 analyzed participants. Although this may simply reflect the distinction between enrolled and analyzed populations, the terminology should be made completely consistent throughout the manuscript, particularly in the Abstract, Results, tables, and supplementary material.

Please explicitly define whether the reported sample size represents enrolled participants, analyzed participants, or participants contributing to a particular quantitative synthesis.

2. Clarify the rationale for the cross-biomarker quantitative synthesis.

The manuscript appropriately emphasizes that different microRNAs, biological matrices, compartments, and assay scales are not directly commensurate. Nevertheless, pooling these biomarkers into exploratory OR and HR models may still be difficult for readers to interpret.

The authors should provide a brief additional justification explaining why the cross-biomarker meta-analysis was considered informative despite the substantial biological and analytical heterogeneity. It would also be helpful to emphasize more explicitly in the Results and Discussion that these pooled estimates should not be interpreted as a biological class effect.

3. Provide additional clarification regarding candidate prioritization.

The candidate-centered framework is an important feature of this review. However, the manuscript states that candidate prioritization was post hoc and that a candidate entered the prospective validation set when direct recurrence evidence was accompanied by support in at least one additional domain.

Please clarify whether this framework was developed before the final results were known and, if not, explain how the authors minimized the possibility of post hoc selection bias. The current wording is generally transparent, but a short additional explanation would improve methodological confidence.

4. Clarify the handling of miRNA nomenclature and mature-arm assignment.

The manuscript appropriately distinguishes between mature-arm-specific nomenclature and legacy arm-unspecified labels. However, terms such as “miR-21,” “miR-21-5p,” “miR-150,” and “miR-150-5p” are used across different sections.

Please consider adopting a standardized nomenclature throughout the main text and tables, while retaining the original terminology of individual studies in supplementary material when necessary. This would make it easier for readers to understand whether apparently similar biomarkers truly represent the same mature miRNA species.

5. Explain more clearly how study overlap was determined.

The manuscript states that 35 publications represented 29 linked studies or cohort programs and that overlapping publications were linked to prevent duplicate quantitative weighting. This is an important methodological strength.

However, because multiple publications from the same or potentially overlapping cohorts appear to contribute evidence for individual candidates, a brief description of the criteria used to determine cohort linkage would be useful in the main manuscript. In particular, readers would benefit from knowing which characteristics were used to distinguish independent cohorts from potentially overlapping reports.

Minor comments

6. Abstract wording

The phrase “Six candidates were identified” could be slightly revised to make clear that these were candidates selected for prospective validation based on the predefined/post hoc evidence-mapping framework rather than established biomarkers.

For example:

“Six candidate microRNAs were identified as priorities for prospective validation.”

This would be more consistent with the cautious interpretation used later in the manuscript.

7. Definition of recurrence

The primary outcome includes “AF or another atrial tachyarrhythmia.” Because recurrence definitions and rhythm-monitoring strategies may substantially influence prognostic estimates, please provide a concise statement in the Results describing the major differences in recurrence ascertainment among included studies.

The manuscript already acknowledges this heterogeneity in the Methods, but a short summary of its practical implications would strengthen the interpretation.

8. Risk-of-bias presentation

The manuscript reports that risk of bias ranged from low/moderate to high, with common concerns involving prognostic-factor measurement, confounding, recurrence ascertainment, and analysis/reporting. Please consider adding one or two representative examples of these limitations in the main text rather than referring predominantly to the supplementary material.

This would help readers understand the practical reasons for downgrading certainty.

9. Statistical terminology

The manuscript uses both “random effects summary” and “random effects model.” Please ensure that the terminology is consistent and specify the exact estimator/model assumptions in the supplementary statistical methods if not already provided.

Given the very high heterogeneity (I² = 89.4% for the OR analysis), the interpretation of the pooled estimate should remain explicitly exploratory.

10. Prediction intervals

The very wide prediction intervals, particularly for the HR analysis, are appropriately reported. I suggest briefly emphasizing in the Discussion that these intervals indicate substantial uncertainty regarding the direction and magnitude of association in a new population.

11. Table 2

Table 2 contains biomarkers with substantially different assay scales and biological interpretations. A short footnote explaining that the numerical ORs cannot be interpreted as directly comparable effect sizes across different microRNAs would be helpful.

12. Mechanistic interpretation

The discussion of miR-21, miR-29b, and miR-150 is informative. Nevertheless, some mechanistic statements could be more explicitly separated into experimentally demonstrated mechanisms versus pathway-level or computational hypotheses.

