High Ultrafiltration Rate Is Associated with Increased All-Cause Mortality in Incident Hemodialysis Patients with a High Cardiothoracic Ratio

Round 1

Reviewer 1 Report

In the manuscript, Dr. Yang and coauthors assessed the association between UFR and all-cause mortality in incident hemodialysis patients, exhibited consistent association in patients with CTR > 50%, and concluded that heart failure with fluid overload followed by a high UFR may be the cause of mortality. The results presented is dedicated to interesting and important findings. However, the statistical analyses are insufficient supporting the conclusion. I have multiple concerns are as follows:

1.  In Results section, the authors presented that “UFR quintiles 5 and 1 were associated with a higher prevalence of all-cause mortality and cardiovascular mortality than UFR quintile 2”. But in Discussion section, the authors claimed that “a high UFR was linearly associated with a higher risk of all-cause mortality”. The results and the conclusion are contradictory. By roughly looking at the summary statistics in Table 1 and model estimates in Table 3, a quadratic relationship might be presented, especially in Table 3 that Q1 and Q5, but not Q3 and Q4, are significantly higher than Q2 given the constant hazard assumption for each UFR group. This finding seems inconsistent with previously reported linear association between URF vs. mortality such as reference cited #19. I would recommend the authors further exploring the association using appropriate modeling approaches, presenting additional evidence supporting current conclusion or providing explanation about the inconsistency.

2.     The interaction between UFR and categorical covariates has been tested with p values reported in Figure 1. Given multiple interactions are significant at 0.05 level, it would be suggested to also include significant interaction terms in the Cox models with covariates adjustment. In Table 4, different patterns were observed across subgroups as well. Authors may consider assessing the interaction, including interaction terms in the model as appropriate and cautiously interpreting the results from models without adjusting for significant interaction terms.

3.     Besides CHF, IDWG might be another confounding factor. The authors should clearly and explicitly discuss the potential impact on the analysis results without IDWG adjusted.

In addition to the above major concerns, a careful revision is recommended to improve the quality of the article addressing the minor concerns and questions listed below:

1.     Please consider rephrasing the sentence “As the UFR increased by 1mL/kg/h...” in Abstract, such as “As the UFR increased by 1 mL/kg/h, the risk of all-cause mortality increased XXX” or “As the UFR increased, the risk of all-cause mortality significantly increased.”

2.     I would suggest the author reporting the Kruskal-Wallis test p values instead of ANOVA test p value for continuous variables with skewed distribution.

3.     Please specify the model used for MCMC imputation in Method section

4.     Please provide the rationale of choosing Q2 as reference level in the analysis

5.     There are quite a bit grammatical errors and ambiguity throughout the manuscript, please try going through the manuscript and revise as much as possible.

Author Response

1.  In Results section, the authors presented that “UFR quintiles 5 and 1 were associated with a higher prevalence of all-cause mortality and cardiovascular mortality than UFR quintile 2”. But in Discussion section, the authors claimed that “a high UFR was linearly associated with a higher risk of all-cause mortality”. The results and the conclusion are contradictory. By roughly looking at the summary statistics in Table 1 and model estimates in Table 3, a quadratic relationship might be presented, especially in Table 3 that Q1 and Q5, but not Q3 and Q4, are significantly higher than Q2 given the constant hazard assumption for each UFR group. This finding seems inconsistent with previously reported linear association between URF vs. mortality such as reference cited #19. I would recommend the authors further exploring the association using appropriate modeling approaches, presenting additional evidence supporting current conclusion or providing explanation about the inconsistency.  

Response 1: We thank the comment on this detail. We have focused on the non-linear association between parameters and all-cause mortality in CKD for a long time (example: BMI and all-cause mortality: PLoS One. 2015 May 5;10(5):e0126668 J Clin Med. 2022 May 15;11(10):2787).

Because the observation of non-linear association between UFR and all-cause mortality in the landmark paper by Flythe et al in KI 2011 (citation #14), we did not assume a linear association in our study. 

In the new Table 3, Q1 was associated with an increased risk of all-cause mortality compared with Q2 in unadjusted model and in the fully-adjusted model with the concern of interaction. We tried to analyze the non-linear association between UFR and other parameters and did find that Q1, Q4, and Q5 were associated with hypoalbuminemia in minimally-adjusted model. But the explanation was still not clear. We believe it is better to present both continuous UFR and categorical UFR.

2. The interaction between UFR and categorical covariates has been tested with p values reported in Figure 1. Given multiple interactions are significant at 0.05 level, it would be suggested to also include significant interaction terms in the Cox models with covariates adjustment. In Table 4, different patterns were observed across subgroups as well. Authors may consider assessing the interaction, including interaction terms in the model as appropriate and cautiously interpreting the results from models without adjusting for significant interaction terms.

Response 2: Thanks for the comments. We did test the interactions according to the literatures and the variables in this cohort. The interactions were shown in Figure 2. We found CTR, gender and BW modified the association. We reported different Cox model including the interaction terms in the new Table 3 according to your comments and Reviewer 2’s comments. We also added p for interaction in Table 4.

