Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

General Comments: This is a robust and well-powered study that addresses a clinically significant question regarding the optimal targets for arterial carbon dioxide in the management of Acute Respiratory Failure 

The following clarifications are needed

The text mentions extracting the maximum PaCO2 value within 48 hours, but also states that a time-weighted average PaCO2 was calculated. It is unclear which of these was used as the primary exposure variable in the analysis

The manuscript treats ARF as a single entity, which is not appropriate. The physiological targets for PaCO2 vary drastically depending on the etiology (e.g., ARDS vs. COPD vs. Neuromuscular failure). A patient with COPD might have a baseline normal PaCO2 far higher than the study's threshold. Please do a sub-group analysis based on aetiology of ARF

The discussion should address how these findings align with current ARDS guidelines that advocate for permissive hypercapnia. Given that the "safe" range extends to nearly 58mmHg, your data may actually support the safety of this practice, which would be a high-impact takeaway

Figure 1 shows that a large number of patients were excluded. A brief discussion or a table comparing the characteristics of included vs. excluded patients would help address concerns regarding selection bias

As a retrospective study, it cannot account for the intent of the clinicians. It is impossible to distinguish between "accidental" hypercapnia and "intentional" permissive hypercapnia, which is a standard of care in lung-protective ventilation. This can be discussed in the limitations of the study

Author Response

Comments 1: The text mentions extracting the maximum PaCO2 value within 48 hours, but also states that a time-weighted average PaCO2 was calculated. It is unclear which of these was used as the primary exposure variable in the analysis。

Response 1:Thank you for pointing this out.We fully agree that the primary exposure variable must be unambiguous. In the revised manuscript we have now explicitly clarified that the time-weighted average (TWA) PaCO₂ during the mechanical ventilation was used as the main exposure.Specific changes:(1).Methods: added the sentence :The primary exposure was the time-weighted averages PaCO₂ (TWA-PaCO₂) during mechanical ventilation in each patient(page 3, lines 122–124). (2).The exact formula is given in the same paragraph  ;n represented the amount of PaCO₂ measurements during each patient's ventilation time.(page 4, lines 127–130).  (3).We have added the corresponding references that used an identical TWA approach for variables in critically ill patients (refs[11] , page 13, lines 443–444). We believe these revisions remove any ambiguity and thank the reviewer for improving the clarity of our work. Should any further detail be required, we will be happy to supply it.

 

Comments 2:The manuscript treats ARF as a single entity, which is not appropriate. The physiological targets for PaCO2 vary drastically depending on the etiology (e.g., ARDS vs. COPD vs. Neuromuscular failure). A patient with COPD might have a baseline normal PaCO2 far higher than the study's threshold. Please do a sub-group analysis based on aetiology of ARF.

Response 2:Thank you for your valuable suggestion that "acute respiratory failure should be stratified by etiology." Following your advice, we categorized patients into subgroups based on the primary causes of acute respiratory failure, using admission diagnoses and ICD codes provided by the database: ① ARDS, ② COPD, ③ severe pneumonia, ④ cerebrovascular diseases, and ⑤neuromuscular/central respiratory failure. Cerebrovascular diseases include stroke, encephalitis, meningitis, etc.; neuromuscular failure diseases include myasthenia gravis, Guillain-Barré syndrome, amyotrophic lateral sclerosis, etc. During data extraction, the number of cases of neuromuscular failure diseases was relatively small and could not be subjected to statistical analysis. Subsequently, within each subgroup, patients were still divided into three groups—hypocapnia, normocapnia, and hypercapnia—using the same PaCO₂cutoff values (36.4 and 57.9 mmHg), and the 28-day survival curves were redrawn (now presented as Supplementary Figure S1).

The results showed:

ARDS subgroup (n = 4884): The survival rate in the normocapnia group was significantly better than in the hypocapnia and hypercapnia groups (log-rank P < 0.001).

COPD subgroup (n = 2485): The survival curve of the hypercapnia group was better than that of the hypocapnia and normocapnia groups (P = 0.035). This aligns with your comment in the review that "patients with COPD may have a normal baseline PaCO₂ significantly higher than the thresholds studied."....