The authors already make this distinction for miR-10b-5p; applying the same wording consistently throughout the mechanistic discussion would further reduce the risk of overinterpretation.

13. Clinical utility

The manuscript appropriately states that clinical utility remains unestablished and recommends evaluation of calibration, discrimination, reclassification, and decision-curve analysis. I suggest adding that any future incremental-value analysis should compare microRNA-based models against well-established clinical and imaging predictors using prespecified model-development and external-validation cohorts.

14. Language and presentation

The manuscript is generally well written, but several sentences are dense and contain multiple methodological concepts. A final language edit would improve readability, particularly in Sections 2.7–2.9 and 4.1–4.3.

Recommendation

Minor Revision

The manuscript addresses an important and emerging topic and provides a thoughtful candidate-centered synthesis of a heterogeneous literature. The requested revisions are primarily intended to improve methodological transparency, consistency of terminology, and interpretation of the exploratory quantitative analyses. I do not believe that major restructuring of the study is necessary.

Author Response

Circulating MicroRNAs as Molecular Biomarkers of Atrial Fibrillation Recurrence After Catheter Ablation

General response. We thank the reviewer for the constructive assessment and for identifying areas that required clearer reporting. We revised the participant denominators, cohort-linkage criteria, recurrence ascertainment, candidate-selection framework, nomenclature, risk-of-bias examples, statistical terminology, mechanistic interpretation, and clinical-validation requirements. Candidate-specific evidence mapping is now the primary synthesis, while cross-biomarker models remain secondary supplementary sensitivity analyses. For verification, changes made specifically in response to Reviewer 1 are shown in blue, whereas changes made in response to Reviewer 2 are shown in red.

Color convention. When the two reviewers requested the same or closely related correction, the passage retains the color of the reviewer whose comment first prompted the revision. Only additional wording introduced specifically for Reviewer 1 is blue; text already revised for Reviewer 2 remains red. This convention is intended solely to facilitate editorial and author verification and does not alter the clean final version that will be prepared after acceptance.

1 Participant counts

Reviewer comment: Please clarify the discrepancy between 3205 analyzed participants in the Abstract and 3309 enrolled participants in the Results, and specify which population contributes to each synthesis.

Response: We clarified the denominator throughout the revision. The 25 independent recurrence cohorts enrolled 3309 participants, of whom 3205 were analyzed. These totals describe the recurrence-specific evidence base and are distinct from the samples contributing to the secondary sensitivity models. The restrictive OR model includes 9 cohorts and 1162 participants; the HR model includes 3 cohorts and 364 participants. The Abstract, Results, Table 1 note, Supplementary Method 2, and Supplementary Tables S8 and S10 now use these definitions consistently.

Revised locations: Abstract; Results 3.2; Table 1 note; Supplementary Method 2; Supplementary Tables S8 and S10.

2 Cross biomarker quantitative synthesis

Reviewer comment: Please justify the exploratory pooling of biologically and analytically heterogeneous microRNAs and state clearly that the summaries do not represent a biological class effect.

Response: We agree that the pooled estimates cannot be interpreted as a biological class effect. Candidate-specific evidence mapping is now the primary synthesis. We removed pooled OR and HR estimates from the Abstract and primary interpretation, and Figure 2 now presents only individual study estimates without a pooled diamond. The random-effects models remain only as secondary supplementary sensitivity analyses. The text explains that a random-effects model cannot make different molecules, matrices, compartments, time points, and assay scales commensurate.

Revised locations: Methods 2.8–2.9; Results 3.5; Discussion; Figure 2; Supplementary Method 4; Supplementary Table S8; Supplementary Figure S1.

3 Candidate prioritization

Reviewer comment: Please clarify whether the candidate framework was developed before the final results were known and how post hoc selection bias was limited.

Response: We now state explicitly that candidate prioritization was post hoc. After formalization, the same rule was applied to every microRNA. A candidate required direct recurrence evidence in at least two analytically distinct samples plus support in at least one additional clinical, mechanistic, analytical, or external-evidence domain. Effect magnitude, direction, statistical significance, pooled-model membership, and title or abstract order were not used. Candidate-specific certainty was assessed separately and is very low for all six candidates.

Revised locations: Methods 2.7; Results 3.4 and 3.6; Supplementary Method 5; Supplementary Tables S3, S5, S7, and S11.

4 microRNA nomenclature

Reviewer comment: Please standardize mature-arm nomenclature while retaining original study terminology when necessary.