3. Besides CHF, IDWG might be another confounding factor. The authors should clearly and explicitly discuss the potential impact on the analysis results without IDWG adjusted.

Response 3: Thanks for the comments. Currently, the association between IDWG and mortality is controversial. IDWG could be a marker for nutrition status, and higher IDWG has been shown to be associated with decreased mortality [DOI: 10.1159/000065228 ]. On the other hand, IDWG could be regarded as a reflection of fluid overload, and higher IDWG has been reported to be associated with higher mortality [DOI: 10.1111/sdi.12159]. By definition, UFR is the ratio of fluid removed (ultrafiltration, UF) to dialysis treatment time. At steady state, UF is equivalent to IDWG, and thus it might be reasonable to assume the effect of IDWG on clinical outcome was positively correlated to UFR. However, if patients with high IDWG could not tolerate the large fluid being removed, the UF may be less than IDWG, and thus these patients might be inappropriately allocated to the low UFR group. This mismatch could lead to biases. In our database, we did not record IDWG but only UF. To minimize the impact of single mismatch, we classified patients on the basis of the average UFR between 4th and 9th month after dialysis initiation. Through multiple adjustment of dry BW, the UF wound be close to IDWG. However, we admitted that without adjustment of IDWG could still cause biases, and we reported this in the Limitations section.

In addition to the above major concerns, a careful revision is recommended to improve the quality of the article addressing the minor concerns and questions listed below:

4. Please consider rephrasing the sentence “As the UFR increased by 1mL/kg/h...” in Abstract, such as “As the UFR increased by 1 mL/kg/h, the risk of all-cause mortality increased XXX” or “As the UFR increased, the risk of all-cause mortality significantly increased.”

Response 4: Thanks for the comment. We modified the sentence as “As the UFR increased by 1 mL/kg/h, the risk of all-cause mortality increased 1.6%.” in the Abstract, and “As the UFR increased by 1 mL/kg/h, the risk of all-cause mortality also significantly increased” in the Results section (Section 3.3).

5.  I would suggest the author reporting the Kruskal-Wallis test p values instead of ANOVA test p value for continuous variables with skewed distribution.

Response 5: Thanks for your comment. We did use KW test in skewed parameters, including cholesterol, triglyceride, intact PTH, ferritin and follow-up days. The results are shown in Table 1.

6.Please specify the model used for MCMC imputation in Method section

Response 6: The data were collected by dialysis centers for the quality control by Taiwan Society of Nephrology and only those dialysis centers with good dialysis quality will get 100% reimbursement from National Health Insurance. Thus, there were only 1% missing data from this database. Markov Chain Monte Carlo (MCMC) method was used to impute these missing values by considering all the variables in Table 1.

7. Please provide the rationale of choosing Q2 as reference level in the analysis

Response 7: To study the linear association between UFR and all-cause mortality, we investigate the HR of UFR per 1 ml/kg/h. To study the non-linear association between UFR and all-cause mortality, as our previous analysis, we use the group of lowest risk for all-cause mortality as the reference group. The new Table 3 also showed that Q2 had the lowest Hazard after the fully-adjusted model considering the interaction. The results were consistent with the landmark paper by Flythe et al in KI 2011 (citation #14).

8. There are quite a bit grammatical errors and ambiguity throughout the manuscript, please try going through the manuscript and revise as much as possible.

Response 8: We apologize for the grammatical errors in the manuscript. We tried our best to revise the manuscript to make it coherent and clear. The manuscript has also been proofread by native English speakers. However, if some grammatical errors still exist, please let us know so that we can correct them.

Reviewer 2 Report

1) Multivariate linear regression and Cox proportional hazards analysis are appropriate and they have been well conducted. However, the most important information is missing. There is no mention of the patients’ condition, nor of the treatments that they were receiving. Lack of echocardiographic diagnosis results, such as cardio-pulmonary function etc.

2) Authors should strongly justify the necessity to conduct the described research. This part of the introduction is insufficient. Both, in the introduction and in the discussion, the following paragraphs are often thematically unrelated. Authors should take care of the quality of the text. Lack of description of the statistical methodology employed in the manuscript and needs to be provided.

3) Inclusion and exclusion criteria should be specific.

4) To account for possible confounding factors, two or more varied models should be developed.

Author Response

1. Multivariate linear regression and Cox proportional hazards analysis are appropriate and they have been well conducted. However, the most important information is missing. There is no mention of the patients’ condition, nor of the treatments that they were receiving. Lack of echocardiographic diagnosis results, such as cardio-pulmonary function etc.

Response 1: Thanks for the comments. The data were collected by dialysis centers for the quality control by Taiwan Society of Nephrology. In this database, comorbidities, dialysis status, and laboratory results were well-documented. However, echocardiography was not mandatory for hemodialysis patient under the regulation of Taiwan Society of Nephrology. In contrast, annual CTR was a must in this database.

