Predictive Value of Epicardial Adipose Tissue Parameters Measured by Cardiac Computed Tomography for Recurrence of Atrial Fibrillation After Pulmonary Vein Isolation

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I read with interest the manuscript. The study is timely and relevant, as better tools for risk stratification are needed. Nevertheless, several aspects deserve further attention and improvement.

Major point

- The study relies on a relatively small, single-center retrospective cohort. This limitation should be more strongly emphasized, particularly with respect to the potential instability of regression models (you need 10 events for every variable included in the model to be stable).

- The stratified analyses (paroxysmal vs persistent AF) provide valuable insights but are clearly underpowered. The very high AUC values in some subgroups (e.g., persistent AF) may reflect sample-size artifacts rather than true discriminative performance. These findings should be presented as exploratory and hypothesis-generating, not as definitive evidence.

- The introduction currently frames CT as the primary imaging modality for EAT analysis. However, it would be mandatory to also mention the potential role of cardiac MRI, which allows a more detailed functional and structural characterization of the left atrium prior to ablation. In particular, atrial MRI has been used to study fibrosis and atrial remodeling, which are central to AF recurrence. I suggest including a brief note on this and citing 10.3390/jcdd12040114

- The discussion interprets EAT attenuation as a marker of local inflammation, but this remains a surrogate. Without histology or PET correlation, conclusions about biological mechanisms are speculative. At line 358 it would be valuable to highlight that different ablation techniques themselves may induce varying degrees of inflammation, which could in turn contribute to AF recurrence.

Minor point

- In the section discussing pharmacological interventions, the manuscript currently cites a meta-analyses on SGLT2 inhibitors (reference 34). To further strengthen this point, I recommend also citing the more recent and specifically arrhythmia-focused meta-analysis 10.1111/jce.16344.

Author Response

I read with interest the manuscript. The study is timely and relevant, as better tools for risk stratification are needed. Nevertheless, several aspects deserve further attention and improvement.

We sincerely thank the Reviewer for the positive appraisal of our manuscript and for recognizing the clinical relevance and timeliness of the study. We also greatly appreciate the constructive comments and suggestions, which we believe will help improve the quality and clarity of our work. Please find below our point-by-point responses to each of the issues raised.

Major point

- The study relies on a relatively small, single-center retrospective cohort. This limitation should be more strongly emphasized, particularly with respect to the potential instability of regression models (you need 10 events for every variable included in the model to be stable).

ANSWER: We thank the reviewer for highlighting the limitations related to the single-center, retrospective design and the relatively small sample size. We fully agree that these aspects restrict the generalizability of our findings and may contribute to the potential instability of regression models. In response, we have strengthened the Limitations section to explicitly acknowledge this concern. As noted in our reply to another reviewer, we applied regularization (LASSO) to select variables and restricted the final model to three clinically justified predictors. Furthermore, we performed cross-validation to mitigate the risk of overfitting. Nevertheless, we emphasize in the revised manuscript that these findings should be interpreted with caution and require confirmation in larger, multicenter prospective studies.

- The stratified analyses (paroxysmal vs persistent AF) provide valuable insights but are clearly underpowered. The very high AUC values in some subgroups (e.g., persistent AF) may reflect sample-size artifacts rather than true discriminative performance. These findings should be presented as exploratory and hypothesis-generating, not as definitive evidence.

ANSWER: Thanks to the reviewer for this important comment. We completely agree that the stratified analyses by AF type (paroxysmal vs. persistent) are based on groups that are too small to be statistically significant. The exceedingly high AUC values noted, especially in persistent AF, may actually indicate sample-size effects rather than strong discriminative performance. In the revised version of the manuscript, we have explicitly stated in the Discussion and Limitations sections that these subgroup findings should be interpreted as exploratory and hypothesis-generating rather than definitive evidence. We also stressed the importance of larger, multicenter studies to confirm these initial findings.

- The introduction currently frames CT as the primary imaging modality for EAT analysis. However, it would be mandatory to also mention the potential role of cardiac MRI, which allows a more detailed functional and structural characterization of the left atrium prior to ablation. In particular, atrial MRI has been used to study fibrosis and atrial remodeling, which are central to AF recurrence. I suggest including a brief note on this and citing 10.3390/jcdd12040114

ANSWER: We thank the Reviewer for this insightful comment. We agree that cardiac MRI plays an important role in the comprehensive evaluation of atrial structure and function, particularly for the assessment of atrial fibrosis and remodeling, which are known contributors to atrial fibrillation recurrence. As suggested, we have added a brief statement to the Introduction acknowledging the complementary role of cardiac MRI alongside CT-based approaches. We have also included the recommended citation (doi:10.3390/jcdd12040114) to support this addition.

