The Prevalence and Correlates of Vitamin D Deficiency and Overweight/Obesity of School-Age Children in Colombia–Findings on the Double Burden of Malnutrition from Nationally-Representative Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The study is well-motivated, uses high-quality data, and applies appropriate general statistical methods. However, some methodological and statistical refinements are necessary to strengthen the robustness of the findings. The main risks are overinterpretation of rare-event associations and insufficient clarity about survey design adjustments. Addressing these issues would substantially improve the credibility of the results.

 Methodological Concerns

1. Definition of Overweight/Obesity
   The cut-off (BMI-for-age z-score > +1) groups together overweight and obese children. This may obscure differential associations. Separate models for overweight and obesity could provide clearer insights.
2. Vitamin D Cut-offs
   While three thresholds were used, no justification was provided for treating them as equally valid. A discussion of clinical versus epidemiological relevance of each threshold would strengthen interpretation.
3. Cross-Sectional Design
   Causality cannot be inferred from associations. The manuscript should explicitly state this limitation, especially when discussing behavioral factors (e.g., screen time, physical activity).
4. Potential Residual Confounding
   Some relevant covariates, such as dietary intake of vitamin D, supplement use, and seasonality of blood collection, were not included. These factors could confound observed associations.

Statistical Concerns

1. Complex Survey Design
   The use of the survey package in R is appropriate. However, the manuscript does not clarify whether clustering and stratification were fully accounted for in variance estimation. Details on PSU, strata, and weighting adjustments should be added.
2. Multiple Testing and Type I Error
   Numerous chi-square tests and logistic regression models were performed without correction for multiple comparisons. The risk of false-positive findings is considerable. Authors should consider methods such as Bonferroni correction, false discovery rate, or emphasize effect sizes and confidence intervals over p-values.
3. Sparse Data Bias
   Some outcomes (e.g., DBM1) had very low prevalence (<1%). Logistic regression estimates for these rare events may be unstable, as reflected in wide confidence intervals. Alternative methods (e.g., penalized regression, Firth’s correction) could improve reliability.
4. Model Specification

All covariates were entered simultaneously into adjusted models, which may introduce multicollinearity. Diagnostics (e.g., VIFs) are not reported.

The rationale for including certain variables (e.g., playground availability) is not fully explained—were these pre-specified or exploratory?

Some adjusted odds ratios changed drastically compared to crude estimates, suggesting possible overfitting or collinearity.

1. Interpretation of Odds Ratios
   In some cases, very large odds ratios with wide CIs are presented (e.g., OR > 9 for physical activity in DBM1). Authors should interpret these cautiously and avoid overstating the strength of associations.
2. Presentation of Results

Prevalence estimates are clearly reported, but confidence intervals for regression models are sometimes extremely wide, indicating low precision.

Tables are dense; summarizing key findings (e.g., highlighting only significant predictors) in the main text would improve readability.

 

Author Response

Methodological Concerns

Comment 1: Definition of Overweight/Obesity

The cut-off (BMI-for-age z-score > +1) groups together overweight and obese children. This may obscure differential associations. Separate models for overweight and obesity could provide clearer insights.

Response:

We appreciate the reviewer’s insightful comment regarding the definition of overweight and obesity. In response, we conducted additional analyses to distinguish between children classified as overweight and those classified as obese when examining the double burden of malnutrition (DBM3). The differences in prevalence between these two groups are presented in the Supplementary Materials (Tables S3 and S4).

For DBM1 and DBM2, we did not perform separate analyses due to their very low prevalence, which limited the statistical power to produce reliable estimates. Nevertheless, we calculated adjusted odds ratios (ORs, 95% CIs) using Firth’s penalized logistic regression for individual-, interpersonal-, and community-level factors across alternative definitions of the double burden of malnutrition (DBM1, DBM2, DBM3). These models were estimated separately for children with overweight, those with obesity, and for the combined group (overweight and obesity). The corresponding results are presented in the Supplementary Materials (Tables S6, S8, and S9).

