Review Reports - Does the Integration of Inflammatory Markers with Clinical Scoring Systems Improve the Prediction of Neurosurgical Intervention in Emergency Department Patients Diagnosed with Traumatic Subarachnoid Hemorrhage?

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for the opportunity to evaluate this paper. First, the topic is clinically relevant, and the attempt to identify readily available parameters that could contribute to early risk stratification in tSAH is worthwhile. The manuscript is generally well structured, and the authors appropriately acknowledge several important limitations. It is suitable for this journal. Nevertheless, I have some recommendations:

I would recommend to describe the characteristics and radiological severity of the tSAH population in considerably greater detail. The major concern is the absence of information regarding the intracranial injuries accompanying tSAH. The decision to perform neurosurgical intervention is primarily driven by neurological and radiological findings, including associated subdural or epidural hematoma, cerebral contusions, mass effect, midline shift, hydrocephalus, and other surgically relevant abnormalities. The authors themselves acknowledge that these variables were unavailable. Was there any other accompanying trauma besides tSAH? There are usually many traumatic alterations in brain after TBI.

The authors should therefore provide, if available, information regarding:

- isolated tSAH versus tSAH associated with other intracranial lesions;

- presence of subdural/epidural hematoma and cerebral contusion;

- oedema, midline shift, or mass effect;

- radiological severity of tSAH;

- indications for the neurosurgical procedures.

The definition of the primary outcome is heterogeneous and requires further clarification.

The primary outcome combines craniotomy, craniectomy, EVD placement, hematoma evacuation, aneurysm clipping, and therapeutic intracranial/cervicocranial vascular procedures. These interventions may be performed for substantially different clinical indications and these are versatile neurosurgical procedures. Please comment on that.

Plus, among the 12 intervention cases, one patient underwent an endovascular intervention and three underwent cranial neurosurgery together with an orthopedic procedure. The manuscript would benefit from a table listing the exact neurosurgical procedures and, importantly, their clinical indications. It should also be clarified whether the intervention was specifically performed because of the tSAH itself or because of an associated traumatic intracranial lesion. If most procedures were performed for associated hematomas or other injuries rather than tSAH, the current framing as prediction of neurosurgical intervention “in tSAH” may be somewhat misleading.

The timing of blood sampling in relation to trauma should be reported. Please report the injury-to-blood-sampling interval, if available, and indicate whether substantial variation existed among patients. If these data are unavailable, this should be added as an explicit limitation.
The clinical meaning of WBC may be confounded by trauma severity. The authors interpret WBC partly as reflecting “systemic inflammatory load.” However, leukocytosis after major trauma can also represent a nonspecific physiological stress response and may correlate strongly with injury burden. Therefore, it is difficult to determine whether WBC provides biologically distinct information related to neuroinflammation or simply acts as another surrogate for severe trauma.
The statement that the 17.4% intervention rate “falls within this expected low range” should be reconsidered. The studies cited in the Discussion report substantially different intervention rates, including a cohort in which no patients required surgery. The comparison should be presented more cautiously because patient populations and tSAH severity may differ.

I recommend major revision.

Author Response

Response to Reviewers X Comments

We thank both reviewers for their careful and constructive evaluation of our manuscript. We have revised the manuscript extensively in response to every comment; changes are marked in the accompanying tracked-changes version. Below we address each point in turn, with reference to the relevant manuscript section.

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

# Reviewer 1 #

R1.1. Reviewer comment: "I would recommend to describe the characteristics and radiological severity of the tSAH population in considerably greater detail. The major concern is the absence of information regarding the intracranial injuries accompanying tSAH. The decision to perform neurosurgical intervention is primarily driven by neurological and radiological findings, including associated subdural or epidural hematoma, cerebral contusions, mass effect, midline shift, hydrocephalus, and other surgically relevant abnormalities. The authors themselves acknowledge that these variables were unavailable. Was there any other accompanying trauma besides tSAH? There are usually many traumatic alterations in brain after TBI.

The authors should therefore provide, if available, information regarding:

- isolated tSAH versus tSAH associated with other intracranial lesions;

- presence of subdural/epidural hematoma and cerebral contusion;

- oedema, midline shift, or mass effect;

- radiological severity of tSAH;

- indications for the neurosurgical procedures."

