Prognostic Value of Nutritional Risk Scores in Septic ICU Patients: A Survival Analysis Using mNUTRIC, PNI, and CONUT

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Overall, the manuscript addresses the important question of whether nutritional risk scores have prognostic value in septic ICU patients. Its focus on survival analysis is a strength, and it includes clinically meaningful outcomes such as mortality, LOS, and mechanical ventilation. The main finding is that mNUTRIC demonstrated the strongest correlation with ICU mortality, whereas PNI distinguished between Kaplan–Meier curves and CONUT did not. However, there are important design, reporting and statistical issues with the current manuscript that must be corrected before it can be considered reliable enough for publication.

Abstract

The abstract requires substantial editing to ensure consistency with the main text. The abstract describes the study as retrospective observational (line 32), whereas the Methods section describes it as a prospective cohort study (line 150). This discrepancy must be resolved. Additionally, while the abstract states that survival analysis was central, the manuscript also presents day-3 reassessments, ROC analyses, LOS and MV outcomes. The primary endpoint should therefore be made explicit.

It should be clear from abstract that only 105 septic patients were included in the survival analysis and that the number of mortality events was limited. The conclusion should be toned down slightly: while the data support the idea that mNUTRIC had the strongest prognostic signal in this cohort, this does not definitively prove that it is the best tool for septic ICU patients in general.

Intro

It is too long and somewhat repetitive, only partially focusing on the actual study question. The final study aim is awkwardly phrased, mixing prognostic discrimination with multiple outcomes.

Do three things more clearly: define the primary hypothesis; justify why septic ICU patients specifically require validation of these scores; explain why it is clinically relevant to compare mNUTRIC with albumin-based tools, given the acute-phase response in sepsis.

Methods

The study design and setting are inconsistent. As it was mentioned above, the manuscript uses the terms “prospective cohort” in the Methods section and “retrospective observational” in the abstract. It later describes limitations as “single-centre”, yet patients appear to come from two hospitals/ICUs. This discrepancy must be resolved.

The inclusion criteria are problematic. Patients were only included if they were 'at risk of malnutrition' according to all three nutritional tools. This introduces strong selection bias and undermines the central claim that these scores are being compared as prognostic instruments. If the cohort is preselected using the predictors under investigation, the distributions and apparent performance of those scores become distorted. Either the authors must justify this design rigorously, or they must reanalyze an unselected septic ICU cohort.

 

Excluding patients who were unable to communicate or participate raises serious concerns in an ICU sepsis study, given that the sickest patients are often intubated, delirious, sedated or otherwise unable to give direct consent. This can produce serious selection bias towards less severe cases, unless proxy or waived consent procedures were used. The authors must clarify exactly how consent was obtained and how many eligible patients were excluded for this reason.

Although the methods section mentions adjusted Cox and logistic models, only the results of univariate Cox analysis are presented. Proportional hazards assumptions are not reported. The day-3 analyses are vulnerable to survivor bias and should be interpreted with caution.

The ROC analysis is not ideal for a time-to-event endpoint, such as mortality. For survival data, time-dependent ROC or concordance metrics would be more appropriate. Alternatively, the authors should justify their use of the simpler ROC approach and clearly state the exact mortality endpoint that the ROC is based on.

Results

The manuscript states that mechanical ventilation was more frequent among non-sepsis patients, but the table shows a higher frequency in the sepsis group. Similarly, a negative chi-squared value is reported, which is impossible.

mNUTRIC takes into account factors such as age, APACHE II, SOFA, comorbidity burden and pre-ICU hospital stay. Therefore, it is partly a composite of severity of illness, rather than a purely nutritional metric. Its superior prognostic performance may reflect the presence of embedded severity variables rather than nutritional state per se. This should be explicitly acknowledged in the Results and Discussion sections of the manuscript.

Discussion

The discussion still overstates the results. Due to the small sample size (only 24 deaths), the selection criteria and the absence of convincing multivariable adjustment, the paper should not draw too strong a conclusion that mNUTRIC is the superior prognostic tool. Instead, it should state that mNUTRIC showed the strongest prognostic association in this selected cohort.

