Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Overview: This is a cross-sectional study addressing a clinically relevant question—namely, the added value of inflammatory indices beyond the comprehensive geriatric assessment (CGA) in predicting frailty. The statistical analysis is fairly thorough, incorporating ROC analysis, bootstrap confidence intervals, and cross-validation. However, several conceptual and methodological issues should be addressed before the manuscript can be considered for publication, as detailed below.

 

Lines 39–41: The cited frailty prevalence is derived from a meta-analysis of community-dwelling older adults (ref. 4), whereas the present study population is clinic-based, with a median age of 84 years (see Table 1)—a substantially different context. The nature of the study setting (e.g., referral clinic, geriatric outpatient clinic) should be clearly stated early in the Introduction so that readers do not mistakenly assume the sample is community-based. The authors should consider revising or supplementing this information to ensure an appropriate basis for comparison.

 

Lines 133–135: The text states that 16 of 300 records were excluded "because of missing core analytic variables or invalid values." However, the Methods section (lines 82–85) also lists a separate set of clinical exclusion criteria (e.g., acute infection, malignancy). No information is provided regarding how many patients were excluded based on these clinical criteria. The authors should consider revising or adding this information accordingly.

 

Lines 215–221: The comparison of frailty prevalence with a community-based study (20.5%) is reasonable; however, a comparison with studies conducted in similar clinical or hospital-based settings (e.g., geriatric outpatient clinics) would provide a more appropriate context for interpreting the findings of this study.

 

Lines 250–258 (Strengths): This section is generally well written; however, the generalizability of the findings to the broader population of adults aged 65 years and older—as implied by the title and abstract—is questionable, given that the study population had a median age of 84 years and was drawn from a clinic-based setting. This limitation should be acknowledged more explicitly.

 

However, the reference list contains duplicated numbering prefixes (e.g., "1. 1. Clegg…", "2. 2. Hoogendijk…") throughout (lines 298, 300, 302, 305, etc.). This formatting error must be corrected prior to publication.

 

Author Response

Response to Reviewer 1 Comments

Manuscript ID: healthcare-4554991
Title (revised): Comprehensive Geriatric Assessment Domains and Routine Inflammatory Indices in Relation to Clinical Frailty Scale-Defined Frailty: A Cross-Sectional Study in Hospital-Based Older Adults
Journal: Healthcare (MDPI)

1. Summary

Thank you very much for taking the time to review this manuscript and for comments that improved it substantially. Two of your points, the clinic-based nature of the sample and the incomplete account of the exclusions, led us to rewrite the framing of the study and the whole of Section 2.2, and one point identified a formatting fault whose cause we had not diagnosed. Please find the detailed responses below; the corresponding revisions are highlighted in yellow in the re-submitted manuscript. All analyses were re-run on the original dataset, and the new analyses described below are additions rather than replacements.

2. Questions for General Evaluation

Reviewer’s Evaluation	Response and Revisions
Does the introduction provide sufficient background and include all relevant references? — Can be improved	Revised. The Introduction now states that reference [4] describes community-dwelling populations, adds a sentence explaining why such estimates do not transfer to hospital-based patients, and names the study setting in the aim paragraph. Six references have been added (28–32 and 35).
Is the research design appropriate? — Can be improved	The design is unchanged, but it is now described accurately. The sampling strategy, the study period and the setting are specified in Section 2.1, the two-stage eligibility process is set out in Section 2.2 with a STROBE flow diagram (new Figure 1), and a post hoc assessment of sample size adequacy has been added.
Are the methods adequately described? — Yes	Thank you. Section 2.5 has nevertheless been expanded with the sample size assessment, calibration and optimism procedures, and six sensitivity analyses.
Are the results clearly presented? — Yes	Thank you. Table 4 now also reports confidence intervals, optimism-corrected discrimination and calibration slopes, and two new paragraphs report the sensitivity analyses.
Are the conclusions supported by the results? — Can be improved	Revised. Predictive wording has been removed throughout and the Conclusions have been rewritten around contemporaneous association, in line with Reviewer 2’s suggested formulation.
Are all figures and tables clear and well-presented? — Yes	Thank you. Table 1 row labels now state the reporting format for each variable, and a STROBE flow diagram has been added as Figure 1.

 

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: Overview: This is a cross-sectional study addressing a clinically relevant question, namely the added value of inflammatory indices beyond the comprehensive geriatric assessment (CGA) in predicting frailty. The statistical analysis is fairly thorough, incorporating ROC analysis, bootstrap confidence intervals, and cross-validation. However, several conceptual and methodological issues should be addressed before the manuscript can be considered for publication.

Response 1: We are grateful for this assessment. We accept the underlying point that the original manuscript used the language of prediction for an analysis that can only describe classification at a single point in time, and we have changed this throughout: in the title, the abstract, the aim paragraph, Section 2.5, the Results and the Conclusions. We have also added the analyses needed to substantiate, rather than merely acknowledge, the two methodological concerns you and Reviewer 2 raise: a post hoc sample size assessment, calibration and bootstrap optimism correction, and two reduced models that quantify how much of the model’s performance depends on measures that share content with the Clinical Frailty Scale (CFS).

Comments 2: Lines 39–41: The cited frailty prevalence is derived from a meta-analysis of community-dwelling older adults (ref. 4), whereas the present study population is clinic-based, with a median age of 84 years (see Table 1), a substantially different context. The nature of the study setting (e.g., referral clinic, geriatric outpatient clinic) should be clearly stated early in the Introduction so that readers do not mistakenly assume the sample is community-based.

Response 2: We agree; this was a fair criticism of the original framing. Reference [4] is now explicitly labelled as a community-dwelling estimate, and the following sentence has been added immediately after it:

“These estimates describe community-dwelling populations and are not directly transferable to hospital-based older patients, among whom age, functional dependence, and comorbidity burden are typically higher.”

The setting is now also stated in four other places so that the reader cannot mistake the sample: the title ends “… A Cross-Sectional Study in Hospital-Based Older Adults”; the abstract describes “consecutive records of adults aged 65 years or older referred for comprehensive geriatric assessment at a university hospital”; the aim paragraph of the Introduction specifies “a hospital-based sample of adults aged 65 years or older who were referred for comprehensive geriatric assessment at a university hospital”; and Section 2.1 gives the clinical service (a healthy ageing outpatient clinic run by the Department of Family Medicine) and the record period (August 2024 to August 2025) in full. We should add, since it bears on how the reader will place the sample, that the assessments were carried out in a healthy ageing outpatient clinic run by a department of family medicine rather than in a specialist geriatrics department. This is now stated in Section 2.1 and noted in the Strengths paragraph, because it means the measures examined here are those actually available to family physicians.

