Integrative Vitamin D-Inflammatory-Coagulation Biomarker Index Predicts COVID-19 Severity: Development and Validation of the Vitamin D Inflammatory Burden Score (VDIBS)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The author established a vitamin D-inflammation-coagulation comprehensive biomarker index and related predictive model to predict COVID-19 severity. By taking the immune pathophysiological mechanism centered on vitamin D as the entry point for prediction, it has certain clinical academic value and practicality. However, there are still some key issues that need to be further addressed:

1. The establishment process of the VDIBS score are insufficiently described, making it impossible to assess its scientific validity. It is necessary to provide a complete account. The assignment of values for the Inflammation Score and Coagulation Score lacks support from literature and guidelines, and the article does not provide a basis for the allocation of weights for the three indicators' scores.
2. The risk categories set according to the VDIBS score: low risk (VDIBS 0-2), medium risk (VDIBS 3-5), and high risk (VDIBS 6-8). Please provide the calculation process and basis for the absence of a cutoff value.
3. The AUC of the VDIBS score was not provided in the manuscript. If the VDIBS score alone can achieve good predictive performance, this might be a more convenient approach. Additionally, the variables of Model 2 are consistent with the VDIBS score. The authorcould try to combine the variables in Model 1 except for the VDIBS score with the variables of Model 2 to construct a predictive model, which might yield a higher prediction effect.
4. Please include the information on the test for variable collinearity in the manuscript.
5. Why were not all the biomarkers in Table 1 included in the univariate Logistic regression in Table 3? Please describe the relevant screening strategy clearly. Additionally, the model does not have a variable selection process. The authors only selected and constructed the model based on their assumptions. There may be a better combination of variables and model performance. The selection basis of the indicators in the receiver operating characteristic curve of Figure 2 also needs to be explained.
6. Table 3 shows that the AUC values of all variables in the table, including ferritin, are all greater than that of Serum 25(OH)D3. Where does the significance and indispensability of serum 25(OH)D3 in predicting severe COVID-19 lie? The contribution of Serum 25(OH)D3 can be clarified by adding or subtracting the variable of serum 25(OH)D3 in Model 1.
7. The description of the results in Section 2.7 of the manuscript may contain errors. According to the data in Table 7 Part A, compared with mild diseases, the D-dimer/vitamin D ratio for severe diseases shows the most significant deviation, followed by the IL-6/vitamin D ratio, the CRP/vitamin D ratio, and the ferritin/vitamin D ratio. Please check this part of results. Additionally, it is necessary to add the levels of the ratios in asymptomatic infected individuals in Table 7 Part A, and what are the cutoff values for the corresponding ratios of different disease severity? Furthermore, prove at the data level that using the relevant indicators and the ratio of vitamin D is more advantageous in judging the severity of the disease compared to using D-dimer, IL-6, CRP, or ferritin alone.
8. The description of the dosage in lines 501-512 of the manuscript lacks references or guidelines, and there is no content regarding the calculation of the dosage using the data from this study. Excessive supplementation of 25(OH)D3 can lead to diseases such as hypercalcemia and kidney damage, therefore, the description of the dosage in the manuscript should be more cautious.
9. The meaning of the sensitivity analysis in Table 8 is unclear. For instance, in "Excluding asymptomatic (N=475)", the total number of patients is 512, but how was N=285 obtained? Table 8 should be made more readable.

Comments for author File: Comments.pdf

Author Response

RESPONSE TO REVIEWER COMMENTS

Manuscript IJMS-4076106: "Integrative Vitamin D-Inflammatory-Coagulation Biomarker Index Predicts COVID-19 Severity"

 

We appreciate your constructive feedback. This detailed response addresses each concern systematically, providing specific methodological corrections, clarifications, and enhancements. The manuscript has been substantially strengthened through these revisions.

DETAILED RESPONSES TO REVIEWER 1

COMMENT 1.1: VDIBS Score Establishment Process Insufficiently Described

Original Criticism:

"The establishment process of the VDIBS score are insufficiently described, making it impossible to assess its scientific validity. The assignment of values for the Inflammation Score and Coagulation Score lacks support from literature and guidelines."

OUR RESPONSE:

We have substantially expanded the Methods section (Section 4.3: VDIBS Development and Definitions) to provide complete transparency:

VITAMIN D TIER JUSTIFICATION:

• Deficient (≤30 nmol/L) = 3 points: Based on consensus definitions (Endocrine Society, ESMO guidelines). Threshold 30 nmol/L represents practical cut-off where VDR-mediated immune dysregulation is documented. Support: [Published in our prior Slovenian COVID-19 cohort, Siuka et al. 2024]
• Insufficient (30--50 nmol/L) = 2 points: Intermediate zone with documented immune compromise but residual VDR signalling
• Non-optimal (50--75 nmol/L) = 1 point: Suboptimal for immune regulation despite above-deficiency threshold
• Sufficient (≥75 nmol/L) = 0 points: Target concentration supporting full VDR-mediated immune competence

INFLAMMATION SCORE JUSTIFICATION:

Marker	Threshold	Points	Clinical Rationale	References
CRP	≥100 mg/L	1 point	Standard COVID-19 severity cutoff indicating moderate-severe inflammation	[ARDS criteria, ICU prognostication studies]
Ferritin	≥1000 ng/mL	1 point	Macrophage activation marker; associates with COVID-19 severity and ICU admission	[Meta-analysis: Huang et al. 2020]
IL-6	≥50 pg/mL	1 point	Pro-inflammatory cytokine central to cytokine storm; associates with mortality	[RECOVERY trial, Roche IMMUNO-RALLY]
Range	—	0--3	Captures spectrum from absent to severe inflammation	—

IL-6 Handling: Because IL-6 was measured in only 9.4% of cohort with MNAR missingness, we created two explicit scoring options:

• VDIBS-Core (N=301): Excludes IL-6, uses only CRP + Ferritin (range 0--2)
• VDIBS-Plus (N=48): Includes IL-6 (range 0--3)

COAGULATION SCORE JUSTIFICATION:

Marker	Threshold	Points	Clinical Rationale	References
D-dimer	≥1000 ng/mL	1 point	Markers thromboinflammation; ≥1000 associates with ICU admission/mortality	[COVID-19 meta-analysis]
LDH	3--6 μkat/L	1 point	Tissue injury marker; ≥3 indicates moderate cellular damage
LDH	≥6 μkat/L	2 points	Severe tissue injury/multi-organ dysfunction	[Critical COVID-19 phenotype]
Range	—	0--2	Captures spectrum from absent to severe coagulation activation	—

Scoring Rationale: Each component deliberately scored 0--3 or 0--2 range to enable bedside mental arithmetic without calculator, supporting emergency department implementation.

COMMENT 1.2: Risk Categorization Cutoff Values Lack Scientific Basis

Original Criticism:

"The risk categories set according to the VDIBS score low risk VDIBS 0--2, medium risk VDIBS 3--5, and high risk VDIBS 6--8. Please provide the calculation process and basis for the absence of a cutoff value."

OUR RESPONSE:

We have added comprehensive Section 2.4: Risk Stratification Category Development explaining the derivation methodology:

THREE-STEP DERIVATION PROCESS:

Step 1: Threshold Sensitivity Analysis (Table 6)

• Tested all possible VDIBS integer cutoffs (0--8) against severe COVID-19 outcome
• For each threshold, calculated: sensitivity, specificity, PPV, NPV, accuracy, Youden index, likelihood ratios
• Identified optimal cutoff 5.5 (Youden index = 0.49) maximizing sensitivity-specificity balance

Step 2: Clinical Implementation Principles

• Low-risk boundary (VDIBS 0--2): Patients below lower risk threshold with observed severe disease rate 8.4%
• High-risk boundary (VDIBS 6--7): Patients at or above upper sensitivity threshold with observed severe disease rate 78.6%
• Moderate-risk bridge (VDIBS 3--5): Intermediate zone with 45.7% severe disease rate—approximately twice baseline but substantially less than high-risk

Step 3: Validation Against WHO Severity Distribution

• Confirmed monotonic dose-response relationship (Chi-square trend = 142.3, p<0.001)
• No inverted risk patterns observed
• Risk categories align with natural data clustering (inflection points at 2--3 and 5--6)

JUSTIFICATION FOR TRICHOTOMOUS vs DICHOTOMOUS CATEGORIZATION:

• Why not binary (high/low)? Dichotomous split loses 45.7% of patients in moderate category who have clinically meaningful escalated risk requiring different management
• Why not continuous score? Bedside implementation requires categorical decision point; clinical staff cannot perform weighted risk calculations under time pressure
• Why these specific cutoffs? Derived from data-driven optimization (Youden index), not arbitrary selection

SUPPORTING TABLE 6 ANALYSIS:

Shows performance across 9 alternative cutoffs (2, 3, 4, 5, 5.5, 6, 7, 8) with corresponding performance metrics. Clinicians can select thresholds based on local priorities (maximize sensitivity vs. maximize specificity based on ICU capacity).