Comment 2: Vitamin D Cut-offs

While three thresholds were used, no justification was provided for treating them as equally valid. A discussion of clinical versus epidemiological relevance of each threshold would strengthen interpretation.

Response

We appreciate the reviewer’s thoughtful comment regarding the justification for the three vitamin D cut-offs. In response, we have expanded the explanation in the Introduction section (lines 85–95) to clarify the clinical and epidemiological rationale for including multiple thresholds. The new paragraph reads as follows:

“The global definition of vitamin D status is complicated by persistent disagreement on clinical thresholds and variability in 25-hydroxyvitamin D [25(OH)D] assays [36]. While international initiatives such as the Vitamin D Standardization Program (VDSP) have substantially improved the analytical accuracy of automated immunoassays—including platforms such as the Siemens ADVIA [37]—significant clinical debate persists over the appropriate cut-off points [36,38]. Generally, the scientific community recognizes two categories: deficiency and insufficiency. The most conservative threshold, proposed by the Institute of Medicine (IOM), defines deficiency as serum 25(OH)D ≤30 nmol/L (≤12 ng/mL), which is sufficient only to prevent skeletal diseases [39]. In contrast, the Endocrine Society defines deficiency more stringently as <50 nmol/L (<20 ng/mL) and insufficiency as 50–75 nmol/L (20–30 ng/mL), reflecting broader endocrine and metabolic benefits [38].”

In addition, we maintained the explanation in the Methods section (lines 157–160) that explicitly describes the three thresholds used in the analysis:

“Due to the lack of consensus on the cutoff point for defining vitamin D deficiency [36,47], we proposed three thresholds: <30 nmol/L [6,40], <37.5 nmol/L [48,49], and <50 nmol/L [6,40]. Among these, the threshold of <50 nmol/L is the most widely accepted and consistently supported across numerous studies.”

This revision provides a clearer justification for treating the three thresholds as complementary and context-dependent definitions, reflecting both clinical practice and epidemiological research perspectives.

Comment 3: Cross-Sectional Design

Causality cannot be inferred from associations. The manuscript should explicitly state this limitation, especially when discussing behavioral factors (e.g., screen time, physical activity).

Response

We appreciate the reviewer’s comment regarding the limitations of the cross-sectional design. In response, we have carefully revised the manuscript to remove any language suggesting causal relationships. The study design is explicitly stated as cross-sectional in the Limitations section, emphasizing that causal inferences cannot be drawn. Additionally, we have ensured that the description of results—particularly for behavioral factors such as screen time and physical activity—avoids any implication of causality.

Comment 4: Potential Residual Confounding

Some relevant covariates, such as dietary intake of vitamin D, supplement use, and seasonality of blood collection, were not included. These factors could confound observed associations.

Response

We appreciate the reviewer’s comment regarding potential residual confounding. In response, we have explicitly addressed this in section 4.3 (Strengths and Limitations) of the Discussion (lines 702–707), where we now state:

“Potential residual confounding may persist due to unmeasured determinants of serum 25(OH)D, including dietary vitamin D intake, supplement use, seasonal or monthly variation in sampling, latitude- and altitude-related differences in ultraviolet exposure, air pollution, and skin pigmentation. These factors could contribute to variability in vitamin D status and partially explain the observed patterns of DBM.”

 

This clarification ensures that readers are aware that, despite the strengths of our study, unmeasured covariates may have influenced the observed associations, reinforcing transparency and the interpretation of our findings.

Statistical Concerns

Comment 5: Multiple Testing and Type I Error

Numerous chi-square tests and logistic regression models were performed without correction for multiple comparisons. The risk of false-positive findings is considerable. Authors should consider methods such as Bonferroni correction, false discovery rate, or emphasize effect sizes and confidence intervals over p-values.

Response

We thank the reviewer for highlighting the potential risk of false-positive findings due to multiple testing. To address this, p-values were adjusted using the False Discovery Rate (FDR) method. This procedure is described in the Methods section (lines 233–236):

“To address the risk of false-positive findings due to multiple comparisons, p-values were adjusted using the False Discovery Rate (FDR) method [52]. Both the raw and FDR-adjusted p-values are reported for each variable.”