Given that the subgroup analysis results are consistent with the main analysis direction, we did not modify the original conclusion. Instead, we added the following statement in the discussion:"After stratification by etiology, except for COPD, the association direction between PaCO2 and prognosis was consistent for ARDS, severe pneumonia, and cerebrovascular diseases, supporting 36.4 - 57.9 mmHg as a safe target range for acute respiratory failure patients on mechanical ventilation." (page 10, lines 273-276).

Comments 3:The discussion should address how these findings align with current ARDS guidelines that advocate for permissive hypercapnia. Given that the "safe" range extends to nearly 58mmHg, your data may actually support the safety of this practice, which would be a high-impact takeaway。

Response 3:We thank the reviewers for their valuable suggestions. We have supplemented the discussion and compared it with current guidelines, with the key points as follows:  

(1) The upper limit of our range (57.9 mmHg) significantly exceeds the typical thresholds of concern in many clinical settings. This finding provides retrospective observational support for the implied safety range suggested by landmark trials establishing lung-protective ventilation. For example, the ARMA trial demonstrated a mortality benefit with low tidal volume ventilation, where the intervention group frequently had PaCO₂ levels exceeding 50 mmHg, and a predefined protocol for managing respiratory acidosis was only triggered at pH < 7.30 [30]. Our data indicate that within this range, there was no increased mortality risk up to approximately 58 mmHg, consistent with the safety profile observed in that pivotal trial. (2) Furthermore, a post hoc analysis of the LUNG SAFE cohort specifically investigating hypercapnia found that mild to moderate hypercapnia (PaCO₂ up to 55 mmHg) on the first day of ARDS was not associated with increased mortality, further reinforcing the view that clinically tolerable hypercapnia encompasses higher values than traditionally assumed [31].(3) The 2023 ESICM Adult ARDS Guidelines do not set an upper limit for PaCO₂, emphasizing only that lung-protective ventilation can continue as long as pH ≥ 7.20 and hemodynamic stability is maintained [32]. (4) Similarly, a recent multicenter randomized controlled trial (RCT) in ventilated neonates set targets of 60–75 mmHg and showed no increase in intraventricular hemorrhage or mortality [33]. The RCS curve in our study suggests 57.9 mmHg as a "safe" upper limit, which aligns closely with the reported tolerable ranges above. Thus, our data not only do not conflict with the "permissive hypercapnia" strategy but instead provide real-world evidence supporting its safety in patients with acute respiratory failure.

We have added corresponding statements in the discussion and cited the above guidelines and RCTs to highlight the clinical value of our study.this change can be found - page11 , lines 344-362.

[30]Acute Respiratory Distress Syndrome Network; Brower RG.;Matthay MA.; Morris A.;et al. Ventilation with lower tidal volumes as compared with traditional tidal volumes for acute lung injury and the acute respiratory distress syndrome. N Engl J Med. 2000,342(18):1301-8. doi: 10.1056/NEJM200005043421801. (page 14, lines 486–488)

[31]Madotto F.;Rezoagli E.; Pham T.;et al. LUNG SAFE Investigators and the ESICM Trials Group. Hyperoxemia and excess oxygen use in early acute respiratory distress syndrome: insights from the LUNG SAFE study. Crit Care. 2020,24(1):125. doi: 10.1186/s13054-020-2826-6.  (page 14, lines 489–491)

[32]Grasselli G.;Calfee CS.; Camporota L.; Poole D.;et al.ESICM guidelines on acute respiratory distress syndrome: definition, phenotyping and respiratory support strategies. Intensive Care Med. 2023 49(7):727-759. doi: 10.1007/s00134-023-07050-7.(page 14, lines 492–493)

[33]Travers CP.;Gentle SJ.;Shukla VV.;et al.Late Permissive Hypercapnia for Mechanically Ventilated Preterm Infants: A Randomized Trial. Pediatr Pulmonol. 2025,60(6):e71165. doi: 10.1002/ppul.71165.  (page 14, lines 494–495)