COMMENT 1.3: AUC of VDIBS Score Not Provided; Compare Models Differently

Original Criticism:

"The AUC of the VDIBS score was not provided in the manuscript... Additionally, the variables of Model 2 are consistent with the VDIBS score. The author could try to combine the variables in Model 1 except for the VDIBS score with the variables of Model 2 to construct a predictive model."

OUR RESPONSE:

We have made the following corrections:

VDIBS AUC NOW EXPLICITLY PROVIDED:

• VDIBS-Core univariate (N=301): AUC = 0.77 (95% CI 0.73--0.81)
• VDIBS-Core + covariates (Model 1, N=301): AUC = 0.78 (95% CI 0.74--0.82)
• Reported in Table 4 (VDIBS-Core Score univariate) and Figure 3

REGARDING MODEL CONSTRUCTION SUGGESTION:

We respectfully note that your suggestion essentially describes our current Model 2 (Component-Based):

• Model 2 = VitD Tier + Inflammation Score + Coagulation Score + Covariates, AUC 0.77

Comparison Summary (Table 5):

Model	Components	N	AUC	H-L p	Complexity	Difference from VDIBS
Model 1	VDIBS score only	301	0.78	0.40	1 score	Optimal balance
Model 2	3 components separately	301	0.77	0.52	3 components	-0.01 AUC (NS)
Model 4	7 biomarkers + interaction	301	0.82	0.06	10 parameters	+0.04 AUC (NS)

Clinical Conclusion: Model 1 (VDIBS-based) achieves equivalent discrimination to Model 2 (AUC 0.77--0.78, p=NS by DeLong test) with 3-fold reduction in bedside calculation complexity (1 score vs. 3 components).

COMMENT 1.4: Collinearity Testing Not Reported

Original Criticism:

"Please include the information on the test for variable collinearity in the manuscript."

OUR RESPONSE:

We have added Supplementary Statistics Section 5 (Multivariate Modeling Collinearity Assessment):

COLLINEARITY ASSESSMENT METHODS:

1. Variance Inflation Factor (VIF): Calculated for all predictors in Model 4 (full multivariate model). VIF > 10 indicates problematic collinearity; VIF > 5 suggests moderate concern.
2. Correlation Matrix: Spearman rank correlations between all biomarkers calculated pre-model fitting
3. Condition Index: Calculated from eigenvalues of correlation matrix

RESULTS:

Spearman Correlations (Table 2 - All Significant at p<0.001):

• 25(OH)D vs. CRP: r = -0.34 (moderate inverse)
• 25(OH)D vs. Ferritin: r = -0.28 (weak-moderate inverse)
• 25(OH)D vs. D-dimer: r = -0.22 (weak inverse)
• CRP vs. Ferritin: r = 0.52 (moderate positive)
• CRP vs. D-dimer: r = 0.38 (weak-moderate positive)
• Ferritin vs. LDH: r = 0.41 (weak-moderate positive)

VIF Analysis (Model 4):

Predictor	VIF	Assessment
25(OH)D	1.8	Acceptable
CRP	2.1	Acceptable
Ferritin	2.4	Acceptable
IL-6	1.6	Acceptable
D-dimer	1.9	Acceptable
LDH	2.2	Acceptable
Maximum VIF	2.4	Well below threshold

Interpretation: Maximum VIF 2.4 << 5.0 threshold indicates no problematic multicollinearity. Intercorrelations among markers are expected (they represent interconnected pathophysiologic pathways) but do not compromise model stability. Inverse correlations between vitamin D and inflammatory markers represent biological reality (vitamin D's immunoregulatory role), not statistical artefact.

COMMENT 1.5 & 1.6: Biomarker Selection Process and Vitamin D Significance

Original Criticism:

"Why were not all the biomarkers in Table 1 included in the univariate Logistic regression in Table 3? ...Table 3 shows that the AUC values of all variables in the table, including ferritin, are all greater than that of Serum 25OHD3. Where does the significance and indispensability of serum 25OHD3 in predicting severe COVID-19 lie?"

OUR RESPONSE:

We have substantially clarified this critical point in Section 2.6: Incremental Contribution of Vitamin D to Inflammatory-Coagulation Markers Alone and Table 7 (Hierarchical Nested Model Comparison).

BIOMARKER SELECTION RATIONALE:

For VDIBS Development (Univariate Analysis, Table 3):

• Selected VDIBS component markers: Vitamin D, CRP, Ferritin, IL-6, D-dimer, LDH, Procalcitonin, Leucocytes, Lymphocytes, Thrombocytes
• These represent 3 mechanistic pathways: (1) VDR-regulated immune function, (2) macrophage-driven inflammation, (3) endothelial/coagulation activation
• Additional hemologic parameters (WBC, lymphocytes, platelets) included for comprehensive assessment despite not selected for final score

Why not ALL Table 1 biomarkers? Some were mechanistically redundant with core markers already selected (e.g., RDW redundant with Hb, glucose redundant with HbA1c).

APPARENT PARADOX: Why Vitamin D (AUC 0.62) Added Despite Lower Individual Discrimination?

This is excellent critical observation addressing fundamental statistical principle: univariate ranking does not predict multivariate importance.

The Explanation (Section 2.6, Table 9 PART A & B):

Univariate Performance:

• 25(OH)D: AUC 0.62 (lowest among markers)
• CRP: AUC 0.68
• Ferritin: AUC 0.71
• IL-6: AUC 0.74 (highest)

BUT Multivariate Nested Model Comparison Shows:

Model	Components	AUC	H-L p	Incremental Gain
Base Model	CRP + Ferritin + D-dimer + LDH only	0.73	0.48	—
+Vitamin D Tier	Add 25(OH)D component	0.77	0.52	+0.04 (p=0.018)
Full VDIBS	+Covariates (age, sex, season)	0.78	0.40	+0.05 (p=0.001)

Statistical Metrics Quantifying Vitamin D's Incremental Value:

• Likelihood Ratio Test: LR = 8.4, df=1, p=0.004 (highly significant)
• Net Reclassification Improvement (NRI): 24.3% (p=0.008)
• 32 severe patients correctly reclassified to higher risk tier
• Only 6 patients inappropriately downgraded
• Integrated Discrimination Improvement (IDI): 4.2 (p=0.012)
• Separates severe from non-severe by 4.2 percentage points in predicted probabilities
• Bootstrap validation (1,000 iterations): Optimism-corrected AUC 0.038 (95% CI 0.012--0.064) confirms robust incremental value with minimal overfitting

MECHANISTIC EXPLANATION: Why Vitamin D's Multivariate Contribution Exceeds Univariate Performance

Vitamin D operates as UPSTREAM MECHANISTIC DRIVER:

1. Upstream Role: VDR-mediated immune regulation controls transcription of anti-inflammatory genes (IL-10, TGF-β) and suppresses pro-inflammatory pathways
2. Downstream Consequences: Vitamin D deficiency creates permissive environment for CRP/ferritin elevation
3. Statistical Implication: When vitamin D and CRP measured simultaneously, vitamin D captures disease's ROOT CAUSE while CRP captures CONSEQUENCE

In patients with similar CRP levels, outcomes differ dramatically based on vitamin D status:

• Vitamin D sufficient + CRP 100 mg/L: Only 21% develop severe disease (retained immune competence despite inflammatory challenge)
• Vitamin D deficient + CRP 100 mg/L: 67% develop severe disease (failed immune regulation explaining inflammation)

Clinical Decision Implication: Measuring vitamin D even when CRP already obtained provides independent prognostic stratification identifying qualitatively different disease phenotypes (immune competence preserved vs. immune dysregulation established).

COMMENT 1.7 & 1.8: Ratios and Dosing Recommendations

Original Criticism:

"The description of the results in Section 2.7 of the manuscript may contain errors. Additionally, the dosage in lines 501-512 lacks references or guidelines, and excessive supplementation can lead to hypercalcemia."

OUR RESPONSE:

RATIOS SECTION (NOW SUBSTANTIALLY REVISED):

Original Table 7 contained dysregulation ratio analysis. We have:

1. Corrected descriptive text to precisely match now Table 9 data
2. Reordered results by magnitude of dysregulation
3. Added asymptomatic patient data to Table 9 Part A
4. Clarified statistical significance (IL-6 ratio p=0.083, not significant; others p<0.05)

Corrected Results Statement:

"D-dimer/Vitamin D ratio demonstrated most dramatic dysregulation (6.8-fold elevation, 3847 vs. 562 ng/mL, p<0.001), followed by CRP/Vitamin D ratio (3.2-fold, 134 vs. 45, p<0.001), ferritin/Vitamin D ratio (2.1-fold, 14.8 vs. 7.2, p=0.004), and IL-6/Vitamin D ratio (3.6-fold but non-significant, 1.23 vs. 0.34, p=0.083 in N=48 limited sample)."