Response: The main synthesis now uses mature-arm-specific names when assay documentation supports assignment. Legacy arm-unspecified terms are shown in parentheses, for example, miR-21-5p (legacy miR-21). Original study labels remain in the detailed supplementary records when mature-arm identity cannot be verified. We do not treat an arm-unspecified legacy label as confirmation of a mature species.

Revised locations: Abstract; Results 3.4; Table 3; Supplementary Tables S5 and S7.

5 Cohort overlap

Reviewer comment: Please describe the criteria used to distinguish independent cohorts from overlapping reports.

Response: We added the linkage criteria to the main Methods and Supplementary Method 2. Reports were compared by author group and institution, recruitment site and period, sample size and AF phenotype, ablation method, biomarker platform, and follow-up characteristics. Shared or probably shared reports received common program and recurrence-cohort identifiers. Uncertain linkage remained classified as probable. This procedure prevented an independent cohort from receiving duplicate quantitative weight.

Revised locations: Methods 2.4; Supplementary Method 2; Supplementary Table S4.

6 Abstract wording

Reviewer comment: Please revise ‘Six candidates were identified’ so that the sentence does not imply established biomarkers.

Response: The Abstract now states that six microRNAs were identified as priorities for prospective validation. The same cautious terminology is used in the Results, Discussion, tables, Figure 3, and graphical abstract.

Revised locations: Abstract; Results 3.4; Table 3; Figure 3; Conclusions; graphical abstract.

7 Recurrence ascertainment

Reviewer comment: Please summarize the main differences in recurrence definitions and rhythm-monitoring strategies and their practical implications.

Response: We added a concise Results summary. Recurrence definitions ranged from AF alone to AF or another atrial tachyarrhythmia after study-specific blanking periods. Surveillance intensity and the use of scheduled, ambulatory, or symptom-triggered assessment differed among reports. Intermittent surveillance can miss asymptomatic episodes and can therefore change event detection and prognostic estimates. These differences remain sources of clinical heterogeneity.

Revised locations: Methods 2.5; Results 3.2; Discussion 4.2.

8 Risk of bias presentation

Reviewer comment: Please add representative examples of the limitations that led to risk-of-bias concerns.

Response: The main Results now give representative examples: incomplete reporting of preanalytical handling or normalization, limited adjustment for established clinical predictors, recurrence ascertainment based on intermittent rather than continuous monitoring, incompletely specified adjustment sets, and selective or internally inconsistent model reporting. Eligibility remained independent of QUIPS judgments.

Revised locations: Results 3.3; Supplementary Figure S3; source-verification audit.

9 Statistical terminology

Reviewer comment: Please use consistent random-effects terminology and report the exact estimator and model assumptions.

Response: We standardized the term ‘random-effects model’ for the secondary sensitivity analyses. The statistical methods specify restricted maximum likelihood, modified Hartung–Knapp confidence intervals, and t-based prediction intervals, with Q, I², and τ² used to describe heterogeneity. ORs and HRs remain separate. These analyses are explicitly post hoc and exploratory.

Revised locations: Methods 2.9; Supplementary Method 4; Supplementary Table S8.

10 Prediction intervals

Reviewer comment: Please emphasize that the wide prediction intervals imply uncertainty about the direction and magnitude of association in a new population.

Response: We added this interpretation to the Discussion. The OR prediction interval is 0.04–16.69 and the HR prediction interval is 0.04–46.36. Both cross 1 widely, so neither model provides a stable expectation for the direction or magnitude of association in a new population.

Revised locations: Discussion; Supplementary Table S8; Supplementary Figure S1.

11 Table 2 comparability

Reviewer comment: Please add a footnote explaining that ORs from different microRNAs and assay scales are not directly comparable.

Response: The Table 2 note now states that the microRNAs, matrices, sampling times, exposure units, and assay scales are not commensurate. Numerical effects cannot be compared directly across rows and do not represent a common biological class effect.

Revised locations: Table 2 note.

12 Mechanistic interpretation

Reviewer comment: Please distinguish experimentally demonstrated mechanisms from computational or pathway-level hypotheses consistently.

Response: We revised the mechanistic synthesis and re-audited the cited references. Experimental target or pathway evidence, computational enrichment, phenotype-level association, and source or carrier inference are now described separately. The revised wording preserves this distinction for miR-21, miR-29b, miR-150, miR-451a, miR-10b-5p, and miR-320a.

Revised locations: Table 3; Discussion 4.1; Supplementary Table S7; Figure 3.