 

Comments 4:Figure 1 shows that a large number of patients were excluded. A brief discussion or a table comparing the characteristics of included vs. excluded patients would help address concerns regarding selection bias。

Response 4：Thank you for pointing this out. The issues you raised regarding "a large number of patients being excluded in the flowchart" and "the possibility of selection bias" have been briefly addressed in the discussion section of the revised manuscript. (page 12 lines 377-379).Detailed explanations are provided as follows:This study is a retrospective study, and the data source is a public database (MIMIC-IV) and（eICU-CRD） , which inherently limits the study design due to the structure and recording methods of the original data. The inclusion and exclusion criteria we set (duration of mechanical ventilation < 6 hours, length of hospital stay < 24 hours, repeated ICU admissions, and missing critical data) are based on considerations regarding the scientific validity of the research question and the completeness of the data. These criteria aim to ensure the representativeness of the study population and the reliability of the analysis results.

Although the final sample size included is relatively small, we believe that these exclusion criteria help reduce the impact of confounding factors on the study conclusions. For example, patients with excessively short duration of mechanical ventilation or length of hospital stay may not have reached a stable pathophysiological state; their PaCO₂ levels may be influenced by factors unrelated to the disease itself. Repeated ICU admissions carry the risk of time-dependent bias. Additionally, missing critical data directly affects the accuracy of defining the primary exposure and outcome variables.

We also recognize that these strict inclusion and exclusion criteria may limit the representativeness of the study population to the overall acute respiratory failure patients. Therefore, in the discussion section, we have clearly pointed out that the extrapolation of the conclusions of this study should be done with caution. Moreover, future research could further relax the inclusion and exclusion criteria or use methods such as multiple imputation to handle missing data; this would help validate the robustness of the findings of this study.

Comments 5:As a retrospective study, it cannot account for the intent of the clinicians. It is impossible to distinguish between "accidental" hypercapnia and "intentional" permissive hypercapnia, which is a standard of care in lung-protective ventilation. This can be discussed in the limitations of the study。

Response 5：Thank you for pointing this out. We agree with this comment. And acknowledge this limitation of our study. In the revised manuscript, we have included a discussion in the limitations section that highlights the inability to differentiate between accidental and intentional hypercapnia due to the retrospective nature of our analysis。Specific changes: “We could only rely on arterial blood gas values extracted from electronic health records and had no access to clinicians’ real-time therapeutic intent, meaning we could not determine whether elevated carbon dioxide levels were intentional or not. Specifically, our database could not differentiate unintentional hypercapnia from intentional permissive hypercapnia”.(page 12, lines 370–375)

We appreciate this important point and believe it adds depth to our discussion of the findings.

Reviewer 2 Report

Comments and Suggestions for Authors

Thankyou for the opportunity to review your manuscript. 

Although it has merit and significant patient numbers, there are a number of concerns with the manuscript.

My major concern is that the paper overstates the findings given that this a retrospective methodology. 

abstract

a very concise and succinct summary. authors should be congratulated for this. minor point would be to include some metrics when stating survival rate in the results rather than just subjectively stating 'highest survival rates' or 'lowest survival rates'. The conclusion is slightly incorrect given this is retrospective - you have not proven that if you maintain PaCO2 in the optimal range that survival will improve. You have only proven that those with a normal PaCO2 have better outcomes. Some confounders will include how refractory PaCO2 was to therapeutic measures, missed diagnosis, safety events etc. 

Introduction

Line 53 "For instance, arterial blood gas parameters, particularly the arterial partial pressure of carbon dioxide (PaCO2), have shown substantial influences on the mortality." needs a reference - also, you need to justify why your retrospective study is needed (and relevant) if a prospective study that this eludes to has already proven benefit?

Line 55 "Prior studies have reported that both low and high extremes of PaCO2 are associated with worse outcomes in ARF patients, particularly those receiving mechanical ventilation" - so how is your study different to these? why is it important that you perform the study too - please justify in your introducttion - you briefly mention in your next sentence but it is unclear how the previous study (ref 4,5) dont do this. 