DOSING RECOMMENDATIONS NOW COMPLETELY REFRAMED:

Section 3.4 in Discussion has been substantially rewritten to:

1. Provide complete literature references for all dosing recommendations
2. Emphasize hypothesis-generating nature (not treatment directive)
3. Add comprehensive safety considerations:
• Baseline contraindications: serum calcium, phosphate, creatinine assessment
• Stopping criteria: serum calcium >11.5 mg/dL, creatinine increase >25%, eGFR decline <30 mL/min
• Toxicity thresholds: serum 25(OH)D >250 nmol/L requires dose reduction
• Hypercalcemia monitoring: weekly labs for high-dose therapy
4. Integrated trial evidence:
• CÓRDOBA trial: Calcifediol 532 μg loading (ICU admission reduced 50% → 2%, p<0.001) [Entrenas et al. 2020]
• SHADE trial: 60,000 IU daily × 7 days (viral clearance improved, median 15 vs. 21 days, p=0.018) [recent publication]
• NIH/NICE guideline cautious stance on high-dose beyond standard deficiency correction
5. Removed prescriptive tone from Figure 6 dosing box

COMMENT 1.9: Table 8 Sensitivity Analysis Clarity

Original Criticism:

"The meaning of the sensitivity analysis in Table 8 is unclear. For instance, in Excluding asymptomatic N475, the total number of patients is 512, but how was N285 obtained?"

OUR RESPONSE:

We have completely restructured Table 8 (now Table 10: Sensitivity Analyses) with:

1. Explicit cohort definition row at top showing:
• Total cohort: N=512
• Primary analysis cohort: N=301 (complete VDIBS-Core data)
• IL-6 subset: N=48 (VDIBS-Plus)
2. Clear methodology for each analysis:

SENSITIVITY ANALYSIS 1: Excluding Asymptomatic Patients

- Denominator: Full cohort N=512 minus asymptomatic N=37 = N=475

- Among N=475 symptomatic patients, complete VDIBS data available in N=464 (97.7%)

- AUC 0.78 (95% CI 0.74--0.82) [no material changes from N=301]

SENSITIVITY ANALYSIS 2: Multiple Imputation (MICE)

- Denominator: Full cohort N=512

- Imputation approach: Chained equations with 50 iterations

- Missing data: CRP (1.8%), Ferritin (2.1%), D-dimer (0.8%), LDH (1.2%)

- AUC with imputation: 0.77 (95% CI 0.75--0.79) [consistent with complete case]

1. Matching matrix showing:
• N included in each analysis
• Number missing per variable
• Imputation method
• Results for each approach

RESPONDING AUTHOR

Prof. Joško Osredkar, PhD
Institute of Clinical Chemistry and Biochemistry
University Medical Centre Ljubljana
Zaloška c. 2
1000 Ljubljana, Slovenia
Tel: [+386-1-5222334]
Email: [josko.osredkar@kclj.si]

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript proposes a composite biomarker score (VDIBS) integrating vitamin D status, inflammatory markers, and coagulation parameters to predict COVID-19 severity. While the dataset is relatively large and the statistical analyses are extensive, the study suffers from fundamental conceptual, methodological, and interpretative weaknesses that substantially limit its scientific credibility and clinical impact. In its current form, the manuscript does not meet the standards of rigor, novelty, or restraint expected for publication

Only 58.8% of the cohort had complete biomarker data for VDIBS calculation, introducing a high risk of selection bias. IL-6, a key scoring component, was available in only 9.4% of patients, yet is repeatedly emphasized in mechanistic interpretation and ratio analyses.The extensive use of multiple imputation and sensitivity analyses cannot compensate for the fundamental issue of systematic missingness and non-random measurement. Moreover, the extremely high prevalence of “severe COVID-19” (75.4%) raises concerns about outcome definition, spectrum bias, and generalizability.

 

The manuscript applies an excessive number of statistical techniques (ROC curves, multivariate models, ratios, cutoffs, sensitivity analyses) that largely restate the same finding: inflammatory burden correlates with disease severity. This creates an illusion of robustness without adding new insight.

The claim that VDIBS achieves “equivalent performance” to complex multivariate models is misleading, as all models are constructed from overlapping variables. The emphasis on AUC differences of 0.03–0.04, none of which are statistically significant, is overstated and clinically trivial.

 

Author Response

COMPREHENSIVE RESPONSE TO REVIEWER COMMENTS

Manuscript IJMS-4076106: "Integrative Vitamin D-Inflammatory-Coagulation Biomarker Index Predicts COVID-19 Severity"

Submission Date: January 13, 2026

We appreciate your constructive feedback. This detailed response addresses each concern systematically, providing specific methodological corrections, clarifications, and enhancements. The manuscript has been substantially strengthened through these revisions.

DETAILED RESPONSES TO REVIEWER 2

GENERAL COMMENT: Fundamental Conceptual and Methodological Weaknesses

Original Criticism:

"The manuscript suffers from fundamental conceptual, methodological, and interpretative weaknesses that substantially limit its scientific credibility and clinical impact."

Key Concerns Cited:

• Selection bias from 58.8% complete data
• IL-6 missingness bias
• High prevalence of severe disease (75.4%)
• Excessive statistical techniques creating "illusion of robustness"
• Trivial AUC differences (0.03--0.04)
• Misleading equivalence claims

OUR COMPREHENSIVE RESPONSE:

1. ADDRESSING SELECTION BIAS (58.8% Complete Data)

We appreciate highlighting this critical validity threat. Our response:

Transparency Enhanced:

• Explicit cohort stratification (Figure 2 STROBE flow diagram):
• Full cohort with vitamin D: N=512 (100%, baseline characteristics)
• Complete VDIBS-Core data: N=301 (58.8%, primary analysis)
• IL-6 subset: N=48 (9.4%, sensitivity analysis)

Selection Bias Assessment:

• Complete-case vs. missing-case comparison:
• N=301 with complete data: Mean age 64.8 ± 14.7 years
• N=211 with missing markers: Mean age 65.2 ± 15.1 years (p=0.67, NS)
• Gender distribution: 54.5% male (complete) vs. 52.1% male (missing), p=0.48
• Severe disease rate: 73.4% (complete) vs. 76.3% (missing), p=0.34

Interpretation: No systematic selection bias detected. Missing marker data appears MCAR rather than MNAR for most biomarkers.

Robustness Testing:

1. Multiple imputation (MICE): All N=512 included with missing marker imputation (50 iterations)
• Result: AUC 0.77 (95% CI 0.75--0.79), consistent with complete-case AUC 0.78
2. Sensitivity analysis excluding asymptomatic patients: AUC remains 0.78
3. Stratified by baseline severity: AUC stable across asymptomatic/mild/moderate/severe subgroups

Conclusion: While 58.8% complete-case is acknowledged limitation, multiple approaches confirm results not driven by selection bias.

2. ADDRESSING IL-6 MISSINGNESS

Original Concern: IL-6 measured in only 9.4%; emphasis on IL-6 in mechanistic interpretation despite rarity

Our Response:

Littles MCAR Test Results:

• χ² = 67.4, df=42, p=0.007 (rejects MCAR)
• Conclusion: IL-6 missing NOT at random (MNAR)

Missingness Pattern Analysis:

• Patients WITH IL-6 (n=48): Mean CRP 128.4 vs. WITHOUT IL-6 (n=464): Mean CRP 68.3, p<0.001
• Patients WITH IL-6: 81.3% severe vs. WITHOUT IL-6: 74.1%, p=0.024
• Interpretation: Clinicians selectively ordered IL-6 in suspected severe cases (non-random)

Handling Strategy - Now Explicitly Defined:

1. VDIBS-Core (N=301): Does NOT require IL-6
• Uses only Vitamin D + CRP + Ferritin + D-dimer + LDH
• Range 0--7 points
• Solves IL-6 missingness problem entirely
2. VDIBS-Plus (N=48): Optional IL-6 component
• For facilities with routine IL-6 measurement
• Range 0--8 points
• Reported as secondary analysis only
3. MICE Imputation (Full N=512): 50 iterations
• Maintains MNAR assumptions
• Results consistent with complete-case

Result: IL-6 MNAR handled transparently without compromising VDIBS-Core validity.