- The discussion interprets EAT attenuation as a marker of local inflammation, but this remains a surrogate. Without histology or PET correlation, conclusions about biological mechanisms are speculative. At line 358 it would be valuable to highlight that different ablation techniques themselves may induce varying degrees of inflammation, which could in turn contribute to AF recurrence.

ANSWER: We thank the Reviewer for this insightful comment. We agree that EAT attenuation remains a surrogate marker of local inflammation, and without direct histological or PET validation, conclusions regarding the underlying biological mechanisms must remain speculative. In response, we have revised the Discussion to acknowledge this important limitation. Additionally, we incorporated the Reviewer’s suggestion by noting that different ablation modalities may elicit varying degrees of postprocedural inflammation, which could influence the risk of AF recurrence. These considerations are now explicitly stated in the revised manuscript to provide a more balanced interpretation of our findings.

Minor point

- In the section discussing pharmacological interventions, the manuscript currently cites a meta-analyses on SGLT2 inhibitors (reference 34). To further strengthen this point, I recommend also citing the more recent and specifically arrhythmia-focused meta-analysis 10.1111/jce.16344.

ANSWER: We appreciate the reviewer’s insightful suggestion. We agree that the recently published meta-analysis focused specifically on arrhythmia outcomes (doi:10.1111/jce.16344) provides additional and relevant support for the role of SGLT2i in reducing AF recurrence. We have now included this citation in the section discussing pharmacological interventions to strengthen the argument and provide a more comprehensive overview of the current evidence.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Congratulations to the authors for their manuscript dealing with such a relevant manuscript; I think that it contains many flaws that need to be corrected:

 

 

The multivariate logistic regression includes approximately ten predictors despite only 26 recurrence events. This contravenes the commonly accepted guideline of at least 10 events per predictor, raising concerns about model overfitting. The authors should therefore consider using regularization techniques such as LASSO regression or, alternatively, restrict the model to a smaller number of clinically relevant predictors.

The odds ratio for obesity (OR = 52.15; 95% CI: 0.45–6079.08) indicates model instability, likely due to sparse data or low event counts in certain subgroups. This result is not statistically robust and should be interpreted with caution or excluded from the final model.

The multivariate model does not include any performance metrics such as the area under the curve or calibration measures like the Hosmer–Lemeshow goodness-of-fit test. Including these statistics is essential to assess the reliability and generalizability of the model.

The analysis does not explore potential interaction effects for instance, between atrial fibrillation subtype (persistent or paroxysmal) and EAT parameters. Given that the ROC analysis is stratified by AF type, formal testing for interaction would add important nuance to the findings.

While univariate analysis demonstrates significant differences in LA-EAT volume between patients with and without AF recurrence, these differences were not retained in the multivariate model. This discrepancy warrants further explanation, possibly due to confounding factors or multicollinearity.

Moreover authors are encouraged to include in their discussion the latest evidences of different power settings in the context of AF ablation (doi: 10.1093/europace/euae265) and how this could be related to the epicardial adipose tissue.

Indeed the segmentation boundaries for LA-EAT, defined from the left atrial appendage to the coronary sinus, align with previously published methods; however, broader consensus and standardization are still lacking. This limitation should be more explicitly addressed in the Discussion section by the authors. The manuscript should also discuss the reproducibility of EAT attenuation measurements and the rationale for using the −50 to −200 Hounsfield Unit (HU) threshold. Some studies have used alternative values such as −190 HU or −30 HU, highlighting the need for standardization.

Author Response

Congratulations to the authors for their manuscript dealing with such a relevant manuscript; I think that it contains many flaws that need to be corrected.

We sincerely thank the reviewer for taking the time to read our manuscript and for acknowledging the relevance of the topic. Below, we provide point-by-point responses addressing each of the concerns raised, along with the corresponding revisions made to the manuscript.

 

- The multivariate logistic regression includes approximately ten predictors despite only 26 recurrence events. This contravenes the commonly accepted guideline of at least 10 events per predictor, raising concerns about model overfitting. The authors should therefore consider using regularization techniques such as LASSO regression or, alternatively, restrict the model to a smaller number of clinically relevant predictors.

ANSWER: We thank the reviewer for pointing out the ratio of events to predictors. We used logistic regression with an L1 penalty (LASSO) to choose the variables and limited the classical model to three clinically justified predictors: LA EAT attenuation, total EAT volume, and LA diameter. The model exhibited commendable discriminative capability (AUCCV = 0.876) and satisfactory calibration (Hosmer–Lemeshow p = 0.085). The outcomes align with the observed effects in the preliminary analysis and existing literature. We think that this simpler model lowers the chance of overfitting and follows your advice.

 

- The odds ratio for obesity (OR = 52.15; 95% CI: 0.45–6079.08) indicates model instability, likely due to sparse data or low event counts in certain subgroups. This result is not statistically robust and should be interpreted with caution or excluded from the final model.

