Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript explores the relationship between the GH–IGF axis and GDF-15 in critically ill patients and attempts to characterize the intercorrelation structure among these biomarkers rather than proposing a clinically applicable prognostic model. The authors have clearly improved the manuscript by explicitly defining the primary objective, emphasizing the exploratory nature of the study, and acknowledging the limitations imposed by the small sample size.

The principal novelty lies in the simultaneous measurement of the complete GH–IGF axis together with GDF-15 in the same ICU cohort, allowing exploration of biomarker clustering rather than isolated biomarker performance. This represents an interesting mechanistic contribution.

However, several important issues still limit the manuscript.

Major comments

1. Sample size remains the major limitation

Although the authors repeatedly acknowledge the exploratory nature of the study, the manuscript still draws relatively strong biological conclusions from only 43 patients (9 non-survivors). Nearly every statistical analysis, including ROC analysis, clustering, partial correlations, longitudinal modelling and logistic regression, is based on this very limited dataset.

2. Heterogeneous ICU population

Approximately half of the cohort consists of patients with brain injury while the remainder includes sepsis and other ICU diagnoses.

Brain injury is known to profoundly alter:

• GH secretion
• pituitary function
• IGF axis
• inflammatory responses
• GDF-15 concentrations

Pooling these populations may generate biomarker relationships that are diagnosis-specific rather than universal.

3. Survival endpoint

The manuscript uses ICU mortality as the primary endpoint.

However,

• ICU discharge policies vary considerably,
• ICU mortality is influenced by organizational factors,
• hospital mortality or 28-day mortality may provide more robust endpoints.

The authors report 28-day mortality but do not justify why ICU mortality was selected as the primary outcome.

4. Adjustment for confounding

Severity adjustment includes APACHE II and SOFA only.

Several additional variables could influence GH–IGF physiology, including

• age
• BMI
• diabetes
• nutritional status
• liver dysfunction
• renal replacement therapy
• vasopressor dose
• mechanical ventilation duration

5. Longitudinal analyses

The manuscript correctly states that longitudinal analyses are exploratory.

Nevertheless, nearly two pages are devoted to analyses that ultimately conclude they are unreliable because of informative missingness and very small numbers.

These results contribute little to the main message.

6. Interpretation of clustering

Hierarchical clustering identifies statistical similarity rather than biological pathways.

Throughout the Discussion the authors occasionally move from statistical clustering toward mechanistic interpretations.

Minor comments

1. Excessive emphasis on novelty

Expressions such as

"to our knowledge"

"first simultaneous characterization"

appear repeatedly throughout the manuscript.

After the Introduction, these statements become repetitive and could be reduced.

2. Introduction

The Introduction is somewhat lengthy.

Several paragraphs describing cytokine-mediated GH resistance could be condensed without affecting readability.

3. Statistical methods

The statistical section is unusually long.

Many details regarding permutation testing, bootstrap confidence intervals and GEE assumptions could be transferred to Supplementary Methods.

4. Multiple testing

Although the authors repeatedly state that no correction for multiple comparisons was applied, nearly every table still highlights "nominal significance."

Because approximately 45 pairwise correlations were examined, readers should be reminded that false-positive findings remain possible.

5. ROC analyses

The ROC analyses are of limited value because only 9 events are available.

Although leave-one-out cross-validation is appreciated, ROC curves may inadvertently suggest clinical usefulness.

6. Figure 2

The correlation heatmap contains numerous coefficients that are difficult to read.

Increasing font size or simplifying the figure would improve readability.

7. Clinical implications

The manuscript appropriately concludes that no clinical application is currently supported.

Nevertheless, a short paragraph explaining how these findings might guide future mechanistic or translational studies would strengthen the final section.

Author Response

Reviewer 1

Major comments

[R1.1] Sample size remains the major limitation. Although the authors repeatedly acknowledge the exploratory nature of the study, the manuscript still draws relatively strong biological conclusions from only 43 patients (9 non-survivors). Nearly every statistical analysis, including ROC analysis, clustering, partial correlations, longitudinal modelling and logistic regression, is based on this very limited dataset.

RESPONSE: We agree, and have acted on this in three ways rather than only re-stating the caveat. First, we substantially expanded the robustness testing of the finding we treat as reasonably solid — the GDF-15–IGFBP-1/IGFBP-2 correlation — adding partial correlations adjusting for IL-6, age+BMI, and mechanical-ventilation duration to the existing APACHE II/SOFA adjustment (new Section 3.7, Supplementary Table S3); the correlation is essentially unchanged under every adjustment we could construct from the available data. Second, we added Benjamini–Hochberg FDR correction across all 45 pairwise correlations tested at T01 (Table 2 footnote, Supplementary Table S3): 9 of 11 nominally significant pairs survive q<0.05, including all three GDF-15–inhibitory-IGFBP/GHBP pairs. Third, we revised the language in the Discussion and Conclusions that read as more assertive than the data support (e.g., "demonstrates" → "is consistent with"; "reveals" → "suggests"), and added an explicit sentence to Section 2.3 stating that this is a pilot dataset not powered for confirmatory inference and that every effect size in the manuscript, including the principal correlation, should be read as an estimate with wide uncertainty. We did not add a multivariable logistic model: a multivariable model adjusting for even 3–4 covariates (e.g., age, lactate, catecholamine exposure) would have fewer than 2–3 events per parameter, far below the ≥10 events-per-variable threshold generally considered necessary for stable logistic regression coefficients; such a model would produce unstable, non-generalizable coefficients and a false impression of adjustment (please see also our response to R3.8). At n=9 events we believe this would create false precision rather than remove it (please see also our response to R2.1).

[R1.2] Heterogeneous ICU population. Approximately half of the cohort consists of patients with brain injury while the remainder includes sepsis and other ICU diagnoses. Brain injury is known to profoundly alter GH secretion, pituitary function, IGF axis, inflammatory responses, GDF-15 concentrations. Pooling these populations may generate biomarker relationships that are diagnosis-specific rather than universal.