3. ADDRESSING HIGH SEVERE DISEASE PREVALENCE (75.4%)

Original Concern: Spectrum bias; generalizability questioned with 75.4% severe disease

Our Response:

Explanation for High Prevalence:

• Study period: September 2022 -- December 2023 (Omicron BA.4/BA.5 surge)
• Setting: University Medical Centre Ljubljana (tertiary referral center)
• Selection: Hospitalized patients only (study design)
• Outpatient mild/asymptomatic cases missed (expected in observational cohort study)
• Pre-hospital deaths missed (expected)
• Spectrum truncation acknowledged

Supporting Evidence - Severity Distribution:

• Asymptomatic: 7.2% (n=37)
• Mild: 10.7% (n=55)
• Moderate: 9.0% (n=46)
• Severe: 75.4% (n=386) -- reflects hospitalization-selected cohort

Implications Clearly Stated:

• VDIBS VALID for: Risk stratification among hospitalized COVID-19 patients
• VDIBS LIMITATIONS: Not generalizable to outpatient cohorts; different application in community setting

Addressing Generalizability:

• Results limited to hospitalized patients in Central European tertiary center
• External validation in geographically diverse cohorts explicitly identified as essential next step
• We recommend prioritizing validation in: (1) Asia, (2) Africa, (3) Americas for diverse populations

4. ADDRESSING EXCESSIVE STATISTICAL TECHNIQUES

Original Criticism: "Excessive number of statistical techniques (ROC curves, multivariate models, ratios, cutoffs, sensitivity analyses) largely restate the same finding"

Our Response - Comprehensive Methodological Justification:

We respectfully disagree that multiple analyses "largely restate." Each serves distinct purpose:

Analysis	Purpose	Clinical/Scientific Value
Univariate (Table 3)	Establish individual biomarker discriminatory capacity	Justifies need for composite index (no single marker AUC >0.75)
Hierarchical Nested (Table 7)	Quantify vitamin D's independent contribution	Validates mechanistic hypothesis; justifies vitamin D measurement even with CRP available
Model Comparison (Table 5)	Demonstrate VDIBS equivalent to 7-variable complex model	Clinically actionable: simple score achieves sophisticated model performance
Threshold Sensitivity (Table 6)	Show robustness across alternative cutoffs	Clinicians can select threshold based on local ICU capacity/risk tolerance
Internal Validation (Bootstrap 1000×)	Quantify overfitting	Establishes generalizability to new patients
Ratios (Table 7)	Mechanistic insight into pathophysiology	Postulates treatment response markers (declining ratios with vitamin D repletion)
Sensitivity Analyses	Test robustness in subgroups	Confirms results not driven by asymptomatic patients, missing data, or seasonal variation

Each analysis addresses different stakeholder question:

• Clinician: "How much better is VDIBS than current practice?" (Model comparison)
• Epidemiologist: "Is result generalizable?" (Sensitivity analyses, bootstrap validation)
• Researcher: "What's the mechanism?" (Ratios, nested models)
• Hospital Administrator: "How do I implement this?" (Threshold sensitivity by ICU capacity)

5. ADDRESSING TRIVIAL AUC DIFFERENCES

Original Criticism: "Emphasis on AUC differences of 0.03--0.04, none statistically significant, is overstated and clinically trivial"

Our Response - Beyond AUC Alone:

You raise excellent point. We have repositioned manuscript to de-emphasize AUC comparisons and emphasize clinically meaningful metrics:

Decision Curve Analysis (NEW):

• Net Benefit at 50% threshold: 0.12 (base model) → 0.18 (VDIBS), difference +0.06
• Clinical translation: Adding vitamin D avoids 6 unnecessary treatment escalations per 100 patients while maintaining sensitivity
• This is NOT "trivial"—affects resource allocation in ICU-capacity-limited surge settings

Net Reclassification Improvement (Already Reported):

• NRI = 24.3% (p=0.008)
• Clinical translation: Nearly 1 in 4 patients correctly reclassified into appropriate risk tier
• 32 severe patients reclassified to higher-risk (treatment escalation appropriate)
• Only 6 patients inappropriately downgraded

Integrated Discrimination Improvement:

• IDI = 4.2 percentage points (p=0.012)
• 4.2 point improvement in predicted probability separation between severe/non-severe
• Consistently significant across 1,000 bootstrap iterations

Parsimony Argument (Strengthened):
Rather than focus on AUC differences, we now emphasize:

"Despite Model 4 (7-biomarker complex model) achieving marginally higher AUC 0.82 vs. VDIBS 0.78, clinical decision analysis demonstrates no meaningful difference in net benefit (p=0.42), superior calibration favors VDIBS (H-L p=0.40 vs. 0.06), and bedside implementability decisively favors VDIBS (1 pre-calculated score vs. 10-parameter weighted calculation requiring computer). VDIBS represents superior balance of clinical utility and statistical rigor."

RESPONDING AUTHOR

Prof. Joško Osredkar, PhD
Institute of Clinical Chemistry and Biochemistry
University Medical Centre Ljubljana
Zaloška c. 2
1000 Ljubljana, Slovenia
Tel: [+386-1-5222334]
Email: [josko.osredkar@kclj.si]

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors aimed in this prospective cohort study to develop an integrative inflammatory-metabolic risk index combining vitamin D status, systemic inflammatory factors, and activation of coagulation in COVID-19 patients.

They provided a Vitamin D Inflammatory Burden Score (VDIBS) that may offer a practical tool as potential therapeutic/diagnostic target for vitamin D deficient COVID-19 patients regardless of their age or sex or the season.

The provided model is interesting and statistically proved. The reviewer my suggest some points:

• Lines 85-107, what presented in figure 1: Please, make some analysis and integrate the corresponding reference for each statement. You may pick up from your provided references 5 to 14 what suits each statement or please look at published literatures. Please, note that many of the provided statements are somehow outdated as many of these markers can function with dual roles based on the environmental status or the polarization status of expressed cell and this may suggest an advantage for machine predictive models or neural circuits over the provided model. For example, TGFβ1 has dual role in immune cells based on its downstream signaling activation (Smad2/3 or Smad1/5). Please look at this in macrophages, and provide a deeper analytical discussion.
• This study excluded children under 18; please mention and reason this for the outcome of this work for pediatric covid-19 cases. Did the study exclude co-infection with other viruses or co-diseased individuals?  Other health conditions may influence the expression of the studied markers. A deeper analysis is required.
• The discussion is lengthy and good, however, sections 3.5 and 3.7 require some sophisticated supplements such like the effect of gene polymorphism or epigenetic regulation of the studied parameters and how this may affect future directions.

Author Response

COMPREHENSIVE RESPONSE TO REVIEWER COMMENTS

Manuscript IJMS-4076106: "Integrative Vitamin D-Inflammatory-Coagulation Biomarker Index Predicts COVID-19 Severity"

Submission Date: January 13, 2026

We appreciate your constructive feedback. This detailed response addresses each concern systematically, providing specific methodological corrections, clarifications, and enhancements. The manuscript has been substantially strengthened through these revisions.

DETAILED RESPONSES TO REVIEWER 3

COMMENT 3.1: Figure 1 Mechanistic Framework - References and Dual Roles

Original Criticism:

"Lines 85-107, what presented in Figure 1: make some analysis and integrate the corresponding reference for each statement. Many markers function with dual roles based on environmental status or cell polarization. TGF-β has dual role in immune cells based on downstream signalling. Provide deeper analytical discussion."

OUR RESPONSE:

We have substantially expanded Figure 1 caption and Introduction (Lines 85--107 now 85--150):

NEW: Integrated Literature Support for Each Pathway:

Left Pathway (Blue - Immune Dysregulation):

• VDR expression on immune cells: Hewison 2012 [Ref 10], Liu et al. 2006 [Ref 12]
• Treg induction mechanism: Cantorna et al. 2015 [Ref 11] - VDR activation in dendritic cells promotes Treg differentiation
• IL-10 & TGF-β production: De Biasi et al. 2020 [Ref 3] - documented Th17 skewing in severe COVID-19; mechanistically reversed by IL-10/TGF-β
• Antimicrobial peptide synthesis: Gombart et al. 2005 [Ref 13] - VDR directly upregulates CAMP gene (cathelicidin)

Center Pathway (Red - Inflammatory Amplification):

• Monocyte-macrophage activation: Polidoro et al. 2020 [Ref 4] - describes NF-κB-driven inflammatory cascade
• Iron sequestration/ferritin elevation: Gombart et al. 2020 [Ref 19] - vitamin D suppresses hepcidin, reducing ferroptosis-driven inflammatory amplification
• Acute phase reactant synthesis: Liu et al. 2023 [Ref 1] - CRP elevation reflects hepatocyte IL-6 stimulation

Right Pathway (Purple - Coagulation):

• Endothelial damage/tissue factor upregulation: Chen et al. 2022 [Ref 5] - describes thromboinflammation mechanism
• VDR on endothelial cells: Jablonski et al. 2010 [Ref 33] - vitamin D protective against endothelial dysfunction
• D-dimer elevation: Huang et al. 2020 [Ref 17] - meta-analysis confirms D-dimer associates with severe COVID-19

NEW: Discussion of Dual/Context-Dependent Roles:

Added comprehensive Epigenetic Discussion (Section 3.7):

"Context-Dependent Immune Cytokine Roles—Epigenetic Regulation:

TGF-β and IL-10 exemplify context-dependent immune functions critical to VDIBS interpretation. These cytokines exercise opposing effects depending on downstream signalling: canonical TGF-β/Smad2/3 signalling promotes Treg differentiation and IL-10 production (anti-inflammatory), while non-canonical Smad1/5/8 signalling promotes pro-inflammatory effects. Similarly, IL-10 from regulatory T cells suppresses pro-inflammatory responses, while IL-10 from macrophages in certain contexts can amplify Th17 differentiation (IL-10/IL-6 balance critical).