Author response: We thank the reviewer for this important point. We have reviewed the radiology-confirmed CT reports for all 12 surgical patients and added a detailed breakdown to Results Section 3.5. Eleven of the 12 interventions (91.7%) were performed in the presence of an identifiable structural lesion accompanying the tSAH — most commonly intraparenchymal/intraventricular hematoma, subdural hematoma, or significant mass effect (midline shift/edema) — rather than for tSAH volume alone. One case additionally demonstrated a cerebral aneurysm, which we now discuss explicitly (Limitations, point 10) given the possibility of an aneurysmal contribution. We have reframed the study's interpretation accordingly throughout the Discussion: the models should be understood as predicting neurosurgical intervention in tSAH-inclusive TBI presentations broadly, consistent with the established observation that isolated tSAH rarely constitutes the sole indication for surgery, rather than intervention attributable to the subarachnoid component specifically. We were unable to retrospectively quantify basal cistern effacement or precise SAH volume, as this would have required systematic re-review of original CT images beyond the scope of the written radiology reports available; this is now explicitly noted (Discussion, Marshall/Rotterdam paragraph).

R1.2. Reviewer comment: " The definition of the primary outcome is heterogeneous and requires further clarification.

Author response: We agree and have added a new table (Table 4: "Neurosurgical intervention procedures and indications") to Results Section 3.5, listing each intervention category, the specific procedure(s) involved, and the CT/operative-report-based indication, directly addressing whether each intervention was tSAH-specific or driven by an associated lesion. We also added a new paragraph explicitly acknowledging that our outcome captures the composite clinical need for neurosurgical intervention in tSAH-inclusive trauma, not intervention attributable to subarachnoid blood in isolation, and have softened the manuscript's framing accordingly (Discussion, Section 3.5 addition, and Limitations point 10).

R1.3. Reviewer comment: "The timing of blood sampling in relation to trauma should be reported. Please report the injury-to-blood-sampling interval, if available, and indicate whether substantial variation existed among patients. If these data are unavailable, this should be added as an explicit limitation."

Author response: The exact injury-to-sampling interval was not systematically recorded in the retrospective medical records and cannot be reliably reconstructed. We have added this as an explicit limitation (Limitations, point 1), noting that while all values were drawn as part of standard initial ED assessment, unmeasured variation in this interval cannot be excluded. The exact injury-to-sampling interval was not individually recorded for each patient in the retrospective records. However, per institutional ED protocol, intravenous access and initial blood sampling are obtained immediately upon hemodynamic stabilization for all trauma patients; for hemodynamically stable patients — who constituted the large majority of this cohort — laboratory samples and initial CT imaging were both obtained within approximately the first hour of ED arrival, providing a reasonably narrow and consistent early time window across patients. We have added this as an explicit limitation (Limitations, point 1), noting that formal per-patient timestamps were nonetheless not captured and that future prospective studies should explicitly record and standardize this interval.

R1.4. Reviewer comment: "The clinical meaning of WBC may be confounded by trauma severity. The authors interpret WBC partly as reflecting “systemic inflammatory load.” However, leukocytosis after major trauma can also represent a nonspecific physiological stress response and may correlate strongly with injury burden. Therefore, it is difficult to determine whether WBC provides biologically distinct information related to neuroinflammation or simply acts as another surrogate for severe trauma."

Author response: This is an excellent point, and we have directly tested it. WBC and ISS (our available index of overall injury severity) were only weakly correlated (r = 0.235), and when entered together in a multivariable model, both remained independently significant predictors of neurosurgical intervention (WBC OR = 3.949, p = 0.003; ISS OR = 2.610, p = 0.012). We have added this analysis and an appropriately hedged interpretation to the Discussion (WBC paragraph): while this suggests WBC is not simply a proxy for overall injury severity as captured by ISS, we explicitly acknowledge that we cannot fully exclude that WBC and ISS each capture only partially overlapping, non-specific components of physiological stress rather than a process specific to neuroinflammation.

R1.5. Reviewer comment: "The statement that the 17.4% intervention rate 'falls within this expected low range' should be reconsidered. The studies cited in the Discussion report substantially different intervention rates, including a cohort in which no patients required surgery. The comparison should be presented more cautiously because patient populations and tSAH severity may differ."

Author response: We agree and have revised this statement to explicitly acknowledge the substantial heterogeneity in reported intervention rates across studies (including a cohort with 0% intervention), attributing this variability to differences in injury severity distribution, referral patterns, and case-mix rather than presenting a single 'expected' rate (Discussion, opening paragraph).