RESPONSE: This is an important concern and we address it with data rather than only discussion. We computed the four key correlations separately within the brain-injury subgroup (n=22) and the non-brain-injury subgroup (n=21) (new Section 3.7, Supplementary Table S3). The GDF-15–IGFBP-1 correlation is strong and significant in both subgroups (brain injury: ρ=0.75, p<00.001; non-brain injury: ρ=0.63, p=0.002), arguing against a purely diagnosis-specific artifact for this pair. However, GDF-15–IGFBP-2 is essentially absent in the brain-injury subgroup (ρ=0.29, p=0.20) and strong in the non-brain-injury subgroup (ρ=0.63, p=0.002) — a genuine, honestly-reported heterogeneity signal that we did not have in the original submission. We have added this as a explicit new limitation/finding in the Discussion (paragraph on heterogeneity) rather than treating the pooled estimate as if it applied uniformly, and we now describe the cluster as "not uniformly present across diagnostic subgroups at this sample size" rather than as a single homogeneous phenomenon. We also clarify in the revised Introduction/Methods (please see our response to R3.1–R3.2) that the study's target construct is a generic, etiology-independent metabolic-stress signature, with diagnosis-heterogeneity treated as a variable to be tested, not a nuisance to be hidden.

[R1.3] Survival endpoint. The manuscript uses ICU mortality as the primary endpoint. However, ICU discharge policies vary considerably, ICU mortality is influenced by organizational factors; hospital mortality or 28-day mortality may provide more robust endpoints. The authors report 28-day mortality but do not justify why ICU mortality was selected as the primary outcome.

RESPONSE: We have added explicit justification to Section 2.1: ICU mortality was chosen a priori as the primary endpoint because the biological question under study (acute neuroendocrine/metabolic stress response) is most proximally linked to the acute critical-illness episode itself, whereas 28-day or hospital mortality also incorporate post-ICU events (ward-level complications, rehabilitation-phase events) that are further removed from the T01–T04 biomarker measurement window. We agree this is a design choice with trade-offs, and we now report a formal sensitivity analysis: GDF-15 AUC for 28-day mortality (5 events) is 0.65 versus 0.74 for ICU mortality (Mann-Whitney p=0.28 vs. 0.03), consistent in direction but, as expected with fewer/later events, weaker and non-significant. This is now reported in new Section 3.7 and referenced in the Limitations paragraph, with an explicit statement that the endpoint choice affects the discrimination estimates and should be considered when interpreting them.

[R1.4] Adjustment for confounding. Severity adjustment includes APACHE II and SOFA only. Several additional variables could influence GH–IGF physiology, including age, BMI, diabetes, nutritional status, liver dysfunction, renal replacement therapy, vasopressor dose, mechanical ventilation duration.

RESPONSE: We have added partial correlations adjusting for age+BMI and for mechanical-ventilation duration to Section 3.7/Supplementary Table S3; the GDF-15–IGFBP-1/IGFBP-2 correlations are essentially unchanged (partial ρ within 0.03–0.11 of unadjusted values for age+BMI; within 0.01–0.05 for MV duration). We also added a derived vasopressor-exposure variable (norepinephrine ± vasopressin at T01, from the ICU medication record): 29/43 patients (67%) received a vasopressor at T01, with no significant difference between survivors (65%) and non-survivors (78%; Fisher p=0.69); this is now reported in Table 1. We were not able to adjust for diabetes status, nutritional status, liver dysfunction, or renal-replacement-therapy exposure, because these were not systematically captured as structured variables in the original data-collection protocol for this biomarker study (as opposed to free-text clinical notes, which we did mine for comorbidity and vasopressor information — see response to R3.6.1/R3.6.2). We now state this explicitly as a specific, named limitation in the Discussion rather than only alluding to "additional confounders."

[R1.5] Longitudinal analyses. The manuscript correctly states that longitudinal analyses are exploratory. Nevertheless, nearly two pages are devoted to analyses that ultimately conclude they are unreliable because of informative missingness and very small numbers. These results contribute little to the main message.

RESPONSE: We agree, and we have shortened this substantially. Section 3.6.2 has been cut from two long paragraphs to a single, tighter paragraph reporting only the headline instability finding and the IGFBP-2 exception; the detailed GEE walkthrough has been moved to Supplementary Methods and Supplementary Table S2, referenced by a single sentence in the main text. We also include Supplementary Figure S1 (longitudinal trajectory plot for all ten biomarkers) so that a reader who wants to inspect the trajectories in full can do so in the Supplementary Materials. The corresponding Discussion paragraph has also been shortened by roughly one-third.

[R1.6] Interpretation of clustering. Hierarchical clustering identifies statistical similarity rather than biological pathways. Throughout the Discussion the authors occasionally move from statistical clustering toward mechanistic interpretations.

RESPONSE: We have reviewed the Discussion line by line for this and revised several sentences. Specifically, the sentence beginning "The observed clustering of GDF-15 with IGFBP-1, IGFBP-2, and GHBP...is consistent with a model in which GDF-15 acts..." has been re-worded to state explicitly that this is a hypothesis generated by, not established by, the clustering, and that clustering identifies statistical co-variation, not a validated pathway. We also softened "This pattern is biologically plausible" to "One biologically plausible interpretation, among others, is..." and added a new sentence clarifying that hierarchical clustering has no mechanistic content and that the biological narrative offered is a candidate interpretation requiring direct testing, not a finding of the clustering itself.

Minor comments

[R1.7] Excessive emphasis on novelty. Expressions such as "to our knowledge", "first simultaneous characterization" appear repeatedly throughout the manuscript. After the Introduction, these statements become repetitive and could be reduced.

RESPONSE: We counted 6 instances of "to our knowledge"/"first" framing outside the Introduction and Abstract and removed 4 of them (Results 3.3 opening sentence, Discussion opening paragraph, Discussion paragraph 6, Conclusions), retaining the claim once in the Abstract and once at first statement in the Introduction, where it is needed to establish rationale, and once in the Discussion where it directly supports the "why this matters" argument. The removed instances have been replaced with plain descriptive language.

[R1.8] Introduction. The Introduction is somewhat lengthy. Several paragraphs describing cytokine-mediated GH resistance could be condensed without affecting readability.

RESPONSE: The paragraph describing IL-1/TNF/IL-6-mediated GH resistance (Introduction, paragraph 1) has been shortened by approximately 40%, removing secondary mechanistic detail (the SOCS3/JAK-STAT sub-mechanism sentence and the "synergistic feedback loop" sentence) while retaining the core citation-supported claim. Total Introduction length is reduced by approximately 25%.

[R1.9] Statistical methods. The statistical section is unusually long. Many details regarding permutation testing, bootstrap confidence intervals and GEE assumptions could be transferred to Supplementary Methods.

RESPONSE: In the revised version we have proceeded accordingly. Section 2.3 has been shortened to describe the analytic strategy and name each method, while the full technical details (resampling counts, rank-residual partial-correlation construction, GEE cluster-count rationale, LOO-CV procedure) have been moved to the Supplementary Materials. Section 2.3 is now approximately half its original length.