Among 2,615 incident hemodialysis patients in the current study, 212 patients in the dialysis centers affiliated to the university received echocardiography exam during follow up. We analyzed the correlation between CTR and echocardiographic parameters. In the univariate linear regression analysis, the factors positively associated with CTR were left atrial diameter, the ratio of early mitral valve flow velocity to mitral annulus early diastolic velocity (E/E’ ratio), left atrial volume index(LAVI), and left ventricular mass index(LVMI). The factor negatively associated with CTR was mitral annulus early diastolic velocity, which was a reflect of left ventricular diastolic function. In the multivariate linear regression analysis, only mitral annulus early diastolic velocity was negatively associated with CTR. Thus, it seems high CTR was correlated with LV diastolic dysfunction. This finding was added in the results section (Section 3.4) and Supplement Table 1.

We addressed this in the Limitations section as “Second, it is difficult to fully assess cardiac condition without functional or structural data from echocardiography. Since echocardiography was not mandatory for hemodialysis patient under the regulation of Taiwan Society of Nephrology, only a small proportion of patients received echocardiography in the current study. Without adjusting baseline cardiac status, some biases may exist.”

As for the treatment (medication or surgery) patients received, since these information was not recorded in the database, some biases could exist without adjusting these factors. We addressed these in the Limitations section.

2. Authors should strongly justify the necessity to conduct the described research. This part of the introduction is insufficient. Both, in the introduction and in the discussion, the following paragraphs are often thematically unrelated. Authors should take care of the quality of the text. Lack of description of the statistical methodology employed in the manuscript and needs to be provided.

Response 2: Thanks for the comment. We modified both the introduction and the discussion sections to make it more coherent and logical. We also added some details about the statistical methodology. (section 2.4)

3. Inclusion and exclusion criteria should be specific.

Response 3: Thanks for the comments. Inclusion criteria: patients underwent stable dialysis for more than 90 days, age more than 18 years, and recruited from three affiliated hospitals and nine associated hemodialysis clinics of Kaohsiung Medical University, southern Taiwan. Exclusion criteria: lost to follow-up in less than 6 months or more than 20% missing data.

4. To account for possible confounding factors, two or more varied models should be developed.

Response 4: Thanks for the comments. We reported different Cox model in Table 3 according to your and Reviewer 1’s suggestion. In Model 1, comorbidities were adjusted. In Model 2, comorbidities and lab data were adjusted. In Model 3, comorbidities, lab data, and interaction term were adjusted. The association between a high UFR and increased mortality remained consistent among all models.

Reviewer 3 Report

I would congratulate with the authors for the very good paper.  A high ultrafiltration rate (UFR) is associated with increased mortality in hemodialysis patients, and in this study, more than 2600 incident hemodialysis patients were categorized according to their initial cardiothoracic ratios (CTRs) to assess whether UFR was associated with mortality in patients with high or low CTRs. From results, high UFRs may be associated with increased all-cause mortality in incident hemodialysis patients with a high CTR but not in those with a low CTR. I currently find these results in clinical practice. Here you find some minor points in order to improve the manuscript

Authors did not discriminate at baseline between patients at higher risk of sudden cardiac death, such those with a cardiac ICD-device that may be very frequent in this patient population where the influence of age may also impact (please cite: doi: 10.1007/s40520-018-1088-5). At the same time since causal relationship between a high UFR and mortality could not be confirmed, ICD data would be interesting Again, authors did not considered patients affected by most common cardiyomyopaties at baseline (please cite DOI: 10.1016/j.ijcard.2022.03.028  : and doi: 10.1002/ejhf.1103). Authors should simply add in study limitations this important point, also including 3 suggested references, since they generically considered at  baseline: CHF, ischemic heart disease, and generically cardiovascular disease [CVD]

Authors should add a nice figure showing all mechanisms by which High ultrafiltration rate is potentially associated with increased all-cause mortality. This would be extremely important for readers

Author Response

1. Authors did not discriminate at baseline between patients at higher risk of sudden cardiac death, such those with a cardiac ICD-device that may be very frequent in this patient population where the influence of age may also impact (please cite: doi: 10.1007/s40520-018-1088-5). At the same time since causal relationship between a high UFR and mortality could not be confirmed, ICD data would be interesting Again, authors did not considered patients affected by most common cardiyomyopaties at baseline (please cite DOI: 10.1016/j.ijcard.2022.03.028  : and doi: 10.1002/ejhf.1103). Authors should simply add in study limitations this important point, also including 3 suggested references, since they generically considered at  baseline: CHF, ischemic heart disease, and generically cardiovascular disease [CVD]

Response 1: Thanks for the suggestion. We addressed this in the Limitations section. “Although we adjusted multiple comorbidities, cardiomyopathy was not taken into consideration. Dilated cardiomyopathy may lead to heart failure [39], and hypertrophic cardiomyopathy could lead to syncope, life-threatening arrhythmic events, and even sudden cardiac death [40]. Both could be confounding factors in the research of mortality in hemodialysis patients. No record or adjustment for cardiomyopathies may lead to biases.”

2. Authors should add a nice figure showing all mechanisms by which High ultrafiltration rate is potentially associated with increased all-cause mortality. This would be extremely important for readers

Response 2: Thanks for the suggestion. We add a figure in the introduction section to explain the possible mechanisms by which high UFR is associated with increased mortality.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Manuscript improved! Congratulations to authors