Reviewer 2 Report

Comments and Suggestions for Authors

What was the rationale for selecting neurosurgical intervention as the primary outcome rather than mortality, functional outcome, or radiological progression?
How were the 69 patients identified, and can the authors provide a patient-selection flowchart showing screened, excluded, and included patients?
What was the exact timing of blood sampling relative to trauma, ED admission, CT imaging, and neurosurgical intervention?
Were WBC, neutrophil count, lymphocyte count, NLR, and CRP measured before any treatment that could influence inflammatory parameters?
How did the authors account for important potential confounders such as anticoagulant/antiplatelet therapy, infection, steroid use, smoking, and other inflammatory conditions?
Why were important CT characteristics such as SAH volume, associated subdural/epidural hematoma, cerebral contusion, midline shift, and basal cistern effacement not incorporated into the prediction model?
The study included only 12 neurosurgical events. How do the authors justify the reliability of the multivariable logistic regression with an EPV of only 6?
Why was bootstrap or other internal validation not performed to evaluate optimism and overfitting of the WBC+GCS model?
The WBC+GCS model had an AUC of 0.904, but the improvement over WBC alone and GCS alone was not statistically significant. How should the claimed incremental predictive value be interpreted?
Why were the diagnostic cut-offs derived from the same dataset using the Youden index? How might this affect their apparent diagnostic performance?
Can the authors provide positive predictive value, negative predictive value, likelihood ratios, and diagnostic accuracy in addition to sensitivity, specificity, and AUC?
Why was decision-curve analysis or another clinical utility analysis not performed to determine whether the WBC+GCS model provides meaningful benefit in clinical decision-making?
Why was ISS evaluated only as a secondary model, and what was the rationale for not developing a more comprehensive comparison between the WBC+GCS and WBC+ISS models?
The surgery group had significantly higher lymphocyte counts, while NLR was not significantly different. How do the authors explain this finding in relation to the proposed inflammatory mechanism?
Why was CRP included despite its known delayed inflammatory response, and was the timing of CRP measurement standardized?
Could the elevated WBC primarily represent a marker of overall trauma severity or physiological stress, rather than an independent predictor of neurosurgical intervention?
How were the different types of interventions—craniotomy, craniectomy, EVD, hematoma evacuation, aneurysm clipping, and endovascular treatment—combined into a single outcome?
Why was the single endovascular intervention considered equivalent to the other neurosurgical procedures, and how might its inclusion influence the results?
Were there differences in trauma mechanism, age, comorbidities, or injury characteristics that could explain the observed differences between surgical and non-surgical groups?
Given the single-center retrospective design and small sample size, how generalizable are the reported AUC and WBC threshold of >11.20 × 10³/µL to other trauma centers?
Could the authors perform external validation or at least bootstrap validation before suggesting that WBC+GCS could be used for early risk stratification?
The manuscript reports that Claude was used for statistical computation, literature-search support, and manuscript preparation. Which specific statistical analyses were generated with AI, and how were these independently verified?
Can the authors provide the complete regression equation for the WBC+GCS prediction model so that it can be independently evaluated or validated?
Was model calibration assessed? If not, why was calibration not reported alongside discrimination?
Do the authors consider the proposed WBC+GCS model a prediction model or an exploratory association model, given the lack of internal/external validation and the small number of events?

Author Response

For research article; Does the Integration of Inflammatory Markers with Clinical Scoring Systems Improve the Prediction of Neurosurgical Intervention in Emergency Department Patients Diagnosed with Traumatic Subarachnoid Hemorrhage?

Response to Reviewers X Comments

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

# Reviewer 2 #

R2.1. Reviewer comment: "What was the rationale for selecting neurosurgical intervention as the primary outcome rather than mortality, functional outcome, or radiological progression?"

Author response: We have added an explicit three-part rationale to Methods Section 2.3: (1) neurosurgical intervention is the most immediate, actionable ED decision, unlike mortality which is a distal outcome shaped by many downstream factors; (2) mortality was too rare an event in this cohort to support meaningful multivariable analysis, providing even less power than the 12 intervention events analyzed; (3) radiological progression and standardized functional outcome scores were not systematically documented in the available retrospective records.

R2.2. Reviewer comment: "How were the 69 patients identified, and can the authors provide a patient-selection flowchart showing screened, excluded, and included patients?"