[R1.10] Multiple testing. Although the authors repeatedly state that no correction for multiple comparisons was applied, nearly every table still highlights "nominal significance." Because approximately 45 pairwise correlations were examined, readers should be reminded that false-positive findings remain possible.

RESPONSE: We have added Benjamini–Hochberg FDR q-values for all 45 T01 pairwise correlations (new Supplementary Table S3, referenced from Table 2 and Section 3.3). Of 11 nominally significant pairs, 9 survive q<0.05, including all pairs central to the principal finding (GDF-15–IGFBP-1, GDF-15–IGFBP-2, GDF-15–GHBP, IGFBP-1–IGFBP-2); 2 borderline pairs (IGF-1–IGFBP-1, GHBP–IGFBP-1) do not survive correction and are now explicitly flagged as such in text. We have also added one sentence to Section 2.3 stating the FDR results and directing readers to Supplementary Table S3 for the full q-value table, addressing the concern directly with numbers rather than only a caveat sentence.

[R1.11] ROC analyses. The ROC analyses are of limited value because only 9 events are available. Although leave-one-out cross-validation is appreciated, ROC curves may inadvertently suggest clinical usefulness.

RESPONSE: We have strengthened the caveat immediately preceding Section 3.6.1 and in the Figure 5 (formerly Figure 4) legend, adding an explicit sentence: "With 9 events, these ROC/AUC estimates carry wide confidence intervals not shown on the curve itself, and their inclusion here is to contextualize the correlation findings against established severity scores, not to propose the biomarkers as diagnostic or prognostic tests." We considered removing the ROC curves entirely but retained them because Reviewer 2 [R2.6] specifically asked us to better discuss "the limited incremental clinical value of these biomarkers over established severity scoring systems," which requires showing the comparison; we believe the strengthened caveat resolves both concerns simultaneously.

[R1.12] Figure 2. The correlation heatmap contains numerous coefficients that are difficult to read. Increasing font size or simplifying the figure would improve readability.

RESPONSE: Figure 3 (formerly Figure 2) has been completely redrawn with larger cell-label and axis-label fonts, higher-contrast text coloring on dark cells, and a larger overall canvas. No data or content was changed, only presentation.

[R1.13] Clinical implications. The manuscript appropriately concludes that no clinical application is currently supported. Nevertheless, a short paragraph explaining how these findings might guide future mechanistic or translational studies would strengthen the final section.

RESPONSE: We added a new paragraph at the end of the Discussion (before Conclusions) outlining three concrete next steps the findings motivate: (1) direct measurement of hepatic JAK2/STAT5 signaling activity relative to GDF-15 and IGFBP levels in mechanistic/animal models to test the proposed downstream-of-GHR hypothesis; (2) a multicenter, adequately powered replication cohort with pre-specified diagnosis-stratified analysis, given the brain-injury/non-brain-injury heterogeneity we now report (R1.2); (3) inclusion of IL-6 and procalcitonin as co-measured covariates in any future GDF-15/IGFBP biomarker panel, given the partial confounding of the GDF-15–GHBP link by IL-6 that we now report (R3.9).

Reviewer 2 Report

Comments and Suggestions for Authors

 

Major Comments

1. The study includes only 43 patients, with 9 non-survivors, which substantially limits the robustness of the conclusions. Many analyses, including ROC curves, correlation analyses, and longitudinal comparisons, are underpowered and susceptible to overfitting. The authors should emphasize this limitation more clearly and avoid drawing strong biological conclusions.
2. All findings were generated from a single-center cohort without validation in an independent dataset. Validation is necessary before proposing the identified biomarker cluster as biologically meaningful or clinically relevant.
3. Although the authors acknowledge that analyses are exploratory, numerous biomarkers and pairwise correlations were tested without correction for multiple testing. Appropriate false discovery rate (FDR) adjustment or additional justification should be provided.
4. The discussion proposes mechanistic links between GDF-15 and GH–IGF signaling; however, the study only reports observational correlations. No functional experiments are presented to support these mechanistic interpretations.
5. Biomarker levels may be influenced by age, sex, BMI, nutritional status, renal function, liver function, medications, infection severity, and underlying diagnoses. Adjustment only for APACHE II and SOFA may not adequately account for these confounders.
6. Although GDF-15 demonstrated the highest biomarker AUC, it remained inferior to APACHE II and SOFA scores. The manuscript should better discuss the limited incremental clinical value of these biomarkers over established severity scoring systems.
7. The cohort includes both neurocritical care and general ICU patients with diverse underlying diseases. This heterogeneity may substantially influence endocrine biomarkers. Subgroup analyses or discussion of disease-specific effects would strengthen the manuscript.

Minor Comments

1. The Introduction is overly long and could be condensed by reducing background information on GH signaling.
2. Several sections of the Results repeat information already presented in tables and figures. Consider shortening the text.
3. Figures should report exact sample sizes for each biomarker where missing values occurred.
4. The manuscript should consistently define all abbreviations at first use (GHBP, ALS, GHR, IGFBPs, etc.).
5. Figure legends should clearly indicate that all statistical analyses are exploratory and unadjusted.
6. A schematic figure summarizing the proposed relationship between GDF-15, IGFBPs, GH resistance, and critical illness would improve readability.
7. The Discussion should compare these findings with larger published ICU biomarker studies and better explain similarities and differences.
8. The limitations section should explicitly discuss the lack of external validation, single-center design, small event number, and possible selection bias.

 

Author Response

Reviewer 2

Major Comments

[R2.1] The study includes only 43 patients, with 9 non-survivors, which substantially limits the robustness of the conclusions. Many analyses, including ROC curves, correlation analyses, and longitudinal comparisons, are underpowered and susceptible to overfitting. The authors should emphasize this limitation more clearly and avoid drawing strong biological conclusions.