The discussion should also explicitly acknowledge that mNUTRIC may outperform PNI/CONUT because it incorporates APACHE II and SOFA, which are powerful predictors of mortality in their own right. This is not just a limitation; it is central to interpreting the findings.

Limitations

Can be expanded after revision

Conclusion

The conclusion should be shortened and softened. It should avoid implying definitive superiority and external generalizability

Author Response

Author's Reply to the Review Report (Reviewer 1)

Dear Reviewer,

We would like to express our sincere gratitude for the time and effort you devoted to reviewing our manuscript. Your thoughtful evaluation, encouraging remarks, and constructive suggestions are deeply appreciated. Your feedback has been invaluable in improving the quality and clarity of our work, and we are truly grateful for your insightful contributions to this manuscript.

All the typing recommended changes were performed in the body of our manuscript, with the Track Changes function activated.

 

Comments and Suggestions for Authors

Overall, the manuscript addresses the important question of whether nutritional risk scores have prognostic value in septic ICU patients. Its focus on survival analysis is a strength, and it includes clinically meaningful outcomes such as mortality, LOS, and mechanical ventilation. The main finding is that mNUTRIC demonstrated the strongest correlation with ICU mortality, whereas PNI distinguished between Kaplan–Meier curves and CONUT did not. However, there are important design, reporting and statistical issues with the current manuscript that must be corrected before it can be considered reliable enough for publication.

Comments 1:

Abstract

The abstract requires substantial editing to ensure consistency with the main text. The abstract describes the study as retrospective observational (line 32), whereas the Methods section describes it as a prospective cohort study (line 150). This discrepancy must be resolved. Additionally, while the abstract states that survival analysis was central, the manuscript also presents day-3 reassessments, ROC analyses, LOS and MV outcomes. The primary endpoint should therefore be made explicit.

It should be clear from abstract that only 105 septic patients were included in the survival analysis and that the number of mortality events was limited. The conclusion should be toned down slightly: while the data support the idea that mNUTRIC had the strongest prognostic signal in this cohort, this does not definitively prove that it is the best tool for septic ICU patients in general.

Response 1:

We thank the reviewer for this important observation. We revised and believe that these revisions better align with your suggestions.

 

Comments 2:

Intro

It is too long and somewhat repetitive, only partially focusing on the actual study question. The final study aim is awkwardly phrased, mixing prognostic discrimination with multiple outcomes.

Do three things more clearly: define the primary hypothesis; justify why septic ICU patients specifically require validation of these scores; explain why it is clinically relevant to compare mNUTRIC with albumin-based tools, given the acute-phase response in sepsis.

Response 2:

We thank the reviewer for this valuable and constructive comment. The Introduction has been revised to improve clarity, focus, and logical flow. Redundant and overly general statements have been removed to better emphasize the study rationale.

Specifically, we have:

(1) Clearly defined the primary hypothesis, now stating that the mNUTRIC score is expected to demonstrate superior prognostic performance for 28-day ICU mortality compared with albumin-based indices (PNI and CONUT).

(2) Strengthened the justification for focusing on septic ICU patients, highlighting the unique pathophysiological features of sepsis, including profound inflammation and metabolic alterations, which may influence the performance of nutritional risk scores.

(3) Expanded the discussion on the limitations of albumin-based indices in sepsis, emphasizing the impact of the acute-phase response, capillary leakage, and fluid shifts on serum albumin levels, which may reduce their reliability as markers of nutritional status in this setting.

(4) Reformulated the study aim to clearly distinguish the primary outcome (28-day ICU mortality) from secondary endpoints, thereby improving precision and readability.

 

Comments 3:

Methods

The study design and setting are inconsistent. As it was mentioned above, the manuscript uses the terms “prospective cohort” in the Methods section and “retrospective observational” in the abstract. It later describes limitations as “single-centre”, yet patients appear to come from two hospitals/ICUs. This discrepancy must be resolved.

The inclusion criteria are problematic. Patients were only included if they were 'at risk of malnutrition' according to all three nutritional tools. This introduces strong selection bias and undermines the central claim that these scores are being compared as prognostic instruments. If the cohort is preselected using the predictors under investigation, the distributions and apparent performance of those scores become distorted. Either the authors must justify this design rigorously, or they must reanalyze an unselected septic ICU cohort.