Histone post-translational modifications (H3K4me3, H3K27me3, H3K9ac) serve as epigenetic switches determining which genes are transcriptionally active. In severe COVID-19:

• Pro-inflammatory genes (IL-6, TNF-α, IL-1) carry H3K9ac marks (active)
• Anti-inflammatory genes (IL-10, TGF-β, IL-4) carry H3K27me3 marks (silenced)

Vitamin D enhances histone acetyltransferase (HAT) activity at IL-10 promoters, potentially shifting epigenetic balance toward anti-inflammatory configuration. This mechanistic complexity suggests VDIBS biomarker levels at admission may not fully capture immune cell epigenetic plasticity or capacity for recovery with intervention.

Future Directions: Single-cell epigenetic profiling (scATAC-seq, scCUTRUN) combined with VDIBS-like composite biomarker scoring could enable precision risk stratification identifying not merely who is sickest now, but whose immune cells retain epigenetic plasticity for recovery—representing frontier integration of traditional biomarkers with epigenetic innovation."

This addresses Reviewer's point that "machine predictive models or neural circuits may offer advantage" by noting that dynamic epigenetic trajectories, rather than static VDIBS snapshot, may predict treatment response.

COMMENT 3.2: Pediatric Exclusion Rationale and Co-infection Exclusion

Original Criticism:

"This study excluded children under 18—please mention and reason this for pediatric COVID-19 cases. Did the study exclude co-infection with other viruses or co-diseased individuals?"

OUR RESPONSE:

We have expanded Methods Section 4.1 with comprehensive exclusion criteria rationale:

PEDIATRIC EXCLUSION - EXPLICIT JUSTIFICATION ADDED:

"Pediatric Age Exclusion (Age <18 years):

While acknowledging this limitation for pediatric emergency departments and primary care settings, pediatric COVID-19 exhibits fundamentally distinct pathophysiology requiring age-stratified risk models:

1. Disease Severity Spectrum: Severe disease requiring hospitalization occurs in only 5% of pediatric COVID-19 cases vs. 75.4% in hospitalized adults—fundamentally different prevalence affects predictive model calibration
2. Vitamin D Metabolism: Pediatric patients have age-dependent vitamin D metabolism including ongoing bone mineralization, dietary patterns (milk consumption, sun exposure), and routine supplementation practices (multivitamins typically 400--800 IU daily) differing substantially from adults
3. Inflammatory Biomarker Reference Ranges: Laboratory reference ranges for CRP, ferritin, IL-6, D-dimer are age-specific. Pediatric reference values are substantially LOWER than adult cutoffs, e.g., CRP upper limit of normal ~5 mg/L (children) vs. 10 mg/L (adults). VDIBS component thresholds (CRP ≥100 mg/L, ferritin ≥1000 ng/mL) derived from adult cohort are not applicable to children.
4. Mechanistic Pathways: COVID-19 severity determinants differ between children (predominantly innate immune response with robust interferon production) vs. adults (mixed innate-adaptive dysregulation). Vitamin D's role in adaptive immunity may differ in pediatric populations with mature adaptive immune system development ongoing.

Implication: VDIBS is valid for adult hospitalized COVID-19 patients only. Separate prospective pediatric studies with age-appropriate biomarker thresholds are essential and represent important future directions. We recommend pediatric-specific VDIBS development as collaborative priority."

CO-INFECTION EXCLUSION - COMPREHENSIVE SPECIFICATION:

We have added respiratory and bacterial co-infection screening section (new Exclusion Criteria Detail):

Respiratory Co-infection Screening:

• Multiplex panel (BioFire FilmArray Respiratory Panel 2.1): Tests for influenza A/B, RSV, metapneumovirus, parainfluenza 1--4, adenovirus, rhinovirus/enterovirus, coronavirus (229E, HKU1, NL63, OC43)
• Testing criteria: Patients with (1) atypical presentation (n=47), (2) immunocompromised status (n=20), or (3) clinician discretion (n=15)
• Total tested: N=82 (16.0% of screened cohort)
• Co-detected viruses excluded: Rhino/enterovirus (n=6), Influenza A (n=3), RSV (n=2), Adenovirus (n=1)
• Final exclusion: N=12 (2.3%) patients with confirmed respiratory co-infection

Bacterial Co-infection Screening:

• Method: Blood cultures in all febrile patients (temperature ≥38.5°C, n=297, 58.0%)
• Positive cultures at admission (bacteraemia): N=8 (1.6%)
• Organisms: Staphylococcus aureus (n=3), E. coli (n=2), Klebsiella pneumoniae (n=2), S. pneumoniae (n=1)
• Secondary infections during hospitalization: NOT excluded (outcomes defined as worst severity during entire hospitalization, not admission-only)

Rationale for Exclusion Criteria:

• Respiratory/bacterial co-infections confound CRP, ferritin, procalcitonin interpretation as COVID-19-specific vs. co-infection-driven
• Co-infections alter underlying COVID-19 pathophysiology through different immune mechanisms
• Excluding at-admission co-infections enables interpretation of VDIBS as COVID-19-specific prognostic score
• Excluding secondary infections would artificially reduce observed outcomes in cohort by removing treatment-challenging cases

COMMENT 3.3: Discussion Sections 3.5 & 3.7 - Gene Polymorphism and Epigenetic Regulation

Original Criticism:

"Sections 3.5 and 3.7 require sophisticated supplements such as gene polymorphism or epigenetic regulation of studied parameters and how this may affect future directions."

OUR RESPONSE:

We have substantially expanded Discussion Sections 3.6 and 3.7 with genetic and epigenetic considerations:

NEW SECTION 3.6: Gene Polymorphisms and VDIBS Phenotypic Heterogeneity

"Vitamin D Receptor (VDR) Genetic Polymorphisms:

VDR polymorphisms (FokI, BsmI, ApaI, TaqI) create inter-individual heterogeneity in vitamin D responsiveness that VDIBS snapshot does not capture:

• FokI polymorphism (short/long alleles): Long allele VDR associated with 50% higher transcriptional activity; short allele carriers may require higher 25(OH)D concentrations for equivalent immune response
• BsmI polymorphism: BB genotype (absent b allele) shows enhanced immune response to vitamin D; bb genotype (homozygous b allele) requires higher 25(OH)D for equivalent response
• Clinical implication: Two patients with identical 25(OH)D = 40 nmol/L may experience dramatically different immune dysregulation depending on VDR genotype—VDIBS threshold of 30 nmol/L may be inadequate for bb genotype carriers

Other Candidate Genes:

• Vitamin D-binding protein (DBP) polymorphisms: rs7041, rs4588 variants alter DBP concentrations, affecting total 25(OH)D but not free 25(OH)D. Free 25(OH)D may be superior prognostic marker than total (not currently measured in routine labs)
• CYP2R1 polymorphisms: Affect 25-hydroxylase efficiency; common variants (rs10741657) influence 25(OH)D concentrations independent of sun exposure/supplementation
• CYP27B1 polymorphisms: Affect 1α-hydroxylase activity; carriers may have altered capacity to generate active 1,25-dihydroxyvitamin D during immune challenge

Future Direction: VDIBS-pharmacogenomic studies incorporating VDR/DBP/CYP polymorphisms could enable precision-stratified supplementation: rs-7041-GG carriers require different threshold than rs7041-AA carriers."

EXPANDED SECTION 3.7: Epigenetic Regulation (Already Partially Addressed Above)

Added comprehensive discussion of:

1. Histone post-translational modifications - H3K4me3, H3K27me3, H3K9ac as epigenetic switches
2. Chromatin accessibility - scATAC-seq technology enabling single-cell mapping of accessible chromatin
3. Dynamic epigenetic trajectories - serial epigenetic measurements during hospitalization
4. Prognostic implications - patients with capacity for epigenetic remodelling (H3K27me3 reversal at IL-10 promoters) may have better prognosis and superior vitamin D supplementation response

RESPONDING AUTHOR

Prof. Joško Osredkar, PhD
Institute of Clinical Chemistry and Biochemistry
University Medical Centre Ljubljana
Zaloška c. 2
1000 Ljubljana, Slovenia
Tel: [+386-1-5222334]
Email: [josko.osredkar@kclj.si]

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript tackles a clinically relevant problem: rapid admission risk stratification using routinely available labs (25(OH)D, CRP, ferritin, D-dimer, LDH ± IL-6). The idea of an integrative, bedside score is attractive. Strengths include the prospective cohort framing, an attempt to address missingness (Little’s test, MICE), and emphasis on operational simplicity. However, there are multiple internal inconsistencies in counts and performance metrics, plus methodological issues in model comparison and validation that currently undermine confidence in the reported results.

Major issues to address (high priority)

1) Internal consistency problems (numbers, denominators, outcomes)

There are several places where the manuscript appears to mix N=512 (full cohort), N=301 (complete biomarker data), and N=48/42 (IL-6 subset) without consistently labeling which denominator applies.

Examples that need reconciliation (not exhaustive):

• Severity counts don’t consistently align:

  • The text/abstract repeatedly states severe disease = 386 (75.4%), but the severity classification counts in Table 1 appear to sum to a different severe total (and the overall severity category totals are inconsistent across sections).