13 Clinical utility

Reviewer comment: Please state that future incremental-value analyses should compare microRNA models with established clinical and imaging predictors in prespecified development and external-validation cohorts.

Response: We added this requirement. Future studies should compare prespecified microRNA models with established clinical and imaging predictors, use separate development and external-validation cohorts, avoid data-driven biomarker or cut-point reselection, and report calibration, discrimination, reclassification, and decision-curve performance.

Revised locations: Discussion 4.3; Conclusions; Figure 3; graphical abstract.

14 Language and presentation

Reviewer comment: Please improve readability in Sections 2.7–2.9 and 4.1–4.3.

Response: We edited these sections to separate candidate selection, molecular annotation, effect selection, statistical sensitivity analyses, mechanistic interpretation, analytical validity, and clinical validation. The revision shortens several dense sentences without changing numerical results or the level of certainty.

Revised locations: Methods 2.7–2.9; Discussion 4.1–4.3.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript addresses an important question: whether circulating microRNAs can help predict atrial fibrillation recurrence after catheter ablation. Its principal strength is the distinction between recurrence-specific prognostic evidence, longitudinal changes after ablation, and mechanistic plausibility. The broad search, inclusion of Chinese-language publications, independent screening and extraction, and explicit handling of overlapping cohorts strengthen the review. The authors also appropriately acknowledge the very low certainty of the evidence and distinguish biomarker association from demonstrated clinical utility. Consequently, the candidate-specific evidence map provides a useful foundation for future research, particularly by highlighting uncertainties concerning sample processing, molecular identity, and the biological source of circulating signals.

The principal methodological weakness is the pooling of biologically distinct microRNAs measured using non-commensurate exposure scales, sampling compartments, and time points. These differences undermine the interpretability of the pooled estimates, beyond simply producing statistical heterogeneity. Combining positive and negative associations across different molecules can yield a near-null estimate without meaningfully addressing any individual biomarker’s prognostic value; random-effects modelling cannot resolve this conceptual problem. Selecting one biomarker per cohort according to its appearance in the title or abstract prevents duplicate weighting but introduces a selection rule based on reporting prominence rather than biological or clinical relevance. Likewise, folding estimates toward the adverse profile produces an upward-biased summary by construction and adds limited information despite the authors’ explicit caveats. The manuscript would be stronger if it omitted these pooled summaries or relegated them to supplementary material, prioritizing candidate-specific synthesis wherever sufficient comparable evidence exists.

The selection of six candidates also requires greater justification. The post hoc requirement of recurrence evidence plus support in only one additional domain is permissive, and the main text does not clearly establish why these candidates qualify while others do not. Candidate-specific certainty assessments would better support the review’s primary objective than certainty ratings limited to the heterogeneous pooled analyses. Several internal inconsistencies also need correction: Table 3 describes increased miR-451a abundance as associated with recurrence, whereas the discussion describes lower abundance, and multiple mechanistic citations in that table do not match the listed references—for example, the miR-320a statement cites a decision-curve methodology paper. Overall, the manuscript offers a potentially valuable research synthesis, but substantial revision is needed to make its candidate selection reproducible, correct its internal discrepancies, and align its quantitative presentation with its appropriately cautious conclusions.

Author Response

Circulating MicroRNAs as Molecular Biomarkers of Atrial Fibrillation Recurrence After Catheter Ablation

General response. We thank the reviewer for identifying the conceptual limitation of pooling biomarkers across different biomarkers and the internal inconsistencies in the presentation of candidates. We revised the manuscript so that candidate-specific evidence mapping is the primary synthesis. For verification, changes made specifically in response to Reviewer 2 are shown in red, whereas subsequent changes made in response to Reviewer 1 are shown in blue.

Color convention. When the two reviewers requested the same or closely related correction, the passage retains the color of the reviewer whose comment first prompted the revision. Only additional wording introduced specifically for Reviewer 1 is blue; text already revised for Reviewer 2 remains red. This convention is intended solely to facilitate editorial and author verification and does not alter the clean final version that will be prepared after acceptance.

1. Cross-biomarker pooling and the former adverse-profile fold

Reviewer comment: The reviewer noted that biologically distinct microRNAs measured on non-commensurate scales cannot yield a clinically interpretable class effect, that title/abstract-order selection was based on reporting prominence, and that outcome-informed folding was upward by construction.