Excluding patients who were unable to communicate or participate raises serious concerns in an ICU sepsis study, given that the sickest patients are often intubated, delirious, sedated or otherwise unable to give direct consent. This can produce serious selection bias towards less severe cases, unless proxy or waived consent procedures were used. The authors must clarify exactly how consent was obtained and how many eligible patients were excluded for this reason.

Although the methods section mentions adjusted Cox and logistic models, only the results of univariate Cox analysis are presented. Proportional hazards assumptions are not reported. The day-3 analyses are vulnerable to survivor bias and should be interpreted with caution.

The ROC analysis is not ideal for a time-to-event endpoint, such as mortality. For survival data, time-dependent ROC or concordance metrics would be more appropriate. Alternatively, the authors should justify their use of the simpler ROC approach and clearly state the exact mortality endpoint that the ROC is based on.

Response 3:

We thank the reviewer for this insightful suggestion. We revised the Methods section and believe that these revisions better align with your suggestions.

Methods 2.1 – Design and center

- Clarified that the study is a prospective observational bicentric (Filantropia Craiova + Spitalul Județean Craiova), with both ethics approvals explicitly mentioned. Single-center inconsistency removed.

Methods 2.2 – Inclusion criteria

- Added explicit paragraph: convergent multi-score design is justified, but score distributions are restricted, and prognostic performance may differ from unselected ICU populations. Results apply to patients with confirmed nutritional risk.

Methods 2.2 – Consent/exclusions

- Clarified that patients unable to consent directly were included by proxy consent (relative/legal representative). Mentioned that 12 eligible patients were excluded due to the unavailability of a proxy. Recognized as a potential limitation of severity.

Methods 2.3.4 – Statistical analysis: ROC

- Added explicit justification: ROC was applied exploratively for the mortality endpoint due to the limited number of events (n=24). Mentioned that Harrell concordance or time-dependent ROC would be preferred and that ROC results for mortality should be interpreted with caution.

Methods 2.3.4 – Cox: multivariable vs. univariate; PH hypothesis

- Clarified that Cox models are univariate for mNUTRIC, PNI, and CONUT (age is the only covariate in the sensitivity model), justified by structural collinearity. Added that the PH hypothesis was checked graphically (log-log plots) without significant violations. Added survivor bias warning for Day 3 analyses.

 

Comments 4:

Results

The manuscript states that mechanical ventilation was more frequent among non-sepsis patients, but the table shows a higher frequency in the sepsis group. Similarly, a negative chi-squared value is reported, which is impossible.

mNUTRIC takes into account factors such as age, APACHE II, SOFA, comorbidity burden and pre-ICU hospital stay. Therefore, it is partly a composite of illness severity rather than a purely nutritional metric. Its superior prognostic performance may reflect the presence of embedded severity variables rather than nutritional state per se. This should be explicitly acknowledged in the Results and Discussion sections of the manuscript.

Response 4:

We thank the reviewer for this insightful suggestion. We revised the Results section and believe that these revisions better align with your suggestions.

Results 3.1 – MV Day 1 error

- Corrected: "significantly more frequent in the septic group compared with non-septic patients (p < 0.0001), consistent with the higher severity of illness in the sepsis cohort."

Results 3.1 – MV Day 3 error

- Corrected: "significantly more frequent in the septic group compared with non-septic patients (8/50, 16.0%; p < 0.0001), reflecting greater severity and organ dysfunction burden in the sepsis cohort."

Results 3.1 – Negative chi-squared value

- Corrected to φ = 0.150 (phi coefficient). Added explanatory note: [Note: the value previously reported as −0.150 was a transcription error; chi-squared statistics are non-negative by definition; the correct value φ=0.150 reflects a small, non-significant association].

 

Comments 5:

Discussion

The discussion still overstates the results. Due to the small sample size (only 24 deaths), the selection criteria and the absence of convincing multivariable adjustment, the paper should not draw too strong a conclusion that mNUTRIC is the superior prognostic tool. Instead, it should state that mNUTRIC showed the strongest prognostic association in this selected cohort.