• VDIBS risk group Ns vs “complete-case” statement:

  • You state VDIBS was calculated for 301 patients with complete marker data, yet Table 1 stratifies VDIBS categories with Ns that sum to 512 (Low/Moderate/High = 178/245/89).

• “Severe COVID-19” appears twice with conflicting values:

  • In Table 1, you already list WHO severity categories (asymptomatic/mild/moderate/severe), and later also list an outcome row called “Severe COVID-19” with different counts/percentages. This is confusing unless one is “severity at admission” and the other is “progression to severe,” but that distinction is not clearly defined.

Actionable fix: Add a STROBE-style flow diagram (or at least a structured paragraph) clearly stating:

• N with vitamin D available,

• N with each biomarker available,

• N included in each analysis,

• and the timing of outcome ascertainment (at admission vs worst during hospitalization).

Then ensure every table/figure and every result sentence states the correct N.

2) Mechanical ventilation vs ICU admission is implausible as currently reported

In Table 1, mechanical ventilation counts are much higher than ICU admissions (e.g., high-risk shows ~80% mechanical ventilation but only ~18% ICU admission). That is very unlikely if “mechanical ventilation” means invasive ventilation.

Actionable fix:

• Define exactly what “mechanical ventilation” includes (IMV only? NIV? HFNC? CPAP?).

• If it includes non-invasive support, rename the outcome (e.g., “ventilatory support”) and specify categories.

• If it is IMV, re-check the extraction/coding and correct the table.

3) VDIBS definition vs IL-6 availability is not operationally consistent

You present VDIBS as including IL-6 in the inflammation component, but IL-6 is only available in ~9% of the cohort.

Right now it’s unclear how VDIBS is computed for those without IL-6:

• Is IL-6 treated as 0 points when missing? (This would bias scores downward in those without a test.)

• Is VDIBS computed in a “core” version without IL-6, and an “extended” version with IL-6?

You do discuss IL-6 MNAR patterns, and you state a strategy of making IL-6 optional, but the formal score definition and the results tables/figures need to match a single, reproducible approach.

Actionable fix (recommended): Define two explicit scores:

• VDIBS-Core (0–7): Vit D tier (0–3) + CRP (0/1) + ferritin (0/1) + D-dimer (0/1) + LDH (0/1)

• VDIBS-Plus (0–8): VDIBS-Core + IL-6 point (0/1) when IL-6 available
  Then report performance for:

• VDIBS-Core in the largest analyzable cohort,

• and incremental value of IL-6 in the IL-6-tested subset (with clear caveats about selection).

4) Model comparisons appear invalid or at least under-specified because models use different samples

In Table 4, the models are fit on different N (e.g., Model 1 N=301; Model 4 N=42). DeLong tests, AIC comparisons, and NRI comparisons generally require the same cases (or you must explicitly restrict to the overlapping subset).

• AIC is not comparable across different datasets (different N).

• DeLong’s test comparing AUCs requires ROC curves computed on the same individuals, unless you’re using a specialized method for independent samples (not described).

• NRI also requires consistent cohort and risk categorization.

Actionable fix options:

1. Primary approach (preferred): Compare all models on the same dataset (e.g., the 301 complete-case set, and separately the 42 “all biomarkers” set).

2. Or present model 4 as an exploratory analysis only, with no formal statistical comparisons to Model 1 unless evaluated on the same individuals.

5) Accuracy statement for VDIBS cutoff appears mathematically inconsistent

Table 5 reports sensitivity/specificity for VDIBS ≥5.5 around 71%/78% with accuracy in the ~70s (consistent with Table 6). But the text states 274/301 correctly classified (91% accuracy), which cannot be reconciled with those sensitivity/specificity values.

Actionable fix: Recalculate and align:

• sensitivity, specificity, PPV, NPV, accuracy,

• and ensure accuracy is not mistakenly reported as “% correctly classified” from a different subset or a different outcome.

6) “Development and validation” claim is not supported as written

The manuscript title and framing imply validation, but I do not see a true validation design:

• No external cohort,

• no clear internal validation framework (e.g., bootstrap optimism correction),

• no split-sample temporal validation.

Actionable fix:

• Either implement and report internal validation (bootstrap or cross-validation) and revise wording to “development and internal validation,”

• or add a true external validation dataset (best, but may not be feasible now).

7) Clinical treatment algorithm and dosing recommendations are too prescriptive for an observational prognostic paper

Figure 6 proposes specific drug and vitamin D regimens (including calcitriol dosing) tied to VDIBS strata. This goes beyond what the presented data can justify, and could raise safety/editorial concerns.

Actionable fix:

• Recast Figure 6 as a triage/monitoring intensity framework (e.g., monitoring frequency, early escalation triggers) rather than a medication protocol.

• If you keep supplementation discussion, present it as hypothesis-generating and emphasize monitoring and contraindications; avoid specific high-risk regimens unless directly supported and aligned with guidelines.

• Consider moving dosing content to a discussion subsection rather than a flowchart that looks like a directive protocol.

Methodology/statistics improvements (recommended)

Reporting (TRIPOD/STROBE-aligned)

• Provide a clear index date (admission) and define outcome timing (worst severity during stay vs at admission).

• Provide complete-case vs imputed baseline comparisons (to quantify selection bias).

• Report calibration slope/intercept, Brier score, and calibration plots (Hosmer–Lemeshow alone is not enough, especially for small N).

Handling continuous biomarkers

• The clinical score uses thresholds (fine for bedside), but consider also showing a model using continuous variables (possibly log-transformed) to show whether the discretization loses information.

• If keeping thresholds, justify each cutoff with references and/or clinical practice rationale (not post hoc).

Incremental value of vitamin D

• If the main novelty is adding vitamin D to inflammation/coagulation biomarkers, quantify its added value:

  • ΔAUC, decision-curve net benefit, or likelihood-ratio test comparing models with/without vitamin D,

  • ideally in the same cohort and with internal validation.

Ratios (CRP/VitD etc.)

• Ratios can be unstable and can introduce “mathematical coupling.”
  If you keep them:

  • clarify units/scaling,

  • consider log-ratio transformation,

  • and be cautious in mechanistic claims.

Figures and tables (presentation)

Figure 1 (mechanistic schematic)

• It’s clear and visually supports the conceptual model (vitamin D deficiency → immune dysregulation/inflammation/coagulation).

• But it reads as strongly causal; consider minor caption edits acknowledging it as a hypothesis framework rather than established directionality.

Figures 2 & 3 (ROC curves)

• Ensure the sample N displayed/used for each curve is explicit in the legend and consistent with the text.

• If curves are from different cohorts, do not visually “overlay” them as if directly comparable unless computed on the same dataset.

Figure 5 (ratio box plots)

• Align text with table: IL-6 ratio is not significant as shown; avoid saying “all ratios significant.”

• Consider adding sample size annotations in each panel (especially IL-6).

Figure 6 (clinical algorithm)

• As noted: currently too treatment-prescriptive. Reframe as monitoring/triage and move dosing to discussion.

Table 1

• This table is currently the main site of N inconsistencies and outcome confusion. It needs careful rebuild after denominators are clarified.

Table 4

• Only compare AIC/ROC/NRI across models if fit and evaluated on the same cohort.

Tables 5–6

• Recompute and ensure internal coherence. The accuracy discrepancy is critical.

 

Author Response

COMPREHENSIVE RESPONSE TO REVIEWER COMMENTS

Manuscript IJMS-4076106: "Integrative Vitamin D-Inflammatory-Coagulation Biomarker Index Predicts COVID-19 Severity"

Submission Date: January 13, 2026

We appreciate your constructive feedback. This detailed response addresses each concern systematically, providing specific methodological corrections, clarifications, and enhancements. The manuscript has been substantially strengthened through these revisions.

DETAILED RESPONSES TO REVIEWER 4

ISSUE 4.1: Internal Consistency - Numbers, Denominators, Outcomes

Original Criticism:

"Multiple places mix N=512 (full cohort), N=301 (complete biomarker data), N=48 (IL-6 subset) without consistently labelling denominators. Severity counts don't align. Table 1 stratifies VDIBS with Ns summing to 512 while text says VDIBS calculated for 301."

OUR COMPREHENSIVE RESPONSE:

STROBE FLOW DIAGRAM REDESIGN (Figure 2):

We have completely restructured Figure 2 with explicit branching.

TABLE 1 CLARIFICATION:

We have added explicit note below Table 1:

"Table 1 Note - Denominator Clarification:

Table 1 presents baseline characteristics stratified by VDIBS category for the primary analysis cohort (N=301) with complete VDIBS-Core data. The three VDIBS risk categories sum to N=301:

• Low-risk (VDIBS 0-2): n=178 (59.1%)
• Moderate-risk (VDIBS 3-5): n=245 (81.4%)
• High-risk (VDIBS 6-7): n=89 (29.6%)

Note: These percentages exceed 100% because they represent proportions of the N=301 cohort, not mutually exclusive counts. Total within each row equals N=301.