RESPONSE: In the revised version of the manuscript we substantially expanded the robustness testing of the one finding we do treat as reasonably solid — the GDF-15–IGFBP-1/IGFBP-2 correlation — adding partial correlations adjusting for IL-6, age+BMI, and mechanical-ventilation duration to the existing APACHE II/SOFA adjustment (new Section 3.7, Supplementary Table S3); the correlation is essentially unchanged under every adjustment we could construct from the available data. Second, we added Benjamini–Hochberg FDR correction across all 45 pairwise correlations tested at T01 (Table 2 footnote, Supplementary Table S3): 9 of 11 nominally significant pairs survive q<0.05, including all three GDF-15–inhibitory-IGFBP/GHBP pairs. Third, we revised the language in the Discussion and Conclusions that read as more assertive than the data support. Moreover, we have added one further explicit sentence at the start of the Discussion: "We emphasize at the outset that every quantitative estimate in this manuscript, including the principal correlation finding, is a pilot estimate from 43 patients and should be treated as hypothesis-generating; none of the analyses in this study, individually or collectively, are powered for confirmatory inference or clinical application." Please see also our response to R1.1.

[R2.2] All findings were generated from a single-center cohort without validation in an independent dataset. Validation is necessary before proposing the identified biomarker cluster as biologically meaningful or clinically relevant.

RESPONSE: We agree and have never intended to propose clinical relevance; we have strengthened the text to make this unambiguous. The Discussion limitations paragraph now opens its single-center-design sentence with an explicit statement that no claim in this manuscript should be acted upon clinically prior to independent multicenter replication, and the Conclusions section now states this requirement as the first sentence of the paragraph. We have also added the phrase "external validation" explicitly to the limitations list (previously present only implicitly via "single-center design").

[R2.3] Although the authors acknowledge that analyses are exploratory, numerous biomarkers and pairwise correlations were tested without correction for multiple testing. Appropriate false discovery rate (FDR) adjustment or additional justification should be provided.

RESPONSE: We revised accordingly: we have added Benjamini–Hochberg FDR q-values for all 45 T01 pairwise correlations (new Supplementary Table S3, referenced from Table 2 and Section 3.3). Of 11 nominally significant pairs, 9 survive q<0.05, including all pairs central to the principal finding (GDF-15–IGFBP-1, GDF-15–IGFBP-2, GDF-15–GHBP, IGFBP-1–IGFBP-2); 2 borderline pairs (IGF-1–IGFBP-1, GHBP–IGFBP-1) do not survive correction and are now explicitly flagged as such in text. We have also added one sentence to Section 2.3 stating the FDR results and directing readers to Supplementary Table S3 for the full q-value table, addressing the concern directly with numbers rather than only a caveat sentence. See also our response to R1.10 for full detail.

[R2.4] The discussion proposes mechanistic links between GDF-15 and GH–IGF signaling; however, the study only reports observational correlations. No functional experiments are presented to support these mechanistic interpretations.

RESPONSE: We have reviewed the Discussion and revised accordingly. We revised the relevant Discussion paragraph to explicitly frame the JAK2/STAT5 mechanistic narrative as a hypothesis motivated by prior cell/animal mechanistic literature (refs 12–13).  The sentence "The observed clustering of GDF-15 with IGFBP-1, IGFBP-2, and GHBP...is consistent with a model in which GDF-15 acts..." has been re-worded to state explicitly that this is a hypothesis generated by, not established by, the clustering, and that clustering identifies statistical co-variation, not a validated pathway. We also changed "This pattern is biologically plausible" to "One biologically plausible interpretation, among others, ...".  The present correlational pattern, not as a finding established by this study, and added a sentence noting that no functional or mechanistic experiments were performed here and that this is a purely observational, correlational dataset. Please see also our response to R1.6.

[R2.5] Biomarker levels may be influenced by age, sex, BMI, nutritional status, renal function, liver function, medications, infection severity, and underlying diagnoses. Adjustment only for APACHE II and SOFA may not adequately account for these confounders.

RESPONSE: See response to R1.4. We added partial correlation adjustment for age+BMI, mechanical-ventilation duration, and IL-6 (as a proxy for infection/inflammatory severity), plus a diagnosis-stratified (brain-injury vs. non-brain-injury) sensitivity analysis. Renal function, liver function, nutritional status, and detailed medication exposure (beyond the vasopressor variable we added) were not systematically captured as structured variables during the original data-collection protocol; we now state this explicitly in Limitations rather than implying comprehensive adjustment. Please see also our response to R1.4.

[R2.6] Although GDF-15 demonstrated the highest biomarker AUC, it remained inferior to APACHE II and SOFA scores. The manuscript should better discuss the limited incremental clinical value of these biomarkers over established severity scoring systems.

RESPONSE: We added a new sentence to the Discussion, immediately following the AUC comparison paragraph stating that the consistent finding that GDF-15 and IGFBP-2, individually or combined, do not approach the discrimination achieved by APACHE II or SOFA argues against any near-term role for these biomarkers as mortality-prediction tools in isolation; their potential value, if any, is more likely to lie in mechanistic stratification (e.g., identifying patients with a specific GDF-15-driven inhibitory-IGFBP signature) than in outcome prediction, which is already well served by validated severity scores. This directly names the incremental-value question rather than leaving it implicit.

[R2.7] The cohort includes both neurocritical care and general ICU patients with diverse underlying diseases. This heterogeneity may substantially influence endocrine biomarkers. Subgroup analyses or discussion of disease-specific effects would strengthen the manuscript.

RESPONSE: We added a brain-injury-stratified sensitivity analysis (new Section 3.7) showing that the GDF-15–IGFBP-1 relationship is consistent across strata but the GDF-15–IGFBP-2 relationship is not, and we discuss this heterogeneity explicitly and honestly in the Discussion rather than only acknowledging pooling as a generic limitation. Please see also our response to R1.2.

Minor Comments

[R2.8] The Introduction is overly long and could be condensed by reducing background information on GH signaling.

RESPONSE: The cytokine-mediated GH-resistance paragraph has been shortened by ~40%. See also our response to R1.8.

[R2.9] Several sections of the Results repeat information already presented in tables and figures. Consider shortening the text.

RESPONSE: We trimmed restated numeric detail from Results Sections 3.1, 3.3, and 3.5, where the same medians/IQRs or ρ/p values were both stated in-text and shown in the adjacent table; the in-text content now generally states the qualitative finding and directs the reader to the table for exact values, retaining full numeric detail only for the single principal finding (GDF-15 vs. IGFBP-1/IGFBP-2).

[R2.10] Figures should report exact sample sizes for each biomarker where missing values occurred.

RESPONSE: Sample sizes (n) were already reported in Table 2 for every biomarker; we have now also added them directly to the affected figure legends (Figure 3/heatmap legend now states "n = 40–43, varying by pair — see Table 2 for exact pairwise n"; Figure 5/ROC-AUC legend now restates n=43, 9 non-survivors explicitly at the start rather than only in the caption's final sentence).