The discussion should also explicitly acknowledge that mNUTRIC may outperform PNI/CONUT because it incorporates APACHE II and SOFA, which are powerful predictors of mortality in their own right. This is not just a limitation; it is central to interpreting the findings.

Limitations

Can be expanded after revision

Response 5:

We thank the reviewer for this insightful suggestion. We revised the Discussion section and believe that these revisions better align with your suggestions.

Discussion – Overestimated superiority of mNUTRIC

- Paragraph completely rewritten: explicitly acknowledged that superior performance of mNUTRIC may reflect embedded severity (APACHE II, SOFA, age, comorbidities) rather than independent nutritional effect. The wording "superiority" was replaced with "strongest prognostic association in this selected cohort". Mentioned that multicenter studies with concordance statistics are needed to disambiguate these contributions.

Discussion – Opening

- Paragraph rewritten: explicitly stated that there are 105 septic patients with 24 events, that the results are hypothesis-generating rather than confirmatory, and that the composite structure of mNUTRIC (APACHE II + SOFA) is a central factor of interpretation.

Limitations – Extension

- Limitations extended with: (a) bicentric but regionally-limited; (b) 24 events = low statistical power; (c) selection effect of inclusion criterion with 3 simultaneous scores; (d) possible severity bias through consent exclusions; (e) survivor bias Day 3; (f) mNUTRIC contains APACHE II + SOFA = pure nutritional effect cannot be disambiguated; (g) exploratory ROC for time-to-event.

 

Comments 6:

Conclusion

The conclusion should be shortened and softened. It should avoid implying definitive superiority and external generalizability

Response 6:

We thank the reviewer for this insightful suggestion. We revised the Conclusion section and believe that these revisions better align with your suggestions

Conclusion rewritten: "exploratory findings from a selected, limited-event cohort"; avoided assertion of definitive superiority; added that the stronger signal of mNUTRIC likely reflects incorporated severity components; recommended multicenter studies with time-dependent metrics.

Reviewer 2 Report

Comments and Suggestions for Authors

1.⁠ ⁠Please strengthen the multivariable survival analysis.
The current survival analysis includes univariate Cox regression, but the manuscript would benefit from a more robust multivariable model to better assess the prognostic value of the nutritional scores. Since mNUTRIC already includes age, APACHE II, and SOFA as structural components, these variables should not be entered together in the same model. However, the authors could still perform an adjusted Cox model that includes mNUTRIC and variables not incorporated into the score, such as sex and mechanical ventilation, to provide a more convincing assessment of its independent prognostic value.

2.⁠ ⁠Please provide a clearer statistical comparison of the prognostic performance of the three nutritional scores.
Although ROC curves are presented for mNUTRIC, PNI, and CONUT, the manuscript does not formally compare their discriminative performance. The authors should clarify whether the observed differences in AUC values are statistically meaningful or, at a minimum, explicitly state that mNUTRIC showed the highest AUC among the evaluated scores. This would strengthen the central message of the study.

3.⁠ ⁠Please improve the description of the survival analysis in the Methods section.
The statistical methods section should explicitly describe the survival analysis procedure, including:
(i) the use of Kaplan–Meier curves,
(ii) the log-rank test for group comparisons, and
(iii) the definition of time-to-event (from ICU admission to death or censoring).
At present, this part of the methodology is not sufficiently clear and should be expanded to ensure reproducibility.

4.⁠ ⁠Please refine the interpretation of the Kaplan–Meier findings in the Discussion.
The Discussion should more clearly explain why PNI reached statistical significance in the Kaplan–Meier analysis, whereas mNUTRIC showed the strongest signal in the Cox regression and ROC analyses. A brief explanation that categorizing continuous variables may reduce statistical power, whereas Cox regression preserves more prognostic information, would improve the interpretation and make the results more coherent.

Author Response

Author's Reply to the Review Report (Reviewer 2)

Dear Reviewer,

We would like to express our sincere gratitude for the time and effort you devoted to reviewing our manuscript. Your thoughtful evaluation, encouraging remarks, and constructive suggestions are deeply appreciated. Your feedback has been invaluable in improving the quality and clarity of our work, and we are truly grateful for your insightful contributions to this manuscript.