Response: We agree. Pooled OR and HR estimates were removed from the Abstract and from the primary interpretation in the Results and Discussion. Figure 2 now displays only study-level estimates and contains no pooled diamond. The adverse-profile fold was removed completely. Cross-biomarker random-effects models are retained only in the Supplementary Material as explicitly labeled sensitivity analyses. We replaced the former title/abstract-order rule with a reviewer-prompted post hoc hierarchy: author-designated primary analysis, then the most fully adjusted compatible estimate, then pre-ablation/prespecified recurrence when otherwise equivalent. If no unique estimate remained, the cohort was excluded from the cross-biomarker model rather than selected arbitrarily. The revised OR sensitivity set contains 9 cohorts (n = 1162; OR 0.84, 95% CI 0.30–2.31; I² = 90.7%; prediction interval 0.04–16.69). The HR sensitivity set remains 3 cohorts (n = 364; HR 1.36, 95% CI 0.66–2.78; I² = 68.5%; prediction interval 0.04–46.36). These values are not interpreted as biological class effects. The full decision audit is provided in Supplementary Table S10.

2. Selection of the six candidates

Reviewer comment: The reviewer considered the former requirement of recurrence evidence plus one additional domain too permissive and requested a reproducible explanation of why the six candidates qualified.

Response: We agree and made the post hoc nature of the framework more explicit. The revised rule requires direct recurrence evidence in at least two analytically distinct samples—independent cohorts or separately recruited derivation and validation phases—plus support in at least one additional clinical, mechanistic, analytical, or external-evidence domain. The same rule was applied to all microRNAs after it was formalized. Effect magnitude, direction, statistical significance, pooled-model inclusion, and title/abstract order were not used. The operational criteria are reported in Methods Section 2.7 and Supplementary Table S5; the candidate-level evidence is reported in Table 3 and Supplementary Table S7.

3. Candidate-specific certainty

Reviewer comment: The reviewer recommended certainty assessments centered on individual candidates rather than heterogeneous pooled analyses.

Response: We agree. The OR- and HR-stratum GRADE profiles were replaced by candidate-specific GRADE assessments for miR-21-5p/miR-21, miR-150-5p/miR-150, miR-29b-3p/miR-29b, miR-451a, miR-10b-5p, and miR-320a/miR-320a-3p. Observational prognostic evidence began at high certainty under GRADE guidance for prognostic-factor research. Candidate-specific downgrading considered risk of bias, inconsistency, indirectness, imprecision, and suspected publication bias. Mechanistic plausibility was not used to upgrade clinical prognostic certainty. Final certainty is very low for all six candidates, with candidate-specific reasons reported in Supplementary Table S11 and summarized in Results Section 3.6.

4. Direction of miR-451a

Reviewer comment: The reviewer identified an inconsistency between Table 3 and the Discussion regarding whether higher or lower miR-451a abundance was associated with recurrence.

Response: Corrected. Table 3, Supplementary Table S7, and the Discussion now state: ‘Lower left-atrial miR-451a abundance was associated with recurrence, whereas greater scar burden was positively associated with recurrence; a second cohort reported an adjusted inverse association [28,37].’ This wording separates miR-451a abundance from scar burden and avoids the erroneous phrase ‘increased abundance in the lower left atrium.’

5. Mechanistic citations in Table 3 / Supplementary Table S7

Reviewer comment: The reviewer identified mismatched mechanistic citations, including a miR-320a statement that cited a decision-curve methodology paper.

Response: We re-audited every mechanistic statement in Table 3 and Supplementary Table S7. The corrected citation mapping is: miR-21/Spry1-supported profibrotic signaling [27,62]; miR-150/c-Myb [65]; miR-29b extracellular-matrix and PDGF-B evidence [63,64]; miR-451a hemolysis susceptibility [57,58]; miR-10b-5p computational enrichment [36]; and the extracellular miR-320a-3p remodeling association [66]. Reference [67] is now restricted to decision-curve methodology in the clinical-validation section. Mechanistic wording now distinguishes experimental support, computational enrichment, phenotype-level association, and unresolved source/carrier inference.

Files revised

• Main manuscript: revised Abstract, Methods 2.6–2.9, Results 3.2–3.6, Discussion, Conclusions, Tables 1–3, Figures 2–3, and supplementary-material description.
• Supplementary material: revised Methods, candidate criteria, Tables S7–S8 and S10–S11, and Supplementary Figures S1–S2; obsolete adverse-profile, meta-regression, funnel-plot, and former selection-rule analyses removed.
• Graphical abstract: rebuilt to emphasize candidate-specific evidence and prospective validation without a pooled class-effect estimate.