In addition, only odds ratios (ORs) whose 95% confidence intervals did not cross 1 were considered statistically significant and interpreted as indicative of an association. This approach emphasizes effect sizes and confidence intervals alongside p-values, reducing the likelihood of Type I errors.

Statistical Concerns

Comment 6: Complex Survey Design

The use of the survey package in R is appropriate. However, the manuscript does not clarify whether clustering and stratification were fully accounted for in variance estimation. Details on PSU, strata, and weighting adjustments should be added.

Response

We thank the reviewer for highlighting the need for clarification regarding the complex survey design. In response, we have provided a detailed description of the survey characteristics in the Materials and Methods section (2.1 Study Population and Data, lines 111–121):

“We used data from the Colombian National Nutrition Survey 2015–2016 [Encuesta Nacional de Situación Nutricional (ENSIN)] [41], a nationally representative survey covering 99% of the Colombian population through a multi-stage stratified sampling design. Non-institutionalized civilian residents were selected to represent the Colombian population. The survey included 44,202 households, organized into 177 strata. The sampling procedure followed the official ENSIN hierarchical design, with 238 primary sampling units (PSUs) comprising groupings of municipalities across all 32 departments and Bogotá. Within these, secondary sampling units (SSUs) were formed by groups of contiguous city blocks within the same sector and census section, each including at least 96 households. Finally, tertiary sampling units (TSUs) consisted of 5,000 segments averaging 12 contiguous households (range 6–17), of which 4,962 were effectively selected and 4,813 contained at least one completed household interview.”

Additionally, in section 2.2 (Statistical Analysis), we describe how these complex survey components—including strata, PSUs, and sampling weights—were incorporated into variance estimation using the survey package in R. This ensures that clustering, stratification, and weighting were fully accounted for in all analyses.

Comment 6: Multiple Testing and Type I Error

Numerous chi-square tests and logistic regression models were performed without correction for multiple comparisons. The risk of false-positive findings is considerable. Authors should consider methods such as Bonferroni correction, false discovery rate, or emphasize effect sizes and confidence intervals over p-values.

Response

We thank the reviewer for highlighting the risk of false-positive findings due to multiple testing. To address this concern, p-values were adjusted using the False Discovery Rate (FDR) method. This procedure is described in the Materials and Methods section, 2.2 Statistical Analysis (lines 233–236):

“To address the risk of false-positive findings due to multiple comparisons, p-values were adjusted using the False Discovery Rate (FDR) method [52]. Both the raw and FDR-adjusted p-values are reported for each variable.”

In addition, only odds ratios (ORs) whose 95% confidence intervals did not include 1 were considered statistically significant and interpreted as indicative of an association. This approach emphasizes effect sizes and confidence intervals alongside p-values, reducing the likelihood of Type I errors and providing more robust evidence for the reported associations.

Comment 7: Sparse Data Bias

Some outcomes (e.g., DBM1) had very low prevalence (<1%). Logistic regression estimates for these rare events may be unstable, as reflected in wide confidence intervals. Alternative methods (e.g., penalized regression, Firth’s correction) could improve reliability.

Response

We thank the reviewer for highlighting the potential issue of sparse data bias, particularly for outcomes with low prevalence. To improve the stability of logistic regression estimates, we implemented multivariable models using penalized regression with Firth’s correction across all DBM outcomes, including DBM3, which had higher prevalence but remained below 10%. This approach is described in the Materials and Methods section (lines 253–262):

“Survey-weighted logistic regression models were then fitted to estimate both crude and adjusted associations between DBM outcomes and explanatory variables. Variables showing a bivariate association with p < 0.20 and not exhibiting problematic multicollinearity (VIF > 5) were considered eligible for inclusion in multivariable models. Multivariable survey-weighted logistic regression models were fitted using penalized regression with Firth’s correction [56] to improve estimation stability across all DBM outcomes. This approach was applied to DBM1, DBM2, and DBM3, with stratified analyses conducted for each outcome by considering overweight and obesity both separately and in combination. Effect estimates were evaluated in terms of their magnitude and precision, with particular attention to confidence intervals rather than relying solely on p-values.”