[R2.11] The manuscript should consistently define all abbreviations at first use (GHBP, ALS, GHR, IGFBPs, etc.).

RESPONSE: We corrected three instances where an abbreviation was used in the Abstract or early Introduction before being spelled out (GHBP, ALS, IGFBPs); each is now spelled out in full at first use in the Abstract and again at first use in the Introduction body text, consistent with journal style.

[R2.12] Figure legends should clearly indicate that all statistical analyses are exploratory and unadjusted.

RESPONSE: We reviewed all figure legends; Figures 2, 4, and 5 (formerly 1, 3, 4) already carried this caveat. We added an equivalent explicit sentence to the Figure 3 (heatmap) legend, which previously stated it only for the p-value thresholds without a general exploratory-analysis caveat, and to the new Figure 6 schematic legend.

[R2.13] A schematic figure summarizing the proposed relationship between GDF-15, IGFBPs, GH resistance, and critical illness would improve readability.

RESPONSE: Added as new Figure 6, referenced from the Discussion. The schematic shows the two statistical clusters (classical GH-resistance signaling vs. stress/inhibitory-IGFBP cluster), the lack of cross-cluster correlation and the partial (GDF-15–GHBP only) confounding by IL-6, with a caption stating explicitly that arrows denote statistical association from this dataset, not established causal direction.

[R2.14] The Discussion should compare these findings with larger published ICU biomarker studies and better explain similarities and differences.

RESPONSE: We searched the recent literature and added a new paragraph to the Discussion comparing our findings with larger cohorts, including the multicenter trajectory-subtype study by Wang et al. (PeerJ 2025; ref 16, already cited) and additional GDF-15/IGFBP-2 ICU cohort literature (refs 14, 19, 20, already cited), explicitly noting where our AUC estimates (0.74 for GDF-15) sit relative to those larger-cohort point estimates and why our estimate carries much wider uncertainty at n=43.

[R2.15] The limitations section should explicitly discuss the lack of external validation, single-center design, small event number, and possible selection bias.

RESPONSE: The Limitations paragraph (Discussion) has been restructured into explicit, separately-stated points covering: (1) small sample size/event number, (2) single-center design and lack of external validation, (3) assay-specific measurement platform, (4) survivorship-driven attrition and informative missingness, (5) selection bias from the ICU-stay-≥3-days and prior-corticosteroid exclusion criteria (newly added, responding also to R3.3), and (6) the observational, non-causal design.

Reviewer 3 Report

Comments and Suggestions for Authors

◆ General Overview

This pilot study investigates the intercorrelation structure of the GH-IGF axis and GDF-15 in a single ICU cohort ($N = 43$) using hierarchical clustering and partial correlation. While the mathematical approach to unearthing the co-varying structures is intriguing in a purely statistical vacuum , the study exhibits several critical clinical ambiguities, insufficient granular baseline descriptions, and a vulnerability to selection bias that must be rigorously addressed before considering publication.

Since this is a prospective pilot study with a small sample size, the strength of the paper must rely on the meticulousness of its clinical description, the transparency of its patient selection, and a clear, well-defined biological hypothesis. Currently, the paper falls short of these standards. The following points are raised to help the authors improve the clinical validity and transparency of their manuscript.

◆ Major Comments

1. Clarification of the Study's Aim and Target Pathophysiology

• The overarching clinical aim of this study remains ambiguous. The authors must explicitly define whether these biomarkers are being evaluated to reflect true biological sepsis, general high-surgical/trauma invasion, or systemic inflammation.

• The title uses the broad term "Critical Illness" , and the Introduction leans heavily into a sepsis narrative. However, 51% of the cohort consists of acute brain injury patients. The biological drivers of GDF-15 in acute brain injury are primarily local tissue destruction and massive central sympathetic/catecholamine surge, which fundamentally differ from cytokine-driven systemic sepsis. The authors must clarify why this specific mixed cohort was selected and clearly state the specific biological phenomenon they are capturing.

2. Demand for Complete Transparency on Patient Selection and Attrition

• Given the small sample size ($N=43$), selection bias cannot be ruled out. For instance, highly metabolic ICU populations—such as those in a Burning Unit (severe burns), major abdominal surgical departments, or cardiovascular surgical units—are not represented. Why were they omitted?

• The authors must integrate a detailed, STROBE-compliant Patient Selection Flowchart into the main body of the text, rather than delegating it entirely to the Supplementary materials (Supplementary Figure S1).

• The main text must explicitly describe the exact inclusion and exclusion criteria in full detail, clearly demonstrating how a heterogeneous "mixed" ICU population was screened down to these exact 43 individuals.

3. Enrichment of Baseline Characteristics and Granular Clinical Data

• Since this is a descriptive pilot study, the baseline clinical description in Table 1 is insufficient. The authors must retrieve and provide the following granular data from their prospective records to improve Table 1 or create an expanded supplementary table:

  • Comorbidities: Detailed background diseases (e.g., chronic heart failure, chronic kidney disease, diabetes, or malignancy) that heavily confound baseline GDF-15 and GH-IGF axes.

  • Hemodynamic Support: The exact proportion of patients requiring vasoactive/inotropic support (catecholamines) and their maximum dose-intensity.

  • Infectious Profile: The exact rate of confirmed bacteremia, source of infection, and the presence/absence of Infective Endocarditis (IE), which severely dictates cardiovascular stress.

  •  

    Respiratory Settings: Granular mechanical ventilation settings (e.g., P/F ratio, PEEP levels) at admission (T01).

4. Absolute Transparency on Outcome and Causes of Death

• Since multi-variable logistic regression to adjust for clinical outcomes is statistically impossible due to the limited number of events ($N=9$ deaths), the authors must describe their clinical outcome with utmost precision.

• The specific, direct cause of death for each of the 9 non-survivors must be disclosed (e.g., brain herniation, septic shock, multi-organ failure). A death driven by physical brain destruction poses a completely different neuroendocrine trajectory compared to systemic septic shock, and mixing them without an explicit breakdown undermines the clinical interpretation of the biomarker trajectories.

5. Justification for Restricting the Scope to Correlational Analyses

• In the secondary analysis, the authors noted that a combined GDF-15 + IGFBP-2 logistic model did not improve discrimination. However, because clustering already showed they are highly collinear, this result is a statistical tautology.

• The authors must explicitly state in the Methods/Discussion why they restricted their primary analysis to correlation and clustering rather than building a multi-variable clinical prediction model. They should clarify that due to the sample size constraint ($N=43$), a robust multi-variable model adjusting for age, lactate, and catecholamine requirements was unfeasible, and therefore, the outcome analysis is strictly exploratory and descriptive.