Your introduction in general is very brief. it would be good to understand some of the underlying physiology as to why targeting PaCO2 as compared with other parameters such as PaO2, or even lactate, should be singled out. And what about non-blood gas parameters, how does targeting PaCO23 compared to say transcutaneous SpO2 compare?

Materials and methods

It is hard to understand the time period for each database. for example, was eICU really 10 years old data (2014-2015) compared with SRRSH and SFH being 2021-2024 and 2023-2025. MIMIC-IV is also unclear. 

Clear articulation of inclusion and exclusion criteria for the database extracts will be helpful for reproducibility and validation studies. 

Results

"The majority of patients in both databases were of Caucasian ethnicity." please include some metrics to define 'majority' so readers do not have to reference the table at that point in time and interrupt the reading. 

"implying a potential correlation between the length of mechanical ventilation and patient prognosis." is a discussion of results and should go in the discussion. 

figure 4. I am unable to read the writing in the figures as they are too small. 

discussion

"Furthermore, we verified the prognostic value of a PaCO2 range, which may help guide clinical decision-making and improve care strategies for patients with ARF" this is a dangerous statement as you have not validated the ranges or proven that it will improve outcomes. 

"The findings of this study have profound clinical implications." This whole paragraph is not proven. this is an overstatement given the retrospective nature. you have not proven that targeting a specific PaCO2 range improves outcomes, just that those who did have PaCO2 in that range had better outcomes. 

Conclusion

As per the abstract .

 

 

Author Response

abstract

Comments1: minor point would be to include some metrics when stating survival rate in the results rather than just subjectively stating 'highest survival rates' or 'lowest survival rates'.

Response1: Agreed. I have made changes to the statement regarding survival rates in the abstract section of the manuscript, adding specific indicators after the original 'highest survival rates' or 'lowest survival rates'. In the revised manuscript, this change can be found - (page 1, abstract, lines 35-36).

 

Comments2: The conclusion is slightly incorrect given this is retrospective - you have not proven that if you maintain PaCO2 in the optimal range that survival will improve. You have only proven that those with a normal PaCO2 have better outcomes. Some confounders will include how refractory PaCO2 was to therapeutic measures, missed diagnosis, safety events etc.  

Response2: Agree.I have clarified our conclusion to reflect the retrospective nature of our study. We have emphasized that our findings indicate an association between normal PaCO₂ levels and better outcomes, rather than a direct causal relationship. The revised conclusion now accurately represents the limitations of our study.   In the revised manuscript, this change can be found - (page 1, lines 40-44).

 

Introduction

Comment3: Line 53 "For instance, arterial blood gas parameters, particularly the arterial partial pressure of carbon dioxide (PaCO2), have shown substantial influences on the mortality." needs a reference-also, you need to justify why your retrospective study is needed (and relevant) if a prospective study that this eludes to has already proven benefit?

Response3: Agree.I have added the appropriate reference to support the statement regarding the influence of arterial blood gas parameters, particularly PaCO₂, on mortality. The reference has been included in the revised manuscript.This reference has also been included in the reference list[4]. (page 2, line 60)However, there is a lack of prospective studies related to the specific issue of the optimal PaCO₂ target value.To the best of our knowledge, no randomized controlled trial (RCT) or prospective interventional study has yet established an optimal PaCO₂ target in mechanically ventilated patients with acute respiratory failure (ARF).Thus, clinical practice continues to rely on expert opinion rather than high-level evidence.

[4]Zhang R, Chen H, Teng R, Li Z, Yang Y, Qiu H, Liu L. Association between the time-varying arterial carbon dioxide pressure and 28-day mortality in mechanically ventilated patients with acute respiratory distress syndrome. BMC Pulm Med. 2023, 23(1):129. doi: 10.1186/s12890-023-02431-6. (page 14, lines 427–429)

Comment 4:Line 55 "Prior studies have reported that both low and high extremes of PaCO2 are associated with worse outcomes in ARF patients, particularly those receiving mechanical ventilation" - so how is your study different to these? why is it important that you perform the study too - please justify in your introducttion - you briefly mention in your next sentence but it is unclear how the previous study (ref 4,5) dont do this. 