All the typing recommended changes were performed in the body of our manuscript, with the Track Changes function activated.

Comments and Suggestions for Authors

Comments 1:

1.⁠ ⁠Please strengthen the multivariable survival analysis.
The current survival analysis includes univariate Cox regression, but the manuscript would benefit from a more robust multivariable model to better assess the prognostic value of the nutritional scores. Since mNUTRIC already includes age, APACHE II, and SOFA as structural components, these variables should not be entered together in the same model. However, the authors could still perform an adjusted Cox model that includes mNUTRIC and variables not incorporated into the score, such as sex and mechanical ventilation, to provide a more convincing assessment of its independent prognostic value.

Response 1:

We thank the reviewer for this important suggestion. We agree that a multivariable analysis provides a more robust evaluation of the prognostic value of the mNUTRIC score. As correctly noted, age, APACHE II, and SOFA are structural components of the mNUTRIC score and were therefore not included in the same model to avoid structural collinearity.

Following the reviewer’s recommendation, we performed an additional adjusted Cox proportional hazards regression analysis including variables not incorporated into the mNUTRIC score, namely sex and mechanical ventilation. In this model, the mNUTRIC score showed a trend toward association with ICU mortality (HR 1.39, 95% CI 0.95–2.04, p = 0.087), while mechanical ventilation was independently associated with mortality (HR 3.30, 95% CI 1.38–7.88, p = 0.007).

These findings further support the role of mNUTRIC as a clinically relevant prognostic score, although its effect appears to be partly influenced by the severity of illness reflected by the need for mechanical ventilation.

 

Comments 2:

2.⁠ ⁠Please provide a clearer statistical comparison of the prognostic performance of the three nutritional scores.
Although ROC curves are presented for mNUTRIC, PNI, and CONUT, the manuscript does not formally compare their discriminative performance. The authors should clarify whether the observed differences in AUC values are statistically meaningful or, at a minimum, explicitly state that mNUTRIC showed the highest AUC among the evaluated scores. This would strengthen the central message of the study.

Response 2:

We thank the reviewer for this helpful suggestion. We agree that a clearer comparison of the discriminative performance of the evaluated nutritional scores strengthens the interpretation of our findings.

We have revised the manuscript to explicitly state this comparison in the Results and Discussion sections.

 

Comments 3:

3.⁠ ⁠Please improve the description of the survival analysis in the Methods section.
The statistical methods section should explicitly describe the survival analysis procedure, including:
(i) the use of Kaplan–Meier curves,
(ii) the log-rank test for group comparisons, and
(iii) the definition of time-to-event (from ICU admission to death or censoring).
At present, this part of the methodology is not sufficiently clear and should be expanded to ensure reproducibility.

Response 3:

We thank the reviewer for this comment. We would like to clarify that the survival analysis methodology was already described in the Methods section, including the definition of time-to-event, the use of Kaplan–Meier curves, and the log-rank test for group comparisons.

However, to improve clarity and ensure better visibility of these elements, the corresponding paragraph has been slightly revised and expanded.

 

Comments 4:

4. Please refine the interpretation of the Kaplan–Meier findings in the Discussion.
The Discussion should more clearly explain why PNI reached statistical significance in the Kaplan–Meier analysis, whereas mNUTRIC showed the strongest signal in the Cox regression and ROC analyses. A brief explanation that categorizing continuous variables may reduce statistical power, whereas Cox regression preserves more prognostic information, would improve the interpretation and make the results more coherent.

Response 4:

We thank the reviewer for this insightful comment. We agree that a clearer explanation of the differences between Kaplan–Meier and Cox regression findings improves the interpretation of our results.

Kaplan–Meier analysis was performed using categorical variables, which may reduce statistical power due to the loss of information inherent in categorizing continuous variables. In contrast, Cox proportional hazards regression preserves the continuous nature of the predictors and may therefore provide a more sensitive assessment of their association with mortality risk.

We have revised the Discussion section to clarify this methodological aspect and to better explain the observed differences between PNI and mNUTRIC across statistical approaches.