Comments 3: Lines 133–135: The text states that 16 of 300 records were excluded “because of missing core analytic variables or invalid values.” However, the Methods section (lines 82–85) also lists a separate set of clinical exclusion criteria (e.g., acute infection, malignancy). No information is provided regarding how many patients were excluded based on these clinical criteria.

Response 3: You identified a genuine inconsistency in our description, although not in the analysis itself. Section 2.2 has been rewritten to set out the eligibility process in two explicit stages. The clinical criteria (acute infection, active malignancy under treatment, systemic corticosteroid use, ongoing chemotherapy and acute inflammatory disease) were applied at the point of record identification, so the 300 screened records had already passed them and no record was excluded on clinical grounds at the later stage. We also now state explicitly that a previously treated malignancy in remission was not an exclusion criterion and was counted within the comorbidity burden, because 36 of the 284 analyzed patients (12.7%) carry that history. All 16 exclusions occurred at the second, data-quality stage and each is now reported by reason:

“Sixteen of the 300 screened records were excluded at this second stage: 5 because a core analytic variable was missing (Mini-Cog in 3, platelet count in 1, and albumin in 1), 4 because a geriatric score lay outside its instrument range (CFS in 1, Katz ADL in 1, Lawton IADL in 1, and Mini-Cog in 1), and 7 because a laboratory value lay outside the plausibility range (neutrophil count in 4, albumin in 2, and lymphocyte count in 1).”

A STROBE flow diagram showing the whole process has been added as the new Figure 1; the previous Figures 1 and 2 are now Figures 2 and 3.

Comments 4: Lines 215–221: The comparison of frailty prevalence with a community-based study (20.5%) is reasonable; however, a comparison with studies conducted in similar clinical or hospital-based settings (e.g., geriatric outpatient clinics) would provide a more appropriate context for interpreting the findings of this study.

Response 4: Added. The Discussion now reports our prevalence with a confidence interval (54.6%; 95% CI: 48.8–60.3) and compares it with an acute inpatient cohort graded with the same instrument, in which 51.9% of patients had a CFS score of 4 or higher [34], a figure much closer to ours than the community estimate. We also cite the Turkish validation of CFS version 2.0 in outpatients [35], which supports the applicability of the instrument in this health system. The same paragraph now works through, one by one, the reasons the community-based comparison is indirect (setting, age distribution, sampling strategy, instrument, cut-off and health system context), which also answers a related request from Reviewer 2.

Comments 5: Lines 250–258 (Strengths): This section is generally well written; however, the generalizability of the findings to the broader population of adults aged 65 years and older, as implied by the title and abstract, is questionable, given that the study population had a median age of 84 years and was drawn from a clinic-based setting. This limitation should be acknowledged more explicitly.

Response 5: We have changed the claim rather than only softening it. The title and abstract now state the setting, and the abstract closes by restricting the findings to “very old, hospital-based patients”. The Strengths paragraph now ends:

“These strengths do not imply generalizability: the analytic population had a median age of 84 years and was drawn from a hospital-based assessment service, so the findings describe very old patients already referred for geriatric assessment rather than the wider population of adults aged 65 years or older.”

The Limitations paragraph adds that the frailty prevalence reported here should not be read as a population estimate, and the sentence on primary care has been softened from “the findings support” to “the findings are consistent with”.

Comments 6: The reference list contains duplicated numbering prefixes (e.g., “1. 1. Clegg…”, “2. 2. Hoogendijk…”) throughout (lines 298, 300, 302, 305, etc.). This formatting error must be corrected prior to publication.

Response 6: Corrected, and we are grateful for the diagnosis. The cause was that the MDPI reference paragraph style (MDPI_8.1_references) carries automatic list numbering while the numbers had also been typed into the text, so Word rendered both. The manually typed numbers have been deleted and the list is now numbered once by the template. While correcting this we also expanded author lists that had been truncated to three names with “et al.” (references 4, 7, 8, 9, 13, 38, 39, 40 and 41), added DOIs, and checked every in-text citation against the list. The list now runs from 1 to 43 with no duplicate or missing numbers.

4. Response to Comments on the Quality of English Language

Point 1: The English is fine and does not require any improvement.

Response 1: Thank you. The newly written passages were prepared with the same care, and the whole manuscript was read through again after revision.

5. Additional clarifications

Two matters that arose while preparing this revision should be brought to your attention, since neither was raised in the reports.

First, to address the construct overlap that Reviewer 2 and the Academic Editor also raise, we fitted two reduced models rather than one. Omitting Katz ADL, Lawton IADL and Mini-Cog lowered the area under the curve from 0.866 to 0.779, and omitting Mini Nutritional Assessment and Timed Up and Go as well lowered it to 0.657, which no longer differed clearly from the basic clinical model (0.611). We report both, because reporting only the first would understate the overlap.

Second, while re-running the analyses we found that the recorded differential counts are not always internally consistent with the recorded total leukocyte count: in three records the neutrophil count exceeds the recorded total, and in 96 of the 284 analyzed records the sum of the neutrophil and lymphocyte counts exceeds it by more than 0.2 ×10³/µL. The adjusted estimates for NLR and SII are not stable to this. We report all three handlings in Table S4, have softened the corresponding sentence in the Results, and have added the point to the Limitations. The main finding is unaffected and is in fact better supported, since no index changed the area under the curve by more than 0.013 under any handling.

We hope that these revisions address your comments satisfactorily, and we thank you again for the time you gave to the manuscript.