6. Inquiry Regarding the Lack of Non-GF Primary Biological Markers

•  

  Reviewer's Note: I recognize that this study is a completed prospective observation and further laboratory experiments cannot be performed. However, from a critical care perspective, evaluating metabolic stress and "mitochondrial dysfunction" without comparing the findings to primary, standard biological markers—such as Interleukin-6 (IL-6), Procalcitionin (PCT), and Serum Lactate—is a major conceptual limitation.

• The authors are requested to provide their scientific opinion on how the lack of these primary biological markers might affect the specificity of the GDF-15/IGFBP cluster. Could this cluster simply be a secondary reflection (an echo) of a massive IL-6 or lactate elevation? This limitation and the authors' perspective must be transparently incorporated into the Discussion.

7. Characterization of the Healthy Control Baseline

• The authors utilized a "GHR vs. Healthy Controls (HC)" metric via RT-PCR, which implies they have access to healthy control data.

• To ground the "profound dysregulation" mentioned in the text, the authors should describe how the absolute values of the primary GH-IGF axis components and GDF-15 behave in these healthy controls. Providing a brief description or reference point of what constitutes a normal baseline for these specific assays will greatly enhance the reader's understanding of the magnitude of stress-induced changes seen in the ICU cohort.

◆ Minor Comments

•  

  Surrogate Nature of PBMC: The measurement of GHR expression in PBMCs is highly surrogate, as the liver is the primary site of GHR-driven IGF-1 production during critical illness. The physiological limitations of using circulating mononuclear cells as a proxy for hepatic GH-resistance should be discussed more critically.

Declaration of AI Assistance:

As a native Japanese speaker, I have utilized AI assistance (Gemini 2.5) to refine the English phrasing, language syntax, and formatting of this peer review report to ensure academic clarity and rigorous terminology. The clinical evaluations, logical reasoning, and final assessments are entirely my own professional responsibilities.

Author Response

Reviewer 3

Major Comments

[R3.1] Clarification of the Study's Aim and Target Pathophysiology. The overarching clinical aim of this study remains ambiguous. The authors must explicitly define whether these biomarkers are being evaluated to reflect true biological sepsis, general high-surgical/trauma invasion, or systemic inflammation.

RESPONSE: We have added an explicit statement to the end of the Introduction (before the aims paragraph) and to Methods 2.1: "The present study targets the GH–IGF axis / GDF-15 response to critical illness as a generic, etiology-independent metabolic-stress construct — not a sepsis-specific or trauma-specific signature — reflecting the case-mix of a general ICU. Diagnosis-specific effects are tested directly, not assumed away, via the brain-injury-stratified sensitivity analysis in Section 3.7." This directly answers what construct is being measured and signals to the reader, before the Methods, that heterogeneity is treated analytically rather than glossed over.

[R3.2] The title uses the broad term "Critical Illness", and the Introduction leans heavily into a sepsis narrative. However, 51% of the cohort consists of acute brain injury patients. The biological drivers of GDF-15 in acute brain injury are primarily local tissue destruction and massive central sympathetic/catecholamine surge, which fundamentally differ from cytokine-driven systemic sepsis. The authors must clarify why this specific mixed cohort was selected and clearly state the specific biological phenomenon they are capturing.

RESPONSE: We agree the original Introduction over-weighted the sepsis/cytokine narrative relative to the actual cohort composition. We have (1) shortened the cytokine-mediated GH-resistance paragraph (see also R1.8/R2.8) and (2) added two new sentences after it, explicitly noting that brain injury constitutes a biologically distinct route to GDF-15 elevation (catecholamine surge and local tissue injury rather than cytokine-driven signaling), citing that this heterogeneity is precisely why we test — rather than assume — a shared GDF-15/IGFBP signature across the two dominant diagnostic groups in Section 3.7. We considered revising the title but concluded "Critical Illness" remains accurate as a description of the study population (a general ICU care cohort is, by definition, a critical-illness cohort, not a sepsis cohort); we have not renamed the paper to avoid overstating a sepsis focus it does not have, and the revised Introduction now makes this explicit rather than implying a sepsis-centric rationale it cannot support. We flag this title question for the Editor's final judgment.

[R3.3] Demand for Complete Transparency on Patient Selection and Attrition. Given the small sample size (N=43), selection bias cannot be ruled out. For instance, highly metabolic ICU populations — such as those in a Burning Unit (severe burns), major abdominal surgical departments, or cardiovascular surgical units — are not represented. Why were they omitted?

RESPONSE: We have added an explanatory sentence to Methods 2.1: Patients with major burns are rarely admitted to our ICU. Individuals undergoing abdominal surgery were not included either because their ICU stay was very short (< 3 days) or, in more severe cases such as peritonitis with septic shock, because corticosteroid therapy constituted an exclusion criterion. Cardiac surgery patients were likewise absent, as they are routinely managed in a dedicated cardiac ICU separate from our unit. Consequently, their non‑representation in this cohort reflects the referral patterns and predefined exclusion criteria of the study rather than any deliberate omission. We have also explicitly added, as a new sentence in Limitations, that the exclusion of ICU stays <3 days (needed to ensure ≥1 follow-up biomarker sample) and of patients started on corticosteroids before sampling could be completed may itself introduce selection bias by under-representing the most rapidly fatal presentations.

[R3.4] (STROBE flowchart) The authors must integrate a detailed, STROBE-compliant Patient Selection Flowchart into the main body of the text, rather than delegating it entirely to the Supplementary materials (Supplementary Figure S1).

RESPONSE: In the revised version of the manuscript the STROBE flow diagram is now main-text Figure 1 (all other figures renumbered +1 accordingly, Figures 2–6). The corresponding in-text citation in Section 3 ("Participant flow...are shown in Supplementary Figure S1") has been changed to "...are shown in Figure 1."

[R3.5] (Inclusion/exclusion detail) The main text must explicitly describe the exact inclusion and exclusion criteria in full detail, clearly demonstrating how a heterogeneous "mixed" ICU population was screened down to these exact 43 individuals.