Author response: A patient-selection flow diagram has been added as Supplementary Figure S1, and the exact counts are also now stated in Methods Section 2.2. Of 1,884 patients with neurosurgical consultation requested in the Emergency Department during the study period, 148 had CT-confirmed tSAH; 71 were excluded for incomplete medical records, active infection/autoimmune disease, or corticosteroid/immunosuppressant use, and a further 8 pediatric patients were excluded, yielding the final analytic cohort of 69 adult patients (12 surgery, 57 no-surgery).

R2.3. Reviewer comment: "What was the exact timing of blood sampling relative to trauma, ED admission, CT imaging, and neurosurgical intervention?"

Author response: As noted in response to R1.3, this interval was not systematically recorded and is now explicitly acknowledged as a limitation (Limitations, point 1). As noted in response to R1.3, the exact blood-sampling and CT-imaging timing relative to injury was not individually recorded but is reasonably narrow and consistent for the hemodynamically stable majority of the cohort (Limitations, point 1). The timing of neurosurgical intervention itself relative to ED arrival was similarly not systematically captured as a discrete variable in the retrospective record.

R2.4. Reviewer comment: "Were WBC, neutrophil count, lymphocyte count, NLR, and CRP measured before any treatment that could influence inflammatory parameters?"

Author response: All laboratory values were drawn as part of the standard initial ED laboratory panel at presentation, prior to surgical or major pharmacological intervention; however, we cannot formally exclude minor variability (e.g., initial fluid resuscitation) before sampling in every case, and this is now covered by the timing limitation referenced above.

R2.5. Reviewer comment: "How did the authors account for important potential confounders such as anticoagulant/antiplatelet therapy, infection, steroid use, smoking, and other inflammatory conditions?"

Author response: Active infection, inflammatory/autoimmune disease, and corticosteroid or immunosuppressive use prior to admission were pre-specified exclusion criteria (Methods 2.3). Anticoagulant/antiplatelet therapy and smoking status were not systematically available in the retrospective records and could not be incorporated; we have added this as an explicit limitation.

R2.6. Reviewer comment: "Why were important CT characteristics such as SAH volume, associated subdural/epidural hematoma, cerebral contusion, midline shift, and basal cistern effacement not incorporated into the prediction model?"

Author response: These detailed radiological severity indices (analogous to Marshall/Rotterdam scoring) were not systematically available as structured data and would have required prospective re-review of original CT images, which was beyond the scope of this retrospective study; this is now explicitly discussed (Discussion, Marshall/Rotterdam paragraph, and Limitations point 7, already present in the prior version). It should be noted that our review of CT reports described in response to R1.1 was limited to confirming the documented surgical indication for the 12 patients who underwent neurosurgical intervention, and did not involve systematic re-analysis of the original imaging across the full cohort required to extract structured Marshall/Rotterdam-type severity indices.

R2.7. Reviewer comment: "The study included only 12 neurosurgical events. How do the authors justify the reliability of the multivariable logistic regression with an EPV of only 6?"

Author response: We agree this is an important limitation and have added explicit reference to Peduzzi et al.'s foundational simulation study establishing the EPV ≥ 10 convention (Limitations, point 2), together with a TRIPOD Type 1b classification (Methods 2.1) clarifying that this is model development with resampling-based internal validation but no external validation, and explicit hypothesis-generating framing throughout.

R2.8. Reviewer comment: "Why was bootstrap or other internal validation not performed to evaluate optimism and overfitting of the WBC+GCS model?"

Author response: We have now performed bootstrap internal validation using Harrell's optimism-correction method (2000 resamples). The apparent AUC of 0.904 corresponded to an optimism-corrected AUC of 0.896 (mean optimism = 0.007), suggesting overfitting was modest despite the low EPV, although meaningful uncertainty remains given the small sample (Results, end of Section 3.4/3.5 area; Limitations, point 8, updated).

R2.9. Reviewer comment: "The WBC+GCS model had an AUC of 0.904, but the improvement over WBC alone and GCS alone was not statistically significant. How should the claimed incremental predictive value be interpreted?"

Author response: We have revised the Discussion to state explicitly that the combined model's incremental value over either predictor alone is statistically unproven in this cohort and should not be overstated, while noting that both WBC and GCS remained independently significant in multivariable analysis, indicating complementary (not necessarily synergistic) information (Discussion, opening paragraph).