 

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The revised manuscript has been substantially improved. The authors have addressed the major concerns raised previously

The paper still needs the final wording to be aligned between the Results, Discussion, and Conclusions sections. The adjusted Cox model shows mNUTRIC with an HR of 1.39 (95% CI: 0.95–2.04; p = 0.087), so the final narrative should not imply an independently significant association after adjustment. The safest formulation is that mNUTRIC had the strongest univariate prognostic signal and the strongest overall association in this selected cohort, but this signal weakened after limited adjustment.

 There is still at least one obvious citation placeholder in the introduction: "[ref]" in the albumin-based tools paragraph. This must be replaced with a real citation.

The manuscript still needs careful final copyediting before acceptance. Even ignoring the deleted text, the wording is awkward in places and reads as revised rather than polished final prose. For instance, parts of the introduction and aims section are verbose and have rough syntax.

The limitations and conclusion sections are much improved, but the final version should explicitly state that the findings are exploratory and apply only to a cohort with confirmed nutritional risk. They should not be generalized to all septic ICU patients. This message is present, but it should be consistent throughout the abstract, discussion, and conclusion.

Author Response

Author's Reply to the Review Report (Reviewer 1)_R2

Dear Reviewer,

We would like to express our sincere gratitude for the time and effort you devoted to reviewing our manuscript. Your thoughtful evaluation, encouraging remarks, and constructive suggestions are deeply appreciated. Your feedback has been invaluable in improving the quality and clarity of our work, and we are truly grateful for your insightful contributions to this manuscript.

All the typing recommended changes were performed in the body of our manuscript, with the Track Changes function activated.

 

Comments and Suggestions for Authors

The revised manuscript has been substantially improved. The authors have addressed the major concerns raised previously

The paper still needs the final wording to be aligned between the Results, Discussion, and Conclusions sections. The adjusted Cox model shows mNUTRIC with an HR of 1.39 (95% CI: 0.95–2.04; p = 0.087), so the final narrative should not imply an independently significant association after adjustment. The safest formulation is that mNUTRIC had the strongest univariate prognostic signal and the strongest overall association in this selected cohort, but this signal weakened after limited adjustment.

There is still at least one obvious citation placeholder in the introduction: "[ref]" in the albumin-based tools paragraph. This must be replaced with a real citation.

The manuscript still needs careful final copy editing before acceptance. Even ignoring the deleted text, the wording is awkward in places and reads as revised rather than polished final prose. For instance, parts of the introduction and aims section are verbose and have rough syntax.

The limitations and conclusion sections are much improved, but the final version should explicitly state that the findings are exploratory and apply only to a cohort with confirmed nutritional risk. They should not be generalized to all septic ICU patients. This message is present, but it should be consistent throughout the abstract, discussion, and conclusion.

 

Response:
We thank the reviewer for this important and insightful comment. We fully agree that the interpretation of the adjusted Cox regression results should be carefully aligned throughout the manuscript.

In response, we have revised the Abstract, Discussion, and Conclusions to ensure a consistent and accurate representation of the findings. Specifically, we now clearly state that the mNUTRIC score demonstrated the strongest univariate prognostic signal, while its association with ICU mortality was attenuated and did not reach statistical significance after limited multivariable adjustment.

We have also ensured that the narrative no longer implies an independent prognostic effect of mNUTRIC after adjustment. Instead, the results are now consistently presented as reflecting the strongest overall association within this selected cohort, while acknowledging the influence of disease-severity components embedded in the score.

Furthermore, we have harmonized the interpretation across all sections to emphasize that the findings are exploratory and apply specifically to a selected cohort of septic ICU patients with confirmed nutritional risk, and therefore should not be generalized to the broader population of critically ill septic patients.

These revisions improve the methodological accuracy and internal consistency of the manuscript.

Changes in Manuscript

• The Abstract Conclusions section was revised to clarify that the association of mNUTRIC with ICU mortality was attenuated and not statistically significant after limited multivariable adjustment.
• The Discussion section was updated to explicitly distinguish between univariate and adjusted analyses and to avoid overinterpretation of the multivariable results.
• The Conclusions section was reformulated to reflect that mNUTRIC showed the strongest univariate prognostic signal, while emphasizing the exploratory nature of the findings and their limited generalizability.

Author Response File: Author Response.pdf