RESPONSE: In the revised version of the manuscript, Methods 2.1 has been expanded into a clearly structured inclusion/exclusion paragraph: Inclusion criteria were: consecutive adult patients (aged ≥18 years) admitted to the ICU. Exclusion criteria were: age under 18 years; an ICU stay shorter than 3 days (needed to ensure at least one follow-up biomarker sample beyond T01); brain death at admission; pregnancy; contagious disease (HIV, hepatitis); ICU readmission or transfer from another ICU; and prior or in-ICU corticosteroid administration before the first blood draw (chronic pre-admission therapy ≥1 mg/kg prednisone-equivalent for over a month, or any in-hospital corticosteroid administration prior to sampling). This now appears as an explicit list rather than a single dense sentence, immediately followed by Figure 1 (the flow diagram) for visual confirmation of how 43 patients resulted.

[R3.6] Enrichment of Baseline Characteristics and Granular Clinical Data. Since this is a descriptive pilot study, the baseline clinical description in Table 1 is insufficient. The authors must retrieve and provide the following granular data...

[R3.6.1] Comorbidities: Detailed background diseases (e.g., chronic heart failure, chronic kidney disease, diabetes, or malignancy) that heavily confound baseline GDF-15 and GH-IGF axes.

RESPONSE: We reviewed the structured past-medical-history field recorded prospectively for each patient and compiled a comorbidity summary, now presented in new Supplementary Table S3. Of 43 patients, 26 (60%) had at least one documented pre-admission condition; the most frequent categories were alcohol/substance use disorder (5, 12%), hypothyroidism/autoimmune thyroid disease (5, 12%), psychiatric illness (4, 9%), prior cerebrovascular disease (3, 7%), cardiac disease (atrial fibrillation, mechanical valve; 2, 5%), and malignancy (2, 5%), with single instances of polycystic kidney disease, ITP, and prior bariatric surgery. We explicitly did not attempt to construct a formal weighted comorbidity index (e.g., Charlson), because the underlying data are free-text clinical notes rather than coded diagnoses, and we say so in the Supplementary Table S3 footnote.

[R3.6.2] Hemodynamic Support: The exact proportion of patients requiring vasoactive/inotropic support (catecholamines) and their maximum dose-intensity.

RESPONSE: We derived a vasopressor-exposure variable (norepinephrine and/or vasopressin) from the T01 ICU medication record: 29/43 patients (67%) received a vasopressor at T01 (survivors 65%, non-survivors 78%; Fisher p=0.69). This is now reported in Table 1. Maximum dose-intensity was not captured as a structured, extractable numeric field in the original data collection, so we report presence/absence only and state this limitation explicitly in the Supplementary Table S3 footnote rather than approximating a dose value.

[R3.6.3] Infectious Profile: The exact rate of confirmed bacteremia, source of infection, and the presence/absence of Infective Endocarditis (IE), which severely dictates cardiovascular stress.

RESPONSE: Sepsis was present in 10/43 patients (23%, already reported); a documented infection source was available for 4 of these 10 (respiratory, intra-abdominal, cholecystitis, and one confirmed Staphylococcus aureus bacteremia with respiratory co-source), now listed in Supplementary Table S3. Source was not systematically documented for the remaining 6/10 sepsis patients, and infective endocarditis was not part of a systematic screening protocol for this biomarker study; we therefore cannot report a cohort-wide IE rate and say so explicitly rather than reporting an unverified 0%.

[R3.6.4] Respiratory Settings: Granular mechanical ventilation settings (e.g., P/F ratio, PEEP levels) at admission (T01).

RESPONSE: Mechanical ventilation status at T01 (70% of the cohort, already reported) and total ventilation days (Table 1) were captured; P/F ratio and PEEP were not recorded as structured variables in the original data-collection protocol for this endocrine biomarker study. We state this explicitly as a limitation (Discussion, Limitations paragraph) rather than omitting the request silently.

[R3.7] (Cause of death) Absolute Transparency on Outcome and Causes of Death. ...The specific, direct cause of death for each of the 9 non-survivors must be disclosed... A death driven by physical brain destruction poses a completely different neuroendocrine trajectory compared to systemic septic shock, and mixing them without an explicit breakdown undermines the clinical interpretation.

RESPONSE: We agree with the underlying concern and provide the most precise, answer the data support. Formally adjudicated proximate cause of death (e.g., brain herniation vs. multi-organ failure, as would be assigned by a mortality review committee) was not part of this study's data collection and we will not construct it retrospectively from free-text notes, as that would risk misclassifying real clinical events. What we report (new Section 3.7 and Supplementary Table S3) is the primary admission diagnosis category for each non-survivor, which is a reasonable, data-supported proxy: 5/9 (56%) had a primary neurological/traumatic admission diagnosis (subarachnoid hemorrhage, intraventricular hemorrhage, or polytrauma with head injury), and 4/9 (44%) had non-neurological admission diagnoses, all four of which had documented sepsis at some point in the ICU stay (heat stroke with cardiac arrest, fever of unknown origin with evolving sepsis, and two further febrile/septic presentations). This maps reasonably well onto the two distinct trajectories the reviewer describes (primary neurological injury vs. systemic septic/inflammatory illness), and we now discuss both explicitly in the Discussion, including the acknowledgment that these are diagnosis-category proxies, not adjudicated causes of death.

[R3.8] (Justification for correlational scope) Justification for Restricting the Scope to Correlational Analyses. ...The authors must explicitly state in the Methods/Discussion why they restricted their primary analysis to correlation and clustering rather than building a multi-variable clinical prediction model. They should clarify that due to the sample size constraint (N=43), a robust multi-variable model... was unfeasible...

RESPONSE: We agree with the Reviewer; this is now stated explicitly rather than left implicit. We added a paragraph to Section 2.3 (and a shorter cross-reference in the Discussion) stating: with 9 events, a multivariable model adjusting for even 3–4 covariates (e.g., age, lactate, catecholamine exposure) would have fewer than 2–3 events per parameter, far below the ≥10 events-per-variable threshold generally considered necessary for stable logistic regression coefficients; such a model would produce unstable, non-generalizable coefficients and a false impression of adjustment. Correlation and hierarchical clustering, by contrast, use the full continuous biomarker distributions across all 43 patients (not conditioned on the rare outcome) and require no comparable events-per-parameter budget, which is why we designated this as the primary analysis and confined outcome-conditioned modelling (ROC/AUC, LOO-CV) to an explicitly secondary, exploratory role. We agree with the reviewer's related observation that the combined GDF-15+IGFBP-2 logistic model not outperforming GDF-15 alone is close to a statistical tautology given their collinearity (Table 3); we have added a sentence stating this directly in Section 3.6.1.