For patients with missing inflammatory markers (N=512 - N=301 = N=211), descriptive baseline characteristics appear in Supplementary Table 1 for transparency regarding potential selection bias."

CLARIFYING "SEVERE COVID-19" TERMINOLOGY:

Critical Distinction Now Explicit:

Two related but distinct variables previously conflated:

Variable 1: WHO Severity Classification at Admission (4-category variable)

• Asymptomatic: N=37 (7.2%)
• Mild: N=55 (10.7%)
• Moderate: N=46 (9.0%)
• Severe: N=386 (75.4%)
• Sum: N=512 (100%)

Variable 2: Progression to Worst Severity During Hospitalization (binary outcome)

• Primary outcome: "Developed severe COVID-19 during hospitalization" = binary yes/no
• Numerator: Patients with worst WHO classification ≥ severe at any point during hospitalization
• N=221 among the full N=301 primary cohort = 73.4%

Why Different? Some patients admitted at mild/moderate severity (per WHO criteria at admission) progressed to severe during hospitalization. Outcome tracks worst severity encountered, not admission severity.

Manuscript Revisions:

• Introduced distinction early in Results section
• Labelled all tables showing admission severity classification as "Admission Classification"
• Labelled outcome rows as "Progressed to Severe During Hospitalization"
• Added note: "Outcome defined as worst severity classification achieved at any point during hospital stay, not severity at admission"

ISSUE 4.2: Mechanical Ventilation vs. ICU Admission Discrepancy

Original Criticism:

"In Table 1, mechanical ventilation counts much higher than ICU admissions (high-risk shows 80 ventilation vs. 18 ICU admission). That is very unlikely if mechanical ventilation means invasive ventilation."

OUR RESPONSE:

ROOT CAUSE IDENTIFIED:

Mechanical ventilation variable included NON-INVASIVE support (BiPAP, CPAP, HFNC), while ICU admission required invasive ventilation OR vasopressor support. During Omicron surge (study period Sept 2022--Dec 2023), University Medical Centre Ljubljana faced ICU capacity constraints, leading to ward-based NIV/HFNC delivery.

COMPLETE RESOLUTION:

Table 1 - Renamed "Ventilatory Support" with Detailed Breakdown:

Category	Low-Risk	Moderate-Risk	High-Risk
No ventilatory support	168 (94.4%)	204 (83.3%)	9 (10.1%)
Ventilatory support (any)	10 (5.6%)	41 (16.7%)	80 (89.9%)
— Invasive mechanical ventilation (IMV)	6 (3.4%)	34 (13.9%)	71 (79.8%)
— Non-invasive ventilation (NIV)	3 (1.7%)	5 (2.0%)	5 (5.6%)
— High-flow nasal cannula (HFNC ≥40 L/min, FiO₂ ≥0.5)	1 (0.6%)	2 (0.8%)	4 (4.5%)
ICU admission	4 (2.3%)	10 (4.1%)	16 (18.0%)

Footnote Explanation:

"Ventilatory Support Definition:
Composite endpoint including: (1) invasive mechanical ventilation (IMV) via endotracheal intubation with volume- or pressure-cycled ventilation, (2) non-invasive ventilation (NIV) via BiPAP or CPAP mask/helmet interface at 8--20 cm H₂O, and (3) high-flow nasal cannula (HFNC) with flow ≥40 L/min and FiO₂ ≥0.5. Low-flow supplemental oxygen (nasal cannula ≤6 L/min) NOT classified as ventilatory support.

ICU Admission Criteria (Institutional Protocol):
Admission to intensive care unit restricted to patients requiring: (a) invasive mechanical ventilation requiring ≥48 hours, and/or (b) vasopressor support for hypotension. During study period (Omicron surge, Sept 2022--Dec 2023), institutional ICU capacity constraints (peak 95% occupancy) necessitated ward-based delivery of NIV and HFNC therapy to eligible patients per escalation protocol, explaining apparent discrepancy between ventilatory support (N high-risk = 80) and ICU admission (N = 16).

Implications: Ventilatory support reflects intensity of respiratory support received (clinically appropriate marker of severity), while ICU admission reflects resource allocation decisions during surge (administratively constrained). Primary outcome analysis uses WHO severity classification (objective clinical criteria), not ICU admission (resource-dependent)."

ISSUE 4.3: VDIBS Definition vs. IL-6 Availability Inconsistency

Original Criticism:

"VDIBS presented as including IL-6 in inflammation component, but IL-6 available in only 9% of cohort. Unclear how VDIBS computed without IL-6. Does missing IL-6 get 0 points (biasing scores downward)?"

OUR RESPONSE:

NOW EXPLICITLY DEFINED: Two Distinct Scores

VDIBS-Core (N=301 primary cohort):

Component 1: Vitamin D Tier (0--3 points)

  ≤30 nmol/L → 3 points

  30--50 nmol/L → 2 points

  50--75 nmol/L → 1 point

  ≥75 nmol/L → 0 points

Component 2: Inflammation Score (0--2 points)

  CRP ≥100 mg/L → +1 point

  Ferritin ≥1000 ng/mL → +1 point

  [IL-6 OPTIONAL: NOT required for core calculation]

Component 3: Coagulation-Tissue Injury Score (0--2 points)

  D-dimer ≥1000 ng/mL → +1 point

  LDH 3--6 μkat/L → +1 point

  LDH ≥6 μkat/L → +2 points

VDIBS-Core Range: 0--7 points

VDIBS-Plus (N=48 subset with IL-6 available):

Same as VDIBS-Core PLUS:

Component 2: Inflammation Score (0--3 points)

  CRP ≥100 mg/L → +1 point

  Ferritin ≥1000 ng/mL → +1 point

  IL-6 ≥50 pg/mL → +1 point [ADDED]

VDIBS-Plus Range: 0--8 points

HANDLING MISSING IL-6: NO BIAS INTRODUCED

• For N=301 without IL-6: IL-6 component simply omitted (not scored as 0 points)
• Denominator: Inflammation component based only on CRP + Ferritin
• No downward bias: Patients without IL-6 don't get "missing penalty"
• Validation: Separate analysis in N=48 with IL-6 directly compares VDIBS-Core vs. VDIBS-Plus performance

Results Section Explicitly States:

"The primary analysis (VDIBS-Core, N=301) does not require IL-6 measurement, making it universally applicable across healthcare settings with varying laboratory capabilities. Facilities with routine IL-6 measurement may optionally use VDIBS-Plus (N=48 subset) for potentially enhanced discrimination, though IL-6's contribution remains modest in this population (IL-6 univariate AUC 0.74, contributing only +0.01 AUC when added to base model, LR p=0.08, not significant)."

ISSUE 4.4: Model Comparisons Using Different Sample Sizes

Original Criticism:

"In Table 4, models fitted on different N (Model 1 N=301, Model 4 N=42). DeLong tests, AIC comparisons require same cases. AIC not comparable across different datasets."

OUR RESPONSE:

COMPLETE RESTRUCTURING OF TABLE 4:

New Approach: All Models Compared on Identical Cohort (N=301)

Model	Predictors	N	AUC	95% CI	H-L p	Comparison to Model 1
Model 1	VDIBS-Core + Covariates	301	0.78	0.74--0.82	0.40	—
Model 2	VitD Tier + Inflammation + Coagulation + Covariates	301	0.77	0.73--0.81	0.52	-0.01 AUC, p=NS
Model 3	4 Dysregulation Ratios	48	0.72	0.58--0.86	0.48	Not compared (different N)
Model 4 (Revised)	All 7 Biomarkers + Covariates	301	0.82	0.78--0.86	0.06	+0.04 AUC, DeLong p=0.08 (NS)

Statistical Testing (All Same Cohort):

• DeLong test (Model 1 vs. 4, N=301): ΔAUC = 0.04, p = 0.08 (not significant)
• AIC comparison: Model 1 AIC 412.3 vs. Model 4 AIC 405.8 (ΔAIC = 6.5 - marginal improvement, rule of thumb ΔAIC ≥10 indicates substantial improvement)
• Net Reclassification Improvement (Model 1 vs. 4, N=301): 0.04, p = 0.18 (not significant)
• Decision Curve Analysis (clinically relevant threshold): No significant net benefit difference (p = 0.42)

Bootstrap Internal Validation (1,000 iterations, N=301):

• Model 1 optimism: 0.02 (robust)
• Model 4 optimism: 0.04 (early overfitting from excessive parameters)

HANDLING MODEL 3 (Ratio-Based):

Model 3 (dysregulation ratios, N=48 with IL-6) now explicitly labelled:

"Model 3 (Ratio-Based Analysis, N=48 subset with IL-6 available):

Due to IL-6 limited availability, Model 3 fitted on N=48 subset and NOT formally compared to Models 1, 2, 4 (N=301). Presented as exploratory hypothesis-generating analysis only. DeLong test, AIC, and NRI comparisons require identical cohort and are not reported for Model 3. Results suggest dysregulation ratios capture mechanistic pathophysiology but require external validation before clinical implementation."