Response 4:Agree.I added a clear paragraph in the introduction that conveys the necessity and relevance of our retrospective analysis.Specific changes:”However, previous studies mainly focused on the impact of ventilation modes and intervention timing on survival rates or emphasized the correlation between dynamic changes in PaCO₂ and prognosis. This study aims to optimize ventilation strategies by setting stable target values for PaCO₂ to improve patient outcomes. There is currently no clear consensus on the optimal PaCO₂ target for mechanically ventilated patients. Determining the ideal range for PaCO₂ can provide valuable reference values for clinicians when implementing lung-protective ventilation.”(page 2, lines 62–68)

Comment 5:Your introduction in general is very brief. it would be good to understand some of the underlying physiology as to why targeting PaCO2 as compared with other parameters such as PaO2, or even lactate, should be singled out. And what about non-blood gas parameters, how does targeting PaCO2 compared to say transcutaneous SpO2 compare?

Response 5；Thank you for pointing this out. PaCO₂ directly reflects the efficiency of alveolar ventilation and is a standard parameter for assessing whether ventilation is adequate. It is inversely related to the alveolar ventilation volume (according to the alveolar ventilation equation), thus it can directly guide the adjustment of mechanical ventilation (such as the settings of tidal volume and respiratory rate) to maintain acid-base homeostasis and avoid under-ventilation or over-ventilation. PaO₂ or transcutaneous SpO₂ mainly reflects the oxygenation function, influenced by multiple factors such as the alveolar-arterial oxygen partial pressure difference, ventilation/perfusion ratio, and diffusion function, and cannot be used alone to assess ventilation. For example, in high flow oxygen therapy or mechanical ventilation, PaO₂ may improve due to increased inhaled oxygen concentration, but PaCO₂ may still rise, indicating inadequate ventilation. Although lactate levels can reflect tissue perfusion and metabolic status, they are non-specific indicators, affected by various factors such as shock, liver and kidney function, and metabolic diseases, and cannot be directly associated with ventilation efficiency.

Materials and methods

Comment 6: It is hard to understand the time period for each database. for example, was eICU really 10 years old data (2014-2015) compared with SRRSH and SFH being 2021-2024 and 2023-2025. MIMIC-IV is also unclear. 

Response 6:We appreciate your attention to the clarity of the timeframes for each database. In the revised manuscript, we have specified the exact timeframes for each database after its mention, outlining the specific timeframes used in our study for each database. It should be noted that the eICU database includes data collected from 2014 to 2015(page 2, line 84), while the SRRSH and SFH databases include data from 2021 to 2024 and 2023 to 2025, respectively. Additionally, we have specified that MIMIC-IV includes data from 2008 to 2021(page 2, line 90). This information has been included in the methods section to ensure clarity for the readers.

Results

Comments 7: "The majority of patients in both databases were of Caucasian ethnicity." please include some metrics to define 'majority' so readers do not have to reference the table at that point in time and interrupt the reading. 

Response 7: Thank you for pointing this out. We have revised the text to include specific metrics defining 'majority.' We now state that '75% in eICU-CRD and 57% in MIMIC-IV' which provides a clear understanding without requiring readers to reference the table. This change has been made in the Baseline characteristics section of the manuscript.this change can be found - (page 4, line 163).

Comment 8:"implying a potential correlation between the length of mechanical ventilation and patient prognosis." is a discussion of results and should go in the discussion. 

Response 8: Thank you for pointing this out.Based on your suggestion, we have removed that sentence from the "Results" section. Furthermore, to further explore this important observation, we have added a paragraph in the "Discussion" section (page 11, lines 317-326).It explains this reverse causal relationship: patients who died early did not have enough time at risk to accumulate ventilation days. In contrast, those who survived continued to receive mechanical ventilation, resulting in increased ICU length of stay. Other factors include refractory hypoxemia, complications related to mechanical ventilation, and delayed weaning from mechanical ventilation; these factors prolong ventilation duration and simultaneously increase mortality. Relevant literature [27] has been cited to support and deepen this discussion and has been added to the reference list.