[R3.9] (IL-6/PCT/lactate) Inquiry Regarding the Lack of Non-GH Primary Biological Markers. ...evaluating metabolic stress and "mitochondrial dysfunction" without comparing the findings to primary, standard biological markers — such as Interleukin-6 (IL-6), Procalcitonin (PCT), and Serum Lactate — is a major conceptual limitation. ...Could this cluster simply be a secondary reflection (an echo) of a massive IL-6 or lactate elevation?

RESPONSE: This is an excellent question and, on investigation, a genuinely important one — we thank the reviewer for bringing attention to it. IL-6, procalcitonin, and lactate were in fact measured/recorded at T01 in this cohort (previously not analyzed against the GH–IGF/GDF-15 panel in this manuscript) and we have now run exactly the analysis the reviewer proposes (new Section 3.7, Supplementary Table S3). The answer is nuanced and we report it transparently: GDF-15 correlates strongly with IL-6 (ρ=0.76, p<0.001) and moderately with procalcitonin (ρ=0.45, p=0.01), but not significantly with lactate (ρ=0.24, p=0.13). When we adjust the key GDF-15–IGFBP partial correlations for IL-6, the GDF-15–IGFBP-1 (partial ρ=0.53, p<0.001) and GDF-15–IGFBP-2 (partial ρ=0.56, p<0.001) correlations remain essentially intact, arguing against these specific links being simply an "echo" of generic IL-6-driven inflammation. However, the GDF-15–GHBP correlation attenuates substantially and loses nominal significance after IL-6 adjustment (ρ 0.47→0.24, p=0.13), indicating that this particular link may indeed be substantially explained by shared inflammatory burden rather than a GDF-15-specific mechanism. We now report this three-way distinction explicitly in the Results (new Section 3.7), the Discussion (a new paragraph directly addressing the reviewer's "echo" framing), and the new schematic Figure 6, and we believe this is one of the most important analytical additions in this revision — it both answers the reviewer's question with real data and meaningfully refines our own claim (the IGFBP-1/IGFBP-2 links appear more specific than the GHBP link).

[R3.10] (Healthy control baseline) Characterization of the Healthy Control Baseline. The authors utilized a "GHR vs. Healthy Controls (HC)" metric via RT-PCR, which implies they have access to healthy control data. ...the authors should describe how the absolute values of the primary GH-IGF axis components and GDF-15 behave in these healthy controls.

RESPONSE: We have added a description of the healthy-control reference group to Methods 2.2: the HC pool used for RT-PCR relative quantification of GHR expression (2^-ΔΔCt method, normalized to the HC baseline) comprises age- and sex-matched healthy blood donors, consistent with the reference population described for PBMC-based receptor expression assays in this same patient cohort in our companion methodological paper (Poupouzas et al., Critical Care 2025;29:390, ref 21), which characterized n=25 age- and sex-matched healthy controls. We have not been able to report absolute GH-IGF-axis/GDF-15 concentrations in the HC group itself, because only the derived GHR ratio, not the underlying HC raw values for the other nine biomarkers, was part of the dataset available for this manuscript; the other nine biomarkers do not have a HC comparator in this study, only the T01–T04 patient measurements, and we have added a sentence clarifying this scope limitation explicitly rather than leaving it ambiguous.

Minor Comments

[R3.11] Surrogate Nature of PBMC: The measurement of GHR expression in PBMCs is highly surrogate, as the liver is the primary site of GHR-driven IGF-1 production during critical illness. The physiological limitations of using circulating mononuclear cells as a proxy for hepatic GH-resistance should be discussed more critically.

RESPONSE: We added a new sentence to the Limitations paragraph: "GHR expression was measured in PBMCs rather than hepatocytes, the primary site of GH-driven IGF-1 production in critical illness; PBMC GHR is a circulating-immune-cell surrogate that may not track hepatic receptor biology, and this measurement therefore constrains our GH-resistance markers” (= GHR vs. HC, GHR vs. T01, to a peripheral-cell readout rather than a direct hepatic measurement, which should be considered when interpreting the 'classical GH-resistance' cluster in Figures 3 and 6). We also detailed in Methods 2.2 the description of GHR measurement, to note explicitly that it is a PBMC-based, not hepatic, assay.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for addressing the major concerns in an appropriate manner. The manuscript is now transparent regarding its exploratory nature, the statistical limitations are acknowledged, additional analyses strengthen the principal findings, and the interpretation is more balanced.

Author Response

Reviewer#1
Thank you for addressing the major concerns in an appropriate manner. The manuscript is now transparent regarding its exploratory nature, the statistical limitations are acknowledged, additional analyses strengthen the principal findings, and the interpretation is more balanced.

Response: We thank the Reviewer for his/her comments.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have revised the manuscript; however, a few minor corrections need to be done before publish the manuscript. 

1. Please increase the font size of Figures 3 and 4 to improve readability, particularly in the printed version.

2. Please ensure consistent formatting of abbreviations (e.g., GDF-15, IGFBP-2, GHBP) throughout the manuscript.

3. Please ensure consistent formatting of abbreviations (e.g., GDF-15, IGFBP-2, GHBP) throughout the manuscript.

Author Response

Reviewer#2
The authors have revised the manuscript; however, a few minor corrections need to be done before publish the manuscript. 
[1]. Please increase the font size of Figures 3 and 4 to improve readability, particularly in the printed version.
Response: In the revised version of the manuscript we have increased the fonts in Figures 3 & 4
[2]. Please ensure consistent formatting of abbreviations (e.g., GDF-15, IGFBP-2, GHBP) throughout the manuscript.

Response: In the revised version of the manuscript care was taken to ensure consistent formatting of abbreviations.

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript has substantially improved and most of my previous concerns have been adequately addressed. One remaining issue is that the manuscript has become considerably longer due to extensive methodological explanations and repeated statements that the analyses are exploratory. I encourage the authors to further streamline redundant descriptions, particularly in the Introduction, Methods, and Discussion, to improve readability without altering the scientific content.

Author Response

Reviewer#3
The manuscript has substantially improved and most of my previous concerns have been adequately addressed. One remaining issue is that the manuscript has become considerably longer due to extensive methodological explanations and repeated statements that the analyses are exploratory. I encourage the authors to further streamline redundant descriptions, particularly in the Introduction, Methods, and Discussion, to improve readability without altering the scientific content.

Response: In the revised version of the manuscript we aimed to be more concise, without altering content. Now the Introduction is 9% shorter compared to the relevant section in the last version, the Methods section is 2% shorter and the Discussion is 10% shorter.