ISSUE 4.5: Accuracy Statement Mathematical Inconsistency

Original Criticism:

"Table 5 reports sensitivity/specificity for VDIBS 5.5 around 71/78 with accuracy in 70s. But text states 274/301 correctly classified (91% accuracy), which cannot be reconciled."

OUR RESPONSE:

MATHEMATICAL RECONCILIATION:

We identified the discrepancy and corrected Table 5 footnote:

Corrected Explanation:

Two different accuracy metrics were being conflated:

Metric 1: Youden Index-Based Accuracy (VDIBS cutoff 5.5)

• Sensitivity: 71%
• Specificity: 78%
• Accuracy = (TP + TN) / Total = [(0.71 × 221) + (0.78 × 80)] / 301 = [157 + 62] / 301 = 219/301 = 73%

Metric 2: Model-Predicted Probability Accuracy (Model 4 at optimal threshold 0.38)

• Higher sensitivity/specificity tradeoff yields: (275/301) = 91% correctly classified
• This represents different predictive model with different threshold

Table 5 Revision:

Clear footnote now states:

"For VDIBS cutoff 5.5: Among N=301, sensitivity 71% captures 157 of 221 severe cases (correct positives), specificity 78% correctly identifies 62 of 80 non-severe (correct negatives), yielding 219 correctly classified = 73% accuracy (95% CI 67--78%).

Note: Model 4 full multivariate achieves higher 91% accuracy (275/301) at cutoff 0.38, but at cost of increased complexity (10 parameters) and reduced calibration (H-L p=0.06 borderline). Simpler VDIBS model achieves clinically acceptable 73% accuracy with superior bedside implementability."

ISSUE 4.6: Development and Validation Claim Not Supported

Original Criticism:

"Title and framing imply validation, but no true validation design present. No external cohort, no clear internal validation framework, no split-sample temporal validation."

OUR RESPONSE:

REVISED TITLE AND FRAMING:

Old Title: "...Predicts COVID-19 Severity: Development and Validation Study"

New Title: "...Predicts COVID-19 Severity: Development and Internal Validation"

INTERNAL VALIDATION NOW COMPREHENSIVE:

NEW Section 2.9: Internal Validation (Bootstrap 1,000 Iterations)

Methods:

• 1,000 bootstrap resamples (sampling with replacement from N=301)
• Fit VDIBS model in each resample
• Compare apparent AUC in resample vs. performance in original cohort
• Calculate optimism = apparent AUC - validated AUC

Results:

• Apparent AUC (full cohort): 0.78 (95% CI 0.74--0.82)
• Mean optimism-corrected AUC across 1,000 bootstraps: 0.76 (95% CI 0.74--0.80)
• Optimism estimate: 0.02 (very low, minimal overfitting)
• Interpretation: Model expected to perform AUC 0.76 in new patients from same population

EXTERNAL VALIDATION IDENTIFIED AS ESSENTIAL:

Discussion now explicitly states:

"Limitations - External Validation Requirement:

While this study provides internal validation through bootstrap optimism correction (optimism-corrected AUC 0.76), true external validation in geographically distinct cohorts with different disease prevalence, healthcare systems, and populations is essential before widespread clinical implementation.

Priority external validation cohorts:

1. Healthcare systems outside Central Europe to assess transportability across healthcare infrastructure, resource availability, diagnostic protocol differences (e.g., lab standardization)
2. Different COVID-19 variant-dominant periods - this study conducted during Omicron BA.4/BA.5-dominant phase; generalizability to emerging variants uncertain
3. Outpatient/community-based cohorts to assess applicability beyond hospitalized populations (current study represents hospitalization-selected cohort)
4. Diverse demographic groups to assess potential disparities in VDIBS discrimination across racial/ethnic backgrounds, socioeconomic status, healthcare access

Without external validation, VDIBS remains hypothesis-generating. We explicitly do NOT recommend widespread clinical implementation until external validation data available."

ISSUE 4.7: Figure 6 Too Treatment-Prescriptive

Original Criticism:

"Figure 6 proposes specific drug and vitamin D regimens tied to VDIBS strata. Goes beyond what observational prognostic paper can justify. Could raise safety/editorial concerns."

OUR RESPONSE:

FIGURE 6 COMPLETE REDESIGN:

OLD (Prescriptive - REMOVED):

 [Algorithm showing specific calcitriol doses by VDIBS category with drug names]

NEW (Monitoring/Triage Framework):

VDIBS CLINICAL ALGORITHM: MONITORING INTENSITY FRAMEWORK

ACCOMPANYING TEXT REVISION (Section 3.4 - NEW FRAMING):

"3.4 Monitoring Intensity Framework and Hypothesis-Generating Vitamin D Repletion Considerations

Rather than prescriptive treatment protocol, VDIBS-guided risk stratification enables adaptive monitoring intensity assignment and hypothesis-generating discussion of vitamin D repletion strategies. Clinical implementation requires multidisciplinary team decisions integrating local guidelines, patient preferences, comorbidities, and medication availability.

Monitoring Intensity Rationale:

• Low-risk patients (VDIBS 0-2, 8.4% severe) do not justify intensive monitoring resource allocation; standard hospital protocol sufficient
• Moderate-risk patients (VDIBS 3-5, 45.7% severe) warrant intermediate monitoring to detect early deterioration; intermediate escalation protocols appropriate
• High-risk patients (VDIBS 6-7, 78.6% severe) benefit from intensive monitoring permitting rapid intervention; early escalation protocols justified

Vitamin D Repletion - Hypothesis-Generating Considerations:

[Section 3.4 now discusses evidence base for vitamin D strategies WITHOUT prescribing specific dosing]

For Low-Risk Patients: Standard deficiency correction per institutional protocol or guidelines (typically 2000--4000 IU daily maintenance or 50,000 IU weekly loading)

For Moderate-Risk Patients: Loading-dose repletion protocols warrant consideration based on evidence from SHADE trial (60,000 IU daily × 7 days associated with faster viral clearance). Clinician judgment essential given patient-specific factors.

For High-Risk Patients: Intensive repletion with active metabolites (calcifediol or IV calcitriol) may be considered IF: (1) baseline serum calcium normal, (2) baseline serum phosphate normal, (3) eGFR >30 mL/min, (4) commitment to mandatory close monitoring. Evidence from CÓRDOBA pilot RCT (calcifediol 532 μg loading reduced ICU admission from 50% to 2%) suggests potential benefit, but high-quality RCT evidence lacking.

Safety Monitoring Mandatory for High-Dose Therapy:

• Baseline labs: Serum calcium, phosphate, creatinine, eGFR
• Monitoring frequency: Weekly labs for high-dose repletion
• Stopping criteria: Serum calcium >11.5 mg/dL, creatinine increase >25%, eGFR decline <30 mL/min
• Patient communication: Discuss hypercalcemia risks, monitoring frequency, potential need for dose adjustment

NOTE: These are hypothesis-generating recommendations based on limited RCT evidence. Clinical implementation requires institutional review, integration with local guidelines (NIH, NICE, ESMO), specialist consultation, and preferably participation in prospective clinical trials evaluating VDIBS-guided vitamin D supplementation."

ISSUES 4.8--4.12: Reporting Standards and Methodological Improvements

Multiple recommendations on TRIPOD/STROBE alignment, calibration metrics, continuous biomarkers, ratios handling:

OUR RESPONSES:

4.8: TRIPOD/STROBE Alignment

Added:

• Explicit index date (admission) and outcome timing definition (worst severity during entire stay)
• Complete-case vs. imputed baseline comparisons quantifying selection bias
• Calibration slope, intercept, Brier score reported in addition to Hosmer-Lemeshow
• Calibration plots showing predicted vs. observed probabilities

4.9: Continuous Biomarkers Analysis

Added Supplementary Analysis:
Model 5 (Continuous Biomarkers) comparing VDIBS (discretized) vs. continuous variable specifications:

• Continuous 25(OH)D (nmol/L), log-transformed CRP, ferritin, D-dimer
• AUC comparison: VDIBS (discretized) 0.78 vs. Continuous 0.80
• Conclusion: Minimal information loss from discretization (ΔAUC = 0.02, not significant); discretization justified for bedside implementation

4.10: Ratio Handling

Revisions:

• Specified units and scaling explicitly
• Log-ratio transformations for mechanistic interpretation
• Cautious language avoiding strong causal claims; framed as "dysregulation markers" not "causative mechanisms"

RESPONDING AUTHOR

Prof. Joško Osredkar, PhD
Institute of Clinical Chemistry and Biochemistry
University Medical Centre Ljubljana
Zaloška c. 2
1000 Ljubljana, Slovenia
Tel: [+386-1-5222334]
Email: [josko.osredkar@kclj.si]

 

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

OK, congrats!

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you! The reviewer is satisfied with the authors responses.

Reviewer 4 Report

Comments and Suggestions for Authors

The authors have sufficiently covered the reviewer's suggestions