[27]Roedl K.;Amann D.; Eichler L.; Fuhrmann V.;et al. The chronic ICU patient: Is intensive care worthwhile for patients with very prolonged ICU-stay (≥ 90 days)? Eur J Intern Med. 2019 ,69:71-76. doi: 10.1016/j.ejim.2019.08.024. (page 14, lines 480-481).

Comment 9:figure 4. I am unable to read the writing in the figures as they are too small. 

Response 9: Thank you for pointing this out. I have revised figure 4 to increase the font size of all text elements, ensuring that they are now easily readable. We appreciate your feedback and believe this enhancement will improve the clarity of the figure.

discussion

Comment 10:"Furthermore, we verified the prognostic value of a PaCO2 range, which may help guide clinical decision-making and improve care strategies for patients with ARF" this is a dangerous statement as you have not validated the ranges or proven that it will improve outcomes. 

Response:10 Thank you for your rigorous review of this key statement. You are absolutely correct that we did not verify the causal relationship or its effect on outcomes in this retrospective study, and as using the term "verified" is inaccurate and potentially misleading, we have carefully revised the wording based on your feedback. In the revised manuscript, the original sentence has been changed to: “Furthermore, we identified a PaCO₂ range associated with a more favorable prognosis. This finding may inform future clinical research and generates a hypothesis for optimizing care strategies in patients with ARF.”this change can be found - (page 10, line 296).

We replaced "verified" with "identified" to more accurately describe our work in identifying a range associated with prognosis. Additionally, we have softened "guide clinical decision-making" to "inform future clinical research" and changed "provides a hypothesis" to emphasize the exploratory and hypothesis-generating nature of this study, thereby avoiding causal inferences that exceed the evidence level of a retrospective study(page 10, lines 297-299). We believe these modifications make the statement more accurate and rigorous. We appreciate your help in enhancing the scientific quality of our paper.

Comment:11 "The findings of this study have profound clinical implications." This whole paragraph is not proven. this is an overstatement given the retrospective nature. you have not proven that targeting a specific PaCO2 range improves outcomes, just that those who did have PaCO2 in that range had better outcomes.

Response 11:Thank you for your rigorous oversight of this important statement. We fully agree with your opinion that, given that this study is a retrospective observational design, we indeed cannot confirm that its findings could directly lead to improvements in clinical outcomes; therefore, the original statement "profound clinical implications" is an overinterpretation. Based on your feedback, we have carefully revised the relevant paragraphs. Specifically, we have revised the original strong statement "The findings of this study have profound clinical implications." to read: "The findings of this study carry potential clinical implications".this change can be found - (page 11, line 339).

 

Conclusion

Comment 12:As per the abstract .

Response 12:Thank you very much for your crucial comments on the conclusion section. We completely agree that as a retrospective study, our analysis cannot establish causality and must adequately consider unmeasured confounding factors. Your specific corrections (such as the potential impact of treatment responsiveness, underdiagnosis, and other factors) have been extremely helpful in refining the logic and presentation of our conclusions. Based on your feedback, we have thoroughly rewritten the conclusion paragraph to ensure it strictly aligns with the study's design and level of evidence.

The revised conclusion is as follows: "In this retrospective analysis, we observed an association between arterial PaCO₂ levels within a specific range and increased survival rates. This was specifically noted in patients with acute respiratory failure on mechanical ventilation. Although this association remains significant after adjusting for available confounding factors, unmeasured factors such as treatment response, undiagnosed conditions or safety events may influence PaCO₂  levels and outcomes. Our findings suggest a hypothesis that targeted management of PaCO₂ may improve survival rates, but its validation in prospective studies is needed before clinical implementation."this change can be found -( page 12, lines 386-393).