ANSWER: We appreciate the reviewer's important point. We concur that the exceedingly broad confidence interval for obesity indicates model instability resulting from limited data within this subgroup. To prevent overfitting and erroneous conclusions, obesity was omitted from the final multivariable model. The modified model now comprises solely three predictors (LA-EAT attenuation, whole-EAT volume, and LA diameter) which exhibited stable estimates and reliable performance metrics (cross-validated AUC and calibration). The revised Results section now includes this change, and the manuscript now acknowledges the limitation of having sparse data.

 

- The multivariate model does not include any performance metrics such as the area under the curve or calibration measures like the Hosmer–Lemeshow goodness-of-fit test. Including these statistics is essential to assess the reliability and generalizability of the model.

ANSWER: Thank you for this valuable suggestion. We have now assessed the performance of the final multivariable logistic regression model.

 

- The analysis does not explore potential interaction effects for instance, between atrial fibrillation subtype (persistent or paroxysmal) and EAT parameters. Given that the ROC analysis is stratified by AF type, formal testing for interaction would add important nuance to the findings.

ANSWER: We appreciate the reviewer's helpful comment. We concur that examining the formal interaction between AF subtypes and EAT parameters would enhance the findings. However, due to the small number of events (26 recurrences) and the need to avoid overfitting the model, the current dataset does not have enough power for reliable interaction analysis. We stratified the ROC analysis by AF subtype to demonstrate possible differences. We have recognized this limitation in the revised manuscript and underscored the necessity for future studies involving larger cohorts to formally investigate such interactions.

 

- While univariate analysis demonstrates significant differences in LA-EAT volume between patients with and without AF recurrence, these differences were not retained in the multivariate model. This discrepancy warrants further explanation, possibly due to confounding factors or multicollinearity.

ANSWER: We are grateful that the reviewer brought this up. In univariate analysis, LA-EAT volume showed a strong link to AF recurrence. Nonetheless, this effect was not maintained in the multivariable model. We assert that this discrepancy is clarified by the correlations between LA-EAT volume and other predictors included in the model, specifically LA diameter and whole-EAT volume, both of which indicate aspects of atrial structural remodeling. The tests for multicollinearity showed that the VIF values were low (<1.4), which means that there was no collinearity that caused problems. But the fact that these variables had some information in common probably made it less clear how much LA-EAT volume contributed on its own. We have made this interpretation clearer in the new Results and Discussion sections, and we want to stress that LA-EAT volume is still clinically important, especially in stratified analyses by AF subtype.

 

- Moreover, authors are encouraged to include in their discussion the latest evidences of different power settings in the context of AF ablation (doi: 10.1093/europace/euae265) and how this could be related to the epicardial adipose tissue.

ANSWER: We appreciate the reviewer's helpful suggestion. We have now included recent evidence on different power settings in AF ablation into the Discussion. Emerging data indicate that customized energy delivery parameters may affect lesion durability, particularly in areas with thickened or inflamed epicardial adipose tissue. This aligns with our findings that increased LA-EAT attenuation and elevated EAT volume correlate with arrhythmia recurrence, as localized inflammatory or structural alterations may hinder the effectiveness of ablation lesions, even with optimized procedural techniques. We have updated the Discussion to emphasize how future ablation strategies, including personalized power settings, could alleviate the adverse effects of EAT on long-term outcomes.

 

- Indeed the segmentation boundaries for LA-EAT, defined from the left atrial appendage to the coronary sinus, align with previously published methods; however, broader consensus and standardization are still lacking. This limitation should be more explicitly addressed in the Discussion section by the authors. The manuscript should also discuss the reproducibility of EAT attenuation measurements and the rationale for using the −50 to −200 Hounsfield Unit (HU) threshold. Some studies have used alternative values such as −190 HU or −30 HU, highlighting the need for standardization.

ANSWER: We thank the reviewer for their helpful comment. We concur that the lack of a universally recognized standard for LA-EAT segmentation and attenuation thresholds constitutes a significant limitation. In the updated Discussion, we clearly state that there is no agreement on the anatomical boundaries used to define LA-EAT and stress the need for future work to create standardized methods. We also talk about how EAT attenuation is usually more reproducible than volume-based measurements, but different operators may still get different results. We used the −50 to −200 HU threshold for the attenuation range, which is in line with most previous studies and is widely used in fat tissue imaging. We recognize that other ranges (like −190 HU or −30 HU) have also been reported, which shows how important it is to standardize methods so that studies can be compared more easily. This conversation has been added to the new version of the manuscript.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Excellent improvement, I don't have other suggestions

Reviewer 2 Report

Comments and Suggestions for Authors

Congratulations to the authors for their great job in revising their manuscript according to my comments.