Applying this method consistently ensures more reliable and precise estimates for all outcomes, accounting for low event prevalence while maintaining comparability across analyses.

Model Specification

Comment 8:

All covariates were entered simultaneously into adjusted models, which may introduce multicollinearity. Diagnostics (e.g., VIFs) are not reported.

Response

We thank the reviewer for raising the concern regarding potential multicollinearity in our adjusted models. To address this, we implemented a structured covariate selection and diagnostic approach, as described in the Materials and Methods section, 2.2 Statistical Analysis (lines 237–252):

“Covariate selection for multivariable logistic regression followed a two-step process. First, each candidate variable was assessed through survey-weighted logistic regression, accounting for clustering, stratification, and sampling weights. For categorical variables with multiple levels, the minimum p-value across levels (excluding the intercept) was used. Variables with a bivariate association of p < 0.20, based on survey-weighted Wald tests, were considered for inclusion. Additionally, age and sex were retained in all models regardless of their statistical significance, as they are potential confounders and key demographic factors commonly associated with child nutritional status. This approach combines empirical selection criteria with theoretical rationale, enhancing model stability and ensuring epidemiological relevance [53].

To assess potential multicollinearity among independent variables, categorical variables were recoded as factors and dummy variables were created for all factor levels. Variance Inflation Factors (VIFs) were then calculated using a weighted linear model that accounted for survey design weights. Variables with VIF values above 5 were considered indicative of problematic multicollinearity, following common epidemiological guidelines [54,55].”

This strategy ensures that multivariable models are both statistically robust and epidemiologically meaningful, minimizing the risk of biased estimates due to collinearity among covariates.

Comment 9:

The rationale for including certain variables (e.g., playground availability) is not fully explained—were these pre-specified or exploratory?

Response

We thank the reviewer for this comment regarding the rationale for including certain community-level variables, such as playground availability. The inclusion of these variables was pre-specified based on their potential relevance to vitamin D exposure. This is described in the Materials and Methods section (lines 208–217):

“Finally, at the community level, we examined the region of residence of schoolchildren, including Atlantico, Oriental, Orinoquia, Amazonia, Bogota, Central, and Pacifico. We also considered urbanicity (rural vs. urban) and degree of urbanization (population size). Additionally, we included whether children lived in neighborhoods with playgrounds or other recreational spaces. Access to safe and well-equipped recreational spaces influences the amount of time children spend outdoors, which directly affects their exposure to ultraviolet B (UVB) radiation—the main source for vitamin D synthesis. Proximity to parks or playgrounds can facilitate sustained sun exposure necessary for adequate 25(OH)D production, and differences in the availability or quality of these environments may help explain variations in vitamin D status among children [51].”

This explanation clarifies that playground availability and other community-level variables were selected based on theoretical and epidemiological considerations, rather than being exploratory.

Comment 10:

Some adjusted odds ratios changed drastically compared to crude estimates, suggesting possible overfitting or collinearity.

Response

We thank the reviewer for raising the concern regarding drastic changes between crude and adjusted odds ratios, which could reflect potential overfitting or collinearity. To address this, we applied a consistent variable selection strategy: only variables with a bivariate association of p < 0.20 (based on survey-weighted Wald tests) and without evidence of problematic multicollinearity (VIF ≤ 5) were considered for inclusion in multivariable models.

Additionally, as described in the Materials and Methods section (lines 253–262):

“Survey-weighted logistic regression models were then fitted to estimate both crude and adjusted associations between DBM outcomes and explanatory variables. Multivariable survey-weighted logistic regression models were fitted using penalized regression with Firth’s correction [56] to improve estimation stability across all DBM outcomes. This approach was applied to DBM1, DBM2, and DBM3, with stratified analyses conducted for each outcome by considering overweight and obesity both separately and in combination. Effect estimates were evaluated in terms of their magnitude and precision, with particular attention to confidence intervals rather than relying solely on p-values.”