On behalf of all authors,

Latife Merve Yıldız

Reviewer 2 Report

Comments and Suggestions for Authors Article summary   This single-center, cross-sectional observational study (n = 284) evaluates how Clinical Frailty Scale (CFS ≥5)-defined frailty relates to comprehensive geriatric assessment domains (function, nutrition, cognition, mood, and mobility) and five routine blood-based inflammatory/prognostic indices (NLR, PLR, SII, CAR, and IPI). Using multivariable logistic regression and ROC analysis, the authors compare a basic clinical model, a core geriatric model, and index-added models. They conclude that geriatric domains are more strongly associated with contemporaneous frailty classification, whereas routine inflammatory indices offer limited incremental discriminative value. They also present an exploratory 6-item geriatric burden score.   The research question is practical and clinically relevant for primary care and geriatrics. However, any causal or prospective predictive interpretation should be carefully limited because of the cross-sectional design and the conceptual overlap between the outcome (CFS) and the core predictors.   Comments regarding general concepts:   • The distinction between contemporaneous cross-sectional classification and actual longitudinal prediction should be maintained consistently throughout the manuscript.   • Defining the primary outcome as CFS ≥5 while including Katz ADL, Lawton IADL, MNA, Mini-Cog, and TUG as predictors creates substantial conceptual overlap between the outcome and the explanatory variables. Because the CFS incorporates functional and cognitive information, the high AUC of the core geriatric model may partly reflect construct convergence rather than independent predictive performance. This issue should be discussed more explicitly.   • Although the Discussion acknowledges this overlap, the limitation should also be reflected consistently in the Title, Abstract, Results, and Conclusions. Phrases such as “driven primarily by” or “explained” may be too strong for a cross-sectional analysis involving overlapping constructs. Consider using wording that refers to association or contemporaneous classification instead.   • The AUC of 0.866 should be interpreted cautiously and should not be taken as evidence of clinical utility by itself. If predictive language is retained, the authors should consider reporting calibration measures, clinically relevant thresholds, decision-curve analysis, and, where feasible, external validation.   Specific comments:   • Table 1 presentation and formatting: Table 1 combines continuous variables reported as median (IQR) with categorical variables reported as n (%) in the same columns. Please make the reporting format explicit for each variable, either in the row labels or in a sufficiently clear table footnote, and ensure that the table is visually aligned.   • Bibliographic formatting: The reference list contains duplicated numbering throughout, for example “1. 1.” and “2. 2.” Please correct the layout according to the journal’s specifications and double-check the mapping between in-text citations and the reference list.   • Sampling strategy: The manuscript states that 300 records were screened but does not specify the sampling methodology. Please clarify whether the records were selected consecutively, by convenience, retrospectively, or through another procedure, and specify the period and setting in which the records were identified.   • Sample size and power: No formal sample size calculation or power analysis is reported. Please justify whether the final sample of 284 participants was adequate for the multivariable logistic regression models and ROC comparisons, taking into account the number of events, predictors, and the planned analyses.   • Participant flow: The reasons for excluding 16 of the 300 initially screened records are not sufficiently detailed. Please provide the exact number of exclusions for each criterion, including missing variables, invalid values, acute infection, malignancy, steroid use, chemotherapy, acute inflammatory disease, and any other reason. Consider presenting this information in a STROBE flow diagram.   • Rationale for “prespecified validity ranges”: The use of prespecified validity ranges requires further methodological justification. Please explain why an upper age limit of 110 years was selected, state whether any records fell outside each threshold, and provide the clinical or statistical basis for these cutoffs. If no records were excluded by these filters, please clarify their practical purpose. If records were excluded, please discuss whether this procedure could have introduced selection bias and provide a sensitivity analysis if feasible.   • Operationalization of exclusion criteria: The term “clinically discordant laboratory values” is not sufficiently operationalized. Please define the criterion, describe who assessed discordance, state whether the assessment was blinded to frailty status, and report how many records were excluded for this reason.   • Construct overlap and circularity: There is substantial conceptual overlap between the CFS outcome and several predictors, particularly Katz ADL, Lawton IADL, and Mini-Cog. Consequently, the high AUC of the core geriatric model may partly reflect the reproduction of components incorporated into the outcome measure rather than independent predictive performance. Please address this issue directly in the Discussion and reframe the interpretation around contemporaneous association or classification rather than independent prediction. If feasible, consider a sensitivity analysis using a reference measure that does not share these components, or explain why such an analysis was not possible.   • Prevalence contextualization: When comparing the observed frailty prevalence with Asian community-dwelling studies, the Discussion should more explicitly consider differences in clinical versus community settings, age distribution, sampling strategy, assessment instrument, cutoff definitions, and health-system context. In particular, the use of the CFS in the present study may partly explain differences from studies using the Fried phenotype or other frailty instruments.   • Exploratory 6-item score: The 6-item score was derived and evaluated in the same dataset, without reported optimism correction or external validation. It should therefore be presented strictly as exploratory and hypothesis-generating. Please consider internal validation or optimism-corrected performance estimates and clearly state that external validation is required before clinical use can be considered.   • Model calibration: The manuscript reports discrimination measures, including AUC, Brier score, and Nagelkerke R², but does not provide calibration measures. If predictive framing is retained, please consider adding a calibration curve, calibration intercept, calibration slope, and a description of how optimism was assessed.   • PLR interpretation: The inverse adjusted association observed for PLR is of interest, but the increase in AUC is small and its confidence interval includes zero. Please frame this finding as exploratory and discuss possible residual confounding, medication effects, biological variation, or random variation.   • Conclusion reframing: Please avoid explanatory or prospective predictive assertions in the Conclusions. A more appropriately cautious formulation could be: “In this sample, comprehensive geriatric assessment domains demonstrated a stronger association with same-time CFS-defined frailty classification than routine blood-based inflammatory indices.”  

Author Response

Response to Reviewer 2 Comments

Manuscript ID: healthcare-4554991
Title (revised): Comprehensive Geriatric Assessment Domains and Routine Inflammatory Indices in Relation to Clinical Frailty Scale-Defined Frailty: A Cross-Sectional Study in Hospital-Based Older Adults
Journal: Healthcare (MDPI)

1. Summary

Thank you very much for a detailed and demanding review. Your central point, that defining the outcome as CFS ≥5 while using Katz ADL, Lawton IADL, MNA, Mini-Cog and TUG as explanatory variables creates substantial construct overlap, is correct, and it has changed both the analysis and the claims of the paper. We have not confined ourselves to acknowledging it in the Discussion: we have measured it, reported the measurement in the Results and in Table 4, and written the conclusions to match the weaker rather than the stronger of the two results it produces. We have also added the calibration, optimism correction, sample size assessment, threshold table and decision curve analysis you asked for. Detailed responses follow; revisions are highlighted in yellow in the re-submitted manuscript.

2. Questions for General Evaluation

Reviewer’s Evaluation	Response and Revisions
Does the introduction provide sufficient background and include all relevant references? — Yes	Thank you. Six references have nevertheless been added (28–32 and 35), covering sample size, calibration, decision curve analysis and a clinical-setting comparison.
Is the research design appropriate? — Must be improved	The design itself is unchanged, but its description and interpretation have been rewritten. The sampling strategy, period and setting are now specified; the eligibility process is given in two stages with a STROBE flow diagram; a post hoc sample size assessment has been added; and every claim has been reframed around contemporaneous classification.
Are the methods adequately described? — Must be improved	Section 2.2 has been rewritten in full (sampling, validity ranges and their rationale, exclusion counts) and Section 2.5 has been expanded with sample size, calibration, optimism correction, decision curve analysis and seven sensitivity analyses. The undefined term “clinically discordant laboratory values” has been removed.
Are the results clearly presented? — Must be improved	Table 4 now reports confidence intervals, optimism-corrected discrimination and calibration slopes, and two reduced models. Two new Results paragraphs report the sensitivity analyses and the operating characteristics. Four supplementary tables and three supplementary figures have been added.
Are the conclusions supported by the results? — Can be improved	The Conclusions now use your proposed formulation, state that a substantial part of the model’s advantage reflects construct overlap, and no longer claim that laboratory information supports frailty assessment.
Are all figures and tables clear and well-presented? — Must be improved	Table 1 row labels now state the reporting format for every variable and the footnote has been rewritten; Table 4 has been restructured; a STROBE flow diagram has been added as Figure 1; and the reference list numbering has been corrected.

 

3. Point-by-point response to the general comments

Comments 1: The distinction between contemporaneous cross-sectional classification and actual longitudinal prediction should be maintained consistently throughout the manuscript.

Response 1: Applied throughout. The title now names the design; the abstract describes “the cross-sectional classification of frailty” and ends by stating that the findings “do not establish prospective predictive value”; the aim paragraph of the Introduction closes with a sentence explaining that assessment, CFS grading and blood sampling were same-day, so all analyses address contemporaneous classification; Section 2.5 states that the models “are therefore not prediction models in the prognostic sense”; the Results describe classification rather than prediction; and the Conclusions follow your wording. We have checked the revised manuscript and no sentence now asserts prospective prediction.

Comments 2: Defining the primary outcome as CFS ≥5 while including Katz ADL, Lawton IADL, MNA, Mini-Cog, and TUG as predictors creates substantial conceptual overlap between the outcome and the explanatory variables. Because the CFS incorporates functional and cognitive information, the high AUC of the core geriatric model may partly reflect construct convergence rather than independent predictive performance. This issue should be discussed more explicitly.

Response 2: We agree, and because your comment names five measures we fitted three reduced models rather than one, so that the reader can see the effect of removing the components that are graded directly, the effect of removing the combination the Academic Editor identifies, and the effect of removing all five. The first model omits Katz ADL, Lawton IADL and Mini-Cog, whose content is what a clinician uses when assigning a CFS grade, and retains MNA and TUG alongside the basic clinical variables and GDS-15. The second omits MNA and TUG as well, leaving GDS-15 as the only geriatric measure. The third omits Katz ADL, Lawton IADL and TUG. The results are:

• Core geriatric model: AUC 0.866 (95% CI: 0.821–0.907).
• Without Katz ADL, Lawton IADL and Mini-Cog: AUC 0.779 (95% CI: 0.721–0.833); difference from the core model 0.088 (95% CI: 0.036–0.140); difference from the basic clinical model 0.167 (95% CI: 0.106–0.232).
• Also without MNA and TUG: AUC 0.657 (95% CI: 0.592–0.719); difference from the basic clinical model 0.046 (95% CI: −0.005 to 0.098), that is, no longer a clear difference.
• A third variant, omitting Katz ADL, Lawton IADL and TUG (the combination named by the Academic Editor): AUC 0.802 (95% CI: 0.747–0.853); difference from the basic clinical model 0.191 (95% CI: 0.118–0.264).
• Basic clinical model: AUC 0.611.

We report all three, because reporting only the most favourable would make the overlap look smaller than it is. The revised text therefore states that construct overlap accounts for a substantial part of the apparent advantage of geriatric assessment, that a clear advantage nevertheless remains once the directly graded components are removed, and that the discrimination of the full core geriatric model should be treated as an upper bound rather than an estimate of independent classification ability. We also note in the Discussion that MNA and TUG are measured rather than graded and are not items of the CFS, but are closely related to the domains a clinician weighs when assigning a grade. The first two reduced models appear as rows in Table 4, in a new Results paragraph, in a new Discussion paragraph, in Table S2 and in Figure S3. The same pattern holds when frailty is defined as CFS ≥6 (0.852 for the core model versus 0.739 for the first reduced model).

Comments 3: Although the Discussion acknowledges this overlap, the limitation should also be reflected consistently in the Title, Abstract, Results, and Conclusions. Phrases such as “driven primarily by” or “explained” may be too strong for a cross-sectional analysis involving overlapping constructs.

Response 3: Both phrases you quoted have been removed. The abstract previously concluded that frailty was “driven primarily by geriatric functional, cognitive, nutritional, and mobility domains”; it now reports that these domains “demonstrated a stronger association with same-time CFS-defined frailty classification”, and it also carries both reduced models. The Conclusions previously stated that the core geriatric model “explained same-time frailty classification substantially better”; they now follow your proposed formulation. The word “explained” survives elsewhere only in two ordinary and unrelated uses (“not explained by chronological age” in the Introduction, “may explain this discrepancy” in the Discussion). The Results paragraph on discrimination now ends:

“Because the core geriatric model shares information with the outcome, these values describe how closely the measured domains reproduce the CFS grading rather than independent classification ability.”

Comments 4: The AUC of 0.866 should be interpreted cautiously and should not be taken as evidence of clinical utility by itself. If predictive language is retained, the authors should consider reporting calibration measures, clinically relevant thresholds, decision-curve analysis, and, where feasible, external validation.

Response 4: We have removed the predictive language, and we have also added the measures you list rather than treating their removal as a reason to omit them. Table 4 now reports, for every model, the AUC with a bootstrap confidence interval, the bootstrap optimism-corrected AUC, the cross-validated AUC and the optimism-corrected calibration slope. Threshold operating characteristics are in Table S3, the calibration curve in Figure S1 and the decision curve analysis in Figure S2. External validation was not possible with these data; the Limitations now state that the optimism-corrected calibration slope of 0.81 implies that the coefficients would need shrinkage before use elsewhere. We add one caveat in the caption of Figure S2: because the predictors are themselves components of comprehensive geriatric assessment and the outcome was recorded at the same visit, the decision curve does not describe a decision a clinician actually faces, and we present it for completeness rather than as evidence of clinical utility.

4. Point-by-point response to the specific comments

Comments 5: Table 1 presentation and formatting: Table 1 combines continuous variables reported as median (IQR) with categorical variables reported as n (%) in the same columns. Please make the reporting format explicit for each variable, either in the row labels or in a sufficiently clear table footnote, and ensure that the table is visually aligned.

Response 5: Every row label in Table 1 now carries its own format marker, for example “Age, years, median (IQR)”, “Male sex, n (%)” and “Timed Up and Go, s, median (IQR)”. The footnote has been rewritten and now opens by stating the convention and that the applicable format is given in each row label. The table has been realigned in the journal template.

Comments 6: Bibliographic formatting: The reference list contains duplicated numbering throughout, for example “1. 1.” and “2. 2.”. Please correct the layout according to the journal’s specifications and double-check the mapping between in-text citations and the reference list.

Response 6: Corrected. The cause was that the MDPI reference paragraph style carries automatic list numbering while the numbers had also been typed into the text, so Word rendered both. The manually typed numbers have been deleted and the list is numbered once by the template. We also expanded author lists truncated to three names with “et al.”, added DOIs, and verified every in-text citation against the list. The list now runs from 1 to 43, every entry is cited, no citation refers to a number outside the list, and the order follows first appearance. Six references were added: Riley et al. [28] and Peduzzi et al. [29] for sample size, Van Calster et al. [30] and Steyerberg et al. [31] for calibration, Vickers and Elkin [32] for decision curve analysis, and Aşık et al. [35] for the Turkish validation of the CFS.

Comments 7: Sampling strategy: The manuscript states that 300 records were screened but does not specify the sampling methodology. Please clarify whether the records were selected consecutively, by convenience, retrospectively, or through another procedure, and specify the period and setting in which the records were identified.

Response 7: Section 2.1 now states that the records were identified retrospectively and consecutively from the hospital information system, that every record meeting the eligibility criteria within the study period was included, and that no sampling, matching or selection by frailty status was applied. The clinical service and the calendar period over which records were identified are given in the same sentence. The records come from a healthy ageing outpatient clinic run by the Department of Family Medicine at Recep Tayyip Erdoğan University Faculty of Medicine, rather than from a specialist geriatrics department; we have also noted this in the Strengths paragraph, since it means the instruments examined here are those available to family physicians in routine practice. The records are those of patients seen between August 2024 and August 2025, which is after the date of ethics approval (18 July 2024), so committee approval preceded access to the records rather than following it.

Comments 8: Sample size and power: No formal sample size calculation or power analysis is reported. Please justify whether the final sample of 284 participants was adequate for the multivariable logistic regression models and ROC comparisons, taking into account the number of events, predictors, and the planned analyses.

Response 8: A post hoc assessment has been added to Section 2.5, with the calculations in Table S1. The study analyzed every eligible record in the study period, so no a priori calculation was performed, and we say so rather than constructing one retrospectively. With 155 frail and 129 nonfrail participants and 11 parameters in the core geriatric model there are 11.7 events per variable for the smaller outcome group [29]. Against the criteria of Riley and colleagues [28], applied to the model actually fitted (apparent Cox–Snell R² 0.365; adjusted Cox–Snell R² 0.334; outcome proportion 0.546), the minimum sample required was 238 for an expected shrinkage factor of at least 0.9 and 236 for a difference of no more than 0.05 between apparent and adjusted Nagelkerke R². Both were met, and the shrinkage factor achievable at n = 284 is 0.92. We also report the criterion the study does not meet: estimating the overall outcome proportion within ±0.05 would have required 381 participants. We preferred to state this plainly, and it is repeated in the Limitations and reflected in the confidence interval we now give for frailty prevalence (54.6%; 95% CI: 48.8–60.3).

Comments 9: Participant flow: The reasons for excluding 16 of the 300 initially screened records are not sufficiently detailed. Please provide the exact number of exclusions for each criterion, including missing variables, invalid values, acute infection, malignancy, steroid use, chemotherapy, acute inflammatory disease, and any other reason. Consider presenting this information in a STROBE flow diagram.

Response 9: Section 2.2 now sets out the eligibility process in two stages. The clinical criteria were applied at the point of record identification, so the 300 screened records had already passed them and no record was excluded on clinical grounds at the later stage; we also state that a previously treated malignancy in remission was not an exclusion criterion, since 36 of the 284 analyzed patients (12.7%) carry that history. The 16 exclusions are reported individually: 5 for a missing core analytic variable (Mini-Cog in 3, platelet count in 1, albumin in 1), 4 for a geriatric score outside its instrument range (CFS 1, Katz ADL 1, Lawton IADL 1, Mini-Cog 1) and 7 for a laboratory value outside the plausibility range (neutrophil count 4, albumin 2, lymphocyte count 1). A STROBE flow diagram has been added as the new Figure 1.

Comments 10: Rationale for “prespecified validity ranges”: Please explain why an upper age limit of 110 years was selected, state whether any records fell outside each threshold, and provide the clinical or statistical basis for these cutoffs. If no records were excluded by these filters, please clarify their practical purpose. If records were excluded, please discuss whether this procedure could have introduced selection bias and provide a sensitivity analysis if feasible.

Response 10: Section 2.2 now distinguishes two kinds of range and explains both. For the geriatric instruments, the ranges simply reproduce the scores each instrument can produce (CFS 1–9, Katz 0–6, Lawton 0–8, MNA 0–14, Mini-Cog 0–5, GDS-15 0–15), so a value outside the range can only be a transcription or data-entry error; four records were excluded on this basis. For age, TUG time and the laboratory values, the limits are physiological plausibility bounds; the age limit of 110 years was chosen because it exceeds the oldest documented age in the national population by a wide margin and therefore flags only implausible date-of-birth entries. We now report which filters did and did not remove records: no record fell outside the age range, and the single record with a TUG time above 60 s had already been excluded because its Lawton IADL and Mini-Cog values were outside their instrument ranges; seven records were removed by the laboratory bounds. Because these filters remove only missing or impossible values, they cannot plausibly induce selection bias; the one exception in principle is the TUG ceiling, which could exclude the slowest walkers, and this is now stated in the Limitations. Two further sensitivity analyses bearing on selection are reported in Table S2: excluding the 36 participants with a recorded malignancy history (n = 248) left the core model AUC unchanged at 0.866, and excluding the 5 participants with CRP above 50 mg/L (n = 279) gave 0.863.

Comments 11: Operationalization of exclusion criteria: The term “clinically discordant laboratory values” is not sufficiently operationalized. Please define the criterion, describe who assessed discordance, state whether the assessment was blinded to frailty status, and report how many records were excluded for this reason.

Response 11: You are right that this phrase was not defensible as written, and we have removed it rather than attempt to define it after the fact. No record was excluded on a subjective judgement of discordance. The only laboratory rule applied was the objective plausibility range now stated in Section 2.2, which is applied automatically to the recorded values and is therefore independent of frailty status and requires no assessor. The phrase no longer appears anywhere in the manuscript.

Comments 12: Construct overlap and circularity: Please address this issue directly in the Discussion and reframe the interpretation around contemporaneous association or classification rather than independent prediction. If feasible, consider a sensitivity analysis using a reference measure that does not share these components, or explain why such an analysis was not possible.

Response 12: The two reduced models described under general comment 2 have been added, together with a dedicated Discussion paragraph that reports them and states the conclusion we draw. On the reference measure: a comparison against an instrument sharing none of the CFS components, such as the Fried phenotype, would indeed be the stronger test, but grip strength, gait speed, exhaustion and unintentional weight loss are not recorded in these routine records. We state this explicitly in the Discussion rather than leave it unexplained, and identify it as a priority for prospective work. The analysis we could perform, removing the overlapping components from the model rather than replacing the outcome, is reported in full, including the version in which the advantage over the basic clinical model disappears.

Comments 13: Prevalence contextualization: When comparing the observed frailty prevalence with Asian community-dwelling studies, the Discussion should more explicitly consider differences in clinical versus community settings, age distribution, sampling strategy, assessment instrument, cutoff definitions, and health-system context. In particular, the use of the CFS in the present study may partly explain differences from studies using the Fried phenotype or other frailty instruments.

Response 13: The prevalence paragraph has been rewritten to work through each of these in turn: the referral setting versus pooled community surveys; a median age of 84 years, close to the oldest stratum of the pooled estimate; consecutive clinical records versus population sampling; the CFS, which incorporates comorbidity, function and cognition, applied at a threshold of 5 or higher, against the Fried phenotype, which weights physical performance and unintentional weight loss and tends to classify fewer people as frail in the same population; and health system context, since referral pathways concentrate functionally impaired patients in hospital-based assessment records. A comparison with a clinical-setting CFS cohort has been added [34].

Comments 14: Exploratory 6-item score: The 6-item score was derived and evaluated in the same dataset, without reported optimism correction or external validation. It should therefore be presented strictly as exploratory and hypothesis-generating. Please consider internal validation or optimism-corrected performance estimates and clearly state that external validation is required before clinical use can be considered.

Response 14: Section 2.5 now states that the score is reported as an exploratory, hypothesis-generating summary because it was both derived and evaluated in the present dataset, and that its performance is presented with bootstrap optimism correction. The Results give the apparent AUC of 0.839 (95% CI: 0.792–0.882) together with the optimism-corrected value of 0.840, and add that these values “describe its behaviour in the present sample and should not be read as validated performance”. The figure caption states that the score was derived and evaluated in the same dataset and was not externally validated, and the Limitations state that external validation is required before any clinical use.

Comments 15: Model calibration: The manuscript reports discrimination measures, including AUC, Brier score, and Nagelkerke R², but does not provide calibration measures. If predictive framing is retained, please consider adding a calibration curve, calibration intercept, calibration slope, and a description of how optimism was assessed.

Response 15: Added. Section 2.5 describes the procedure and makes the necessary point explicitly: because every model was fitted and evaluated in the same dataset, apparent calibration is perfect by construction, so optimism in the AUC, the Brier score and the calibration slope was estimated by bootstrap resampling with 1000 replications, and optimism-corrected values are reported alongside apparent values [31]. For the core geriatric model the optimism-corrected calibration slope was 0.81, the calibration-in-the-large was 0.00 and the Hosmer–Lemeshow test was not significant (χ² = 15.11, df = 8, p = 0.06). Table 4 carries the calibration slope and the optimism-corrected AUC for every model, and Figure S1 shows the calibration curve with deciles of predicted risk and a loess smoother.

Comments 16: PLR interpretation: The inverse adjusted association observed for PLR is of interest, but the increase in AUC is small and its confidence interval includes zero. Please frame this finding as exploratory and discuss possible residual confounding, medication effects, biological variation, or random variation.

Response 16: The relevant Discussion paragraph now opens by calling the finding exploratory rather than biological, and lists residual confounding by conditions and medications that affect platelet and lymphocyte counts, day-to-day biological variation in a single measurement, and chance arising from the separate testing of five indices. We also report that the association was attenuated when participants with a recorded malignancy history were excluded (OR: 0.64; 95% CI: 0.45–0.90) and that under the CFS ≥6 definition it was no longer significant (OR: 0.79; 95% CI: 0.58–1.09; Table S2). A related caution now applies to NLR and SII, whose estimates were not stable to the handling of internally inconsistent leukocyte counts (see section 5 below and Table S4).

Comments 17: Conclusion reframing: Please avoid explanatory or prospective predictive assertions in the Conclusions. A more appropriately cautious formulation could be: “In this sample, comprehensive geriatric assessment domains demonstrated a stronger association with same-time CFS-defined frailty classification than routine blood-based inflammatory indices.”

Response 17: We have adopted your sentence. The Conclusions now open:

“In this sample of hospital-based older adults, comprehensive geriatric assessment domains demonstrated a stronger association with same-time CFS-defined frailty classification than routine blood-based inflammatory indices.”

They continue by stating that a substantial part of this advantage reflects the conceptual overlap between the CFS and the functional and cognitive measures it incorporates, although a clear advantage over basic clinical variables remained after the directly graded components were removed. We have also deleted the clause “frailty assessment may be supported by laboratory information”, which the data did not support, and replaced it with “routine laboratory indices added no measurable information beyond the geriatric domains”. The closing sentence now calls for multicenter prospective studies with external validation.

5. Additional clarification: a data-quality issue we found during the revision

While re-running the analyses we found that the recorded differential counts are not always internally consistent with the recorded total leukocyte count. In three records the neutrophil count exceeds the recorded total, and in 96 of the 284 analyzed records the sum of the neutrophil and lymphocyte counts exceeds it by more than 0.2 ×10³/µL. This matters because the adjusted estimates for NLR and SII are not stable to it: excluding the three impossible records changes them from null (OR 1.00 and 0.89) to inverse (OR 0.68, 95% CI: 0.48–0.96, and OR 0.65, 95% CI: 0.47–0.92), whereas restricting the analysis to the 188 records with an internally consistent differential returns them to null (OR 0.86 and 0.79). We report all three handlings in Table S4, have softened the corresponding sentence in the Results, and have added the point to the Limitations. The main finding is unaffected and is in fact better supported: no index changed the area under the curve by more than 0.013 under any handling.

6. Response to Comments on the Quality of English Language

Point 1: The English is fine and does not require any improvement.

Response 1: Thank you. The newly written passages were prepared with the same care, and the whole manuscript was read through again after revision.

We hope that these revisions address your comments satisfactorily. The manuscript is, in our view, a considerably more careful piece of work than the version you reviewed, and we are grateful for the rigour of your report.

On behalf of all authors,

Latife Merve Yıldız

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I thank the authors for a thorough revision. Most of my comments have been fully addressed, and I appreciate that construct overlap was quantified and that unfavourable results were reported.

Main points

1. Conclusions and Limitations. Both state only that a clear advantage remained after removing Katz ADL, Lawton IADL and Mini-Cog. As the Abstract and Discussion already do, they should also report that once MNA and TUG are removed as well, the advantage over the basic clinical model is no longer clear. The Discussion should also note that the MNA-SF includes mobility and neuropsychological items, which makes it closer to CFS content.
2. 6-item score. (a) The optimism-corrected AUC (0.840) exceeds the apparent AUC (0.839). The bootstrap only captures optimism from steps repeated in each resample. If items, cut-offs, weights and grouping were prespecified, there is nothing to correct, and the score should not be described as "derived" in these data. If any step was data-driven, it must be repeated in each resample. Please state how these were chosen. (b) Three of the six items (ADL dependence, IADL dependence, Mini-Cog <3) are the measures the authors identify as used directly in CFS grading, so the overlap caveat applies to the score as well. Ideally, please report its AUC without these items.
3. Clinical exclusions. My comment 9 asked for the number of exclusions for acute infection, malignancy, steroid use, chemotherapy and acute inflammatory disease. These criteria were applied before the 300 records were screened, so Figure 1 shows only data-quality exclusions. Please report how many records were identified in the study period and how many were excluded for each clinical criterion and confirm that 300 was not a predefined target.
4. Leukocyte inconsistency. (a) From Table S2, the inconsistent records comprise 65 of 155 frail participants (41.9%) and 31 of 129 nonfrail participants (24.0% aprox), and all three records with neutrophils above the total count were frail. The error is therefore differential, so it cannot be assumed to bias estimates towards the null. This bears directly on the conclusion about the indices and should be discussed. (b) Removing three records moves the OR for NLR from 1.00 to 0.68 (Table S4), which suggests influential extreme values. (c) The cross-reference in the paragraph introducing Table 2 should be to Table S4.
5. Overfitting. Table S1 gives a van Houwelingen shrinkage factor of 0.915, but the bootstrap calibration slope is 0.81. For a prespecified model these should be close, so please explain how the slope was obtained and reconcile the two.

Minor points

1. PLR: the non-significant result under CFS ≥6 and the borderline result in records with a consistent differential belong in the Discussion. The change from 0.60 to 0.64 after excluding malignancy history is not an attenuation.
2. CFS ≥6: please add both reduced models to Table S2.
3. The findings in [34] are not directly comparable to those of the present article. 
4. Discussion: please mention the third reduced model.
5. References 27 (missing DOI) and 37 ("et al." after three authors) still need correcting.

Author Response

Dear Reviewer,

Thank you for a second reading as careful as the first. Every numerical claim in your report is correct: we reproduced the distribution of the inconsistent leukocyte records across the outcome groups, and we confirmed that the optimism estimate for the 6-item score is indistinguishable from zero. Four of your main points have changed what the manuscript claims rather than only how it is worded, and we are grateful for them. One point, the per-criterion clinical exclusion counts, we are unable to answer, and we explain why below rather than offering a partial figure. All revisions are highlighted in yellow in the re-submitted manuscript.

MAIN POINTS

Comments 1: Conclusions and Limitations. Both state only that a clear advantage remained after removing Katz ADL, Lawton IADL and Mini-Cog. As the Abstract and Discussion already do, they should also report that once MNA and TUG are removed as well, the advantage over the basic clinical model is no longer clear. The Discussion should also note that the MNA-SF includes mobility and neuropsychological items, which makes it closer to CFS content.

Response 1: Corrected in both places. The Conclusions now read that a clear advantage remained after the directly graded components were removed, "when the Mini Nutritional Assessment and Timed Up and Go were removed as well, that advantage was no longer clear". The Limitations now state that the advantage "did not survive the further removal of the Mini Nutritional Assessment and Timed Up and Go". We have also added your point about the instrument itself to the Discussion: "The MNA-SF in particular contains a mobility item and an item on neuropsychological problems, so it is not a purely nutritional measure and shares content with CFS grading more directly than its name suggests."

Comments 2: 6-item score. (a) The optimism-corrected AUC (0.840) exceeds the apparent AUC (0.839). The bootstrap only captures optimism from steps repeated in each resample. If items, cut-offs, weights and grouping were prespecified, there is nothing to correct, and the score should not be described as "derived" in these data. If any step was data-driven, it must be repeated in each resample. Please state how these were chosen. (b) Three of the six items are the measures the authors identify as used directly in CFS grading, so the overlap caveat applies to the score as well. Ideally, please report its AUC without these items.

Response 2: (a) You are right, and the inconsistency you spotted is diagnostic of the error. Re-running the procedure gives a mean optimism of -0.0006 with a Monte Carlo standard error of 0.0005, that is, zero to within simulation noise, which is exactly what should be expected when nothing data-driven is repeated in the resamples. To answer your question directly: the six indicators and their thresholds were specified on clinical grounds before the score was examined. Each threshold is the standard published cut-off for the instrument concerned (Katz <6, Lawton <8, MNA <12, Mini-Cog <3) and is the same one already applied elsewhere in the analysis; the score is an unweighted count; and no item weights, thresholds or item selections were estimated from these data. We have therefore removed the optimism-corrected figure, removed the description of the score as "derived" in this dataset, and stated the above explicitly in Section 2.5. Apparent performance is now reported alone, with the note that bootstrap optimism was below 0.001 in absolute value.

(b) Added. A reduced score built only from malnutrition risk, fall risk and recorded depression history/diagnosis, that is, with the three CFS-graded items removed, had an AUC of 0.687 (95% CI: 0.629-0.743) against 0.839 for the full score. This is reported in the Results, and the Results now also state that "three of the six items are measures whose content is used directly in CFS grading, so the overlap described above applies to the score as well". The drop is of the same order as the one seen for the regression models, which we regard as a consistent and useful finding rather than an awkward one.

Comments 3: Clinical exclusions. My comment 9 asked for the number of exclusions for acute infection, malignancy, steroid use, chemotherapy and acute inflammatory disease. These criteria were applied before the 300 records were screened, so Figure 1 shows only data-quality exclusions. Please report how many records were identified in the study period and how many were excluded for each clinical criterion and confirm that 300 was not a predefined target.

Response 3: We can answer the second question but not the first, and we prefer to say so plainly. The clinical criteria were applied at the point of record identification, as part of the routine eligibility screening carried out before the analytic dataset was assembled, and the number of records removed under each criterion was not logged at that stage. We cannot reconstruct these counts retrospectively, and we are not willing to supply an approximation that could not be verified. The Limitations now state this explicitly: "The clinical exclusion criteria were applied when records were identified, as part of routine eligibility screening, and the number of records excluded under each criterion was not retained; we are therefore unable to report these counts." On your second question: 300 was not a predefined target. All records meeting the eligibility criteria within the study period of August 2024 to August 2025 were carried forward, and the screening was not stopped at a round number. The sentence already in Section 2.1 states this, and we have verified with the clinical team that it is accurate as written.

Comments 4: Leukocyte inconsistency. (a) From Table S2, the inconsistent records comprise 65 of 155 frail participants (41.9%) and 31 of 129 nonfrail participants (24.0% approx), and all three records with neutrophils above the total count were frail. The error is therefore differential, so it cannot be assumed to bias estimates towards the null. This bears directly on the conclusion about the indices and should be discussed. (b) Removing three records moves the OR for NLR from 1.00 to 0.68, which suggests influential extreme values. (c) The cross-reference in the paragraph introducing Table 2 should be to Table S4.

Response 4: (a) Your figures are exactly right; we recomputed them and obtained 65 of 155 frail (41.9%) and 31 of 129 nonfrail (24.0%), chi-square p = 0.002, with all three impossible records in the frail group. We had treated the inconsistency as non-differential without checking, which was an error of reasoning on our part. The Results now report the distribution, the Table S2 footnote gives it, and the Discussion has been rewritten: "The inconsistency is also differential with respect to the outcome, since it was almost twice as frequent among frail as among nonfrail participants and all three impossible records were frail. Measurement error of this kind cannot be assumed to bias estimates towards the null, and its direction here cannot be determined from the available data; the null findings for NLR and SII should therefore be read as uninformative about any underlying association rather than as evidence against one." The Limitations carry the same point. We note that this weakens the specific claim about NLR and SII while leaving the paper's principal finding intact, since no index improved discrimination by more than 0.013 under any handling of the data.

(b) Agreed, and now stated: "that the removal of three records moves the adjusted odds ratio for NLR from 1.00 to 0.68 shows that these estimates rest on a small number of influential extreme values".

(c) Corrected; the cross-reference now points to Table S4.

Comments 5: Overfitting. Table S1 gives a van Houwelingen shrinkage factor of 0.915, but the bootstrap calibration slope is 0.81. For a prespecified model these should be close, so please explain how the slope was obtained and reconcile the two.

Response 5: The two quantities were reported side by side without explanation, which was a fair thing to query. Section 2.5 now describes the derivation: "The optimism-corrected calibration slope was obtained by refitting the full model in each bootstrap resample and evaluating the calibration slope of that resample's linear predictor in the original dataset, averaged over replications." The model fitted in each resample is the same prespecified model, so no selection is involved. The difference from 0.915 is therefore not evidence of an unreported data-driven step but a difference between two estimators of the same quantity: the Van Houwelingen value is a closed-form approximation based on the likelihood ratio statistic, which is known to be the more optimistic of the two at this sample size and number of parameters, while the bootstrap value is empirical and incorporates the instability of the coefficients across resamples. We now say this in Section 2.5 and in the Table S1 note, and we use the bootstrap estimate when judging how much shrinkage the model would require.

MINOR POINTS

Comments 6: PLR: the non-significant result under CFS >=6 and the borderline result in records with a consistent differential belong in the Discussion. The change from 0.60 to 0.64 after excluding malignancy history is not an attenuation.

Response 6: Both corrected. You are right that a shift from 0.60 to 0.64 is not an attenuation, and the word has been removed; the text now reads that the estimate "was essentially unchanged". The two results you mention have been moved into the Discussion paragraph on PLR, which now reports that the association "was no longer significant when frailty was defined as a CFS score of 6 or higher (OR: 0.79; 95% CI: 0.58-1.09) and was only borderline among the records whose differential counts were internally consistent (OR: 0.66; 95% CI: 0.44-1.00)".

Comments 7: CFS >=6: please add both reduced models to Table S2.

Response 7: Added, together with the third reduced model and the basic clinical model for reference. Under CFS >=6 the values are 0.852 for the core geriatric model, 0.739 without Katz ADL, Lawton IADL and Mini-Cog, 0.630 when MNA and TUG are also removed, 0.782 without Katz ADL, Lawton IADL and TUG, and 0.576 for the basic clinical model. The Results now give the same sequence. The pattern reproduces the one seen at CFS >=5.

Comments 8: The findings in [34] are not directly comparable to those of the present article.

Response 8: Agreed, and the sentence has been qualified rather than left to imply a closer correspondence than exists. It now reads that estimates from other clinical settings "are closer to the present one, although none is directly comparable", and identifies the two specific differences: that cohort was an acute inpatient surgical population and the threshold applied was CFS >=4 rather than >=5.

Comments 9: Discussion: please mention the third reduced model.

Response 9: Added. The Discussion now reports that "a third reduced model, omitting Katz ADL, Lawton IADL, and TUG, gave an intermediate area under the curve of 0.802 (95% CI: 0.747-0.853)", which sits between the two models already described and makes the gradient across the three easier to follow.

Comments 10: References 27 (missing DOI) and 37 ("et al." after three authors) still need correcting.

Response 10: Reference 37 is corrected: the full author list now reads Moon, S.; Oh, E.; Chung, D.; Choi, R.; Son Hong, G.-R. We also re-checked every other entry and confirmed that the remaining uses of "et al." all follow ten listed authors, as the journal style requires. Reference 27 is a different case. This 2016 article in the Asian Pacific Journal of Cancer Prevention has no DOI registered with Crossref; we searched by title, author and bibliographic record and found none, and the journal did not assign DOIs to articles of that period. We have therefore left the reference without a DOI rather than supply one that cannot be resolved, and the full author list, volume and page range are given so that the article can be located. We would of course add a DOI if the editorial office is able to identify one.

We hope these revisions are satisfactory. Two of your points, the differential nature of the leukocyte inconsistency and the inapplicability of optimism correction to a prespecified score, corrected errors we would not have found ourselves, and the manuscript is more accurate for them.

On behalf of all authors,
Latife Merve Yıldız