R2.10. Reviewer comment: "Why were the diagnostic cut-offs derived from the same dataset using the Youden index? How might this affect their apparent diagnostic performance?"

Author response: We agree these cut-offs are exploratory. Bootstrap 95% confidence intervals for each cut-off (2000 resamples) were already reported in the prior version of the manuscript (e.g., WBC 8.20–16.10) and are explicitly labeled as study-derived and exploratory, not externally validated clinical thresholds (Table 2 footnote; Results Section 3.3).

R2.11. Reviewer comment: "Can the authors provide positive predictive value, negative predictive value, likelihood ratios, and diagnostic accuracy in addition to sensitivity, specificity, and AUC?"

Author response: These have been added to Results Section 3.4: at the Youden-optimal threshold, PPV = 69.2% (95% CI 42.4–87.3%), NPV = 94.6% (95% CI 85.4–98.2%), LR+ = 10.69, LR− = 0.27, and overall accuracy = 89.9%.

R2.12. Reviewer comment: "Why was decision-curve analysis or another clinical utility analysis not performed to determine whether the WBC+GCS model provides meaningful benefit in clinical decision-making?"

Author response: We have now performed decision curve analysis across threshold probabilities of 5–50%. The WBC+GCS model showed positive net benefit across this entire range and outperformed a 'treat-all' strategy at threshold probabilities above approximately 15%, while consistently outperforming a 'treat-none' strategy; these findings have been added to the Results and are summarized in the Discussion, with the caveat that this analysis reflects apparent (in-sample) performance in a small single-center cohort.

R2.13. Reviewer comment: "Why was ISS evaluated only as a secondary model, and what was the rationale for not developing a more comprehensive comparison between the WBC+GCS and WBC+ISS models?"

Author response: ISS was analyzed as a secondary rather than primary model because it is not immediately available at the point of initial ED triage (its calculation requires complete injury documentation), unlike WBC and GCS; this rationale is stated in the Introduction and Methods. We have not combined GCS and ISS in the same model due to their high collinearity (r = −0.680, already reported in Methods 2.5).

R2.14. Reviewer comment: "The surgery group had significantly higher lymphocyte counts, while NLR was not significantly different. How do the authors explain this finding in relation to the proposed inflammatory mechanism?"

Author response: We thank the reviewer for identifying this important point, which also revealed an internal inconsistency in our prior text (which had incorrectly invoked lymphopenia). We have corrected this: lymphocyte count was significantly higher, not lower, in the surgery group, alongside elevated neutrophil count, consistent with a stress-related leukocytosis affecting both subpopulations in parallel rather than the classic neutrophilia-with-lymphopenia pattern. Because NLR is a ratio of two components moving in the same direction in this cohort, much of each component's discriminatory signal may be attenuated in the ratio itself (Discussion, NLR paragraph, revised).

R2.15. Reviewer comment: "Why was CRP included despite its known delayed inflammatory response, and was the timing of CRP measurement standardized?"

Author response: CRP was included as a widely available, routinely measured acute-phase reactant for comparative completeness, not because we expected it to be highly discriminatory at ED presentation; its delayed kinetics (rising only around 6 h after the inciting stimulus, peaking at approximately 48 h) together with its 19 h plasma half-life are explicitly discussed as the biological rationale for its poor early discriminatory performance in this cohort (Discussion, CRP paragraph, already present). Measurement timing was not separately standardized beyond the standard initial ED laboratory panel (see response to R2.3/R2.4).

R2.16. Reviewer comment: "Could the elevated WBC primarily represent a marker of overall trauma severity or physiological stress, rather than an independent predictor of neurosurgical intervention?"

Author response: Please see our response to R1.4 above, where we directly test and discuss this question using a WBC+ISS multivariable model.

R2.17. Reviewer comment: "How were the different types of interventions—craniotomy, craniectomy, EVD, hematoma evacuation, aneurysm clipping, and endovascular treatment—combined into a single outcome?"

Author response: These were combined under a pre-specified definition of neurosurgical intervention as any cranial or therapeutic intracranial vascular procedure directly addressing intracranial pathology (Methods 2.3). We have added text explicitly acknowledging that, despite differing surgical techniques, all 12 interventions were performed emergently for acute, life-threatening mass effect or vascular pathology, providing a shared clinical rationale for their inclusion under one outcome (Results 3.5). Furthermore, our subsequent review of CT/operative-report indications (added in response to R1.1/R1.2; Table 4) provides empirical support for this shared rationale: 11 of the 12 interventions (91.7%) were performed in the presence of an identifiable structural lesion accompanying the tSAH, indicating that these procedurally heterogeneous interventions were, in the great majority of cases, triggered by a comparable underlying pathology rather than by fundamentally different clinical drivers.