By combining systematic variable selection, assessment of multicollinearity, and the use of penalized regression with Firth’s correction, we minimized the risk of overfitting and ensured that adjusted estimates are reliable and interpretable.

Interpretation of Odds Ratios
Comment 11:

In some cases, very large odds ratios with wide CIs are presented (e.g., OR > 9 for physical activity in DBM1). Authors should interpret these cautiously and avoid overstating the strength of associations.

Response

We thank the reviewer for highlighting the importance of cautious interpretation of large odds ratios (ORs) with wide confidence intervals (CIs). In the revised manuscript, we have explicitly addressed this issue throughout the Results and Discussion sections. For instance:

• We now note that some subgroup analyses, particularly in DBM1 and DBM3 (e.g., physical inactivity in the obesity-only subgroup, OR = 11.73, 95% CI: 1.52–1507.37), are based on small numbers of cases, resulting in wide CIs and limited precision.
• We clarify that these ORs should be interpreted as indicative of potential associations rather than definitive causal effects.
• The text now emphasizes the cross-sectional design and the exploratory nature of these analyses, highlighting that observed associations may reflect behavioral and contextual correlations rather than direct causal relationships.
• We have also included language cautioning the reader about overinterpreting large ORs and suggesting that future longitudinal or intervention studies are needed to confirm these findings.

These revisions ensure that the manuscript conveys appropriate caution in interpreting ORs while still presenting the observed associations transparently.

Presentation of Results

Comment 12:

Prevalence estimates are clearly reported, but confidence intervals for regression models are sometimes extremely wide, indicating low precision.

Response

We thank the reviewer for this observation. We acknowledge that some regression estimates, particularly for rare DBM categories or subgroup analyses (e.g., obesity-only with physical inactivity), have wide confidence intervals, reflecting limited precision due to small numbers of cases.

In response, we have revised the manuscript to:

• Explicitly highlight instances where ORs have wide CIs and caution readers that these estimates should be interpreted carefully.
• Emphasize the exploratory nature of these analyses and that associations may reflect potential correlations rather than definitive causal effects.
• Note that sparse data may contribute to unstable estimates and that Firth’s correction or penalized approaches could improve reliability in future studies.
• Maintain transparent reporting of the ORs and CIs, while guiding interpretation toward contextual and behavioral relevance rather than overstatement of effect size.

These changes ensure that readers are aware of the limitations in precision and interpret regression results appropriately.

Comment 13:

Tables are dense; summarizing key findings (e.g., highlighting only significant predictors) in the main text would improve readability.

Response

We thank the reviewer for this suggestion. To improve clarity and readability, we divided the tables related to DBM3 (Tables 3–8), including both prevalence estimates and regression models (adjusted and unadjusted) at the individual, interpersonal, and community levels. For each table, we provide a concise description in the main text, highlighting the most relevant findings. Therefore, these tables remain in the main text to ensure transparency and facilitate interpretation of the key results.

For DBM1 and DBM2, the tables related to prevalence and regression analyses (Tables S1, S2, S5–S8), as well as sensitivity analyses for DBM3 (Tables S3, S4, S9), are included in the Supplementary Material. Descriptive summaries of these supplementary tables are fully described in the main text to guide the reader through the most important patterns and findings.

This approach ensures that the manuscript remains readable, emphasizes significant predictors, and maintains complete reporting of all analyses in either the main text or supplementary material.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript entitled “The Prevalence and Correlates of Vitamin D Deficiency and Overweight/Obesity of School-Age Children in Colombia – Findings on the Double Burden of Malnutrition from Nationally-Representative Data” is primarily aimed at identifying socioeconomic factors associated with the double burden (both vitamin D deficiency and obesity) in Colombian children aged 5-12 years. The authors use a nationally representative dataset (ENSIN 2015). The topic of the article is important but raises the question of why authors considered only vitamin D deficiency as a burden of malnutrition, even though it is known to be low in obese people. It would be interesting to review the status of the other micronutrients as well.

The manuscript is well structured, but some things need to be clarified and improved.

Although it is clear in the main text what the abbreviations DBM1-3 stand for, the summary itself is confusing. I recommend reformulating this section a little.

Results - The question remains as to the vitamin D status of the 73.9% of subjects with normal body weight. Is obesity one of the most important factors for a deficiency or can a correlation perhaps be established with another factor?

- In Table 2, for which of the 2011 respondents do these data apply, since according to the first three rows of the table there should be 79 undernourished, 4565 adequately nourished and 1419 overweight/obese respondents? And for which of these respondents do we have a distribution on DBM1-3, since there is data for 609 respondents?

- The tables are comprehensive but lengthy. Consider moving some detailed tables or parts of them to supplementary material to improve readability.

- The decision to categorise DBM into three categories is understandable given the lack of consensus on the thresholds. However, this approach makes interpretation somewhat difficult, especially as DBM1 and DBM2 are based on very rare events and lead to unstable estimates. The manuscript would perhaps benefit from a greater emphasis on DBM3 (<50 nmol/L) as the most widely accepted threshold for public health purposes, while DBM1 and DBM2 are presented in a complementary analysis

Conclusion of the paper is very general.

The manuscript requires a clearer interpretation of the results and a refinement of the presentation. Some of the tables (at least parts of them) are superfluous and detract from the text of the paper itself.

Author Response

Comments and Suggestions for Authors

The manuscript entitled “The Prevalence and Correlates of Vitamin D Deficiency and Overweight/Obesity of School-Age Children in Colombia – Findings on the Double Burden of Malnutrition from Nationally-Representative Data” is primarily aimed at identifying socioeconomic factors associated with the double burden (both vitamin D deficiency and obesity) in Colombian children aged 5-12 years. The authors use a nationally representative dataset (ENSIN 2015). The topic of the article is important but raises the question of why authors considered only vitamin D deficiency as a burden of malnutrition, even though it is known to be low in obese people. It would be interesting to review the status of the other micronutrients as well.

The manuscript is well structured, but some things need to be clarified and improved.

Comment 1:

Although it is clear in the main text what the abbreviations DBM1-3 stand for, the summary itself is confusing. I recommend reformulating this section a little.

Response

We thank the reviewer for this suggestion. Although the abbreviations DBM1–3 were clearly defined in the main text, we agree that the previous abstract was somewhat confusing in summarizing these definitions. In response, we have reformulated and reorganized the abstract to improve clarity and readability.

The revised abstract now:

• Clearly introduces the concept of the double burden of malnutrition (DBM) and its public health relevance.
• Explicitly defines DBM1, DBM2, and DBM3 based on concurrent overweight/obesity and vitamin D deficiency thresholds (<30, <37.5, and <50 nmol/L).
• Summarizes prevalence estimates and key findings separately for each DBM definition, highlighting differences in behavioral, household, and community-level associations.
• Clarifies how stricter versus broader DBM thresholds capture different aspects of vulnerability, from behavioral/household factors to structural/contextual conditions.

These revisions ensure that the abstract is concise, readable, and conveys the study’s objectives, methods, and main findings clearly, directly addressing the reviewer’s concern about the previously confusing abstract.

Comment 2:

Results - The question remains as to the vitamin D status of the 73.9% of subjects with normal body weight. Is obesity one of the most important factors for a deficiency or can a correlation perhaps be established with another factor?

Response

We thank the reviewer for raising this point. Our study specifically focused on the double burden of malnutrition (DBM), defined as the co-occurrence of overweight/obesity and vitamin D deficiency, and did not assess vitamin D deficiency among children with normal body weight. Within this DBM framework, overweight and obesity were considered key characteristics of the population with low vitamin D, reflecting patterns observed in previous research (e.g., lower circulating 25(OH)D in children with excess adiposity). Other behavioral and contextual factors—such as physical activity, household composition, and geographic region—were also examined for associations with DBM. Our analyses therefore describe associations among children with overweight/obesity, rather than factors associated with vitamin D deficiency in the general population.

Comment 3:

- In Table 2, for which of