R2.18. Reviewer comment: "Why was the single endovascular intervention considered equivalent to the other neurosurgical procedures, and how might its inclusion influence the results?"

Author response: Chart review confirmed this procedure was therapeutic (not purely diagnostic) and performed for trauma-related vascular pathology directly associated with the tSAH (Results 3.5). A sensitivity analysis excluding this case (n = 68) yielded materially unchanged results, as already reported in the manuscript.

R2.19. Reviewer comment: "Were there differences in trauma mechanism, age, comorbidities, or injury characteristics that could explain the observed differences between surgical and non-surgical groups?"

Author response: These comparisons are reported in Table 1, which presents age, sex, comorbidities (diabetes, hypertension, active malignancy), and trauma mechanism distribution by surgical status, with associated p-values.

R2.20. Reviewer comment: "Given the single-center retrospective design and small sample size, how generalizable are the reported AUC and WBC threshold of >11.20 × 10³/µL to other trauma centers?"

Author response: We have strengthened the Limitations section to explicitly state that generalizability is limited by the single-center retrospective design (Limitations, point 1) and that the reported cut-offs are study-derived and exploratory, not externally validated (Table 2 footnote; Results 3.3).

R2.21. Reviewer comment: "Could the authors perform external validation or at least bootstrap validation before suggesting that WBC+GCS could be used for early risk stratification?"

Author response: Bootstrap internal validation has now been performed (see response to R2.8). External validation was not possible given the retrospective, single-center nature of this exploratory study; we have softened all clinical-application language accordingly and explicitly framed the WBC+GCS model as hypothesis-generating, pending external validation (Conclusions, revised).

R2.22. Reviewer comment: "The manuscript reports that Claude was used for statistical computation, literature-search support, and manuscript preparation. Which specific statistical analyses were generated with AI, and how were these independently verified?"

Author response: We have substantially expanded the AI-transparency statement (Methods 2.6) to specify that AI assistance was used for descriptive statistics, ROC/AUC analysis, DeLong comparisons, univariable/multivariable logistic regression, and bootstrap validation, all computed from the authors' source dataset; every output was independently cross-checked by the authors against the raw dataset and, for key analyses, manual or spreadsheet-based recalculation, before inclusion. All cited literature was independently retrieved and verified by the authors. The verified underlying statistical outputs and computation logs can be provided to the editors or reviewers upon request, should further independent verification be desired.

R2.23. Reviewer comment: "Can the authors provide the complete regression equation for the WBC+GCS prediction model so that it can be independently evaluated or validated?"

Author response: The complete equation, in raw predictor units, has been added to Results Section 3.4: Logit(P) = −1.3676 + (0.2764 × WBC) + (−0.3051 × GCS), with P = 1/(1+e−logit).

R2.24. Reviewer comment: "Was model calibration assessed? If not, why was calibration not reported alongside discrimination?"

Author response: We have now performed a calibration assessment. Calibration-in-the-large showed close agreement between mean predicted probability and observed event rate (0.174 vs. 0.174); a 3-group Hosmer-Lemeshow test was not statistically significant (χ² = 3.19, p = 0.074), though we explicitly note this test is underpowered given the small event count and should be interpreted with caution. This has been added to Results and the Limitations section has been revised accordingly.

R2.25. Reviewer comment: "Do the authors consider the proposed WBC+GCS model a prediction model or an exploratory association model, given the lack of internal/external validation and the small number of events?"

Author response: We agree the model should be regarded as exploratory/hypothesis-generating rather than a validated clinical prediction model at this stage; we have made this explicit throughout the manuscript (Abstract Conclusions, Discussion, and Conclusions section), consistent with its TRIPOD Type 1b classification (Methods 2.1).

We believe these revisions substantially strengthen the manuscript's methodological transparency and appropriately calibrate its clinical claims. We thank the reviewers again for their thorough and constructive feedback.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for the revisions. The authors have corrected and improved the paper and I think it is ready now for the acceptance.

Reviewer 2 Report

Comments and Suggestions for Authors

Authors address all the comments, accept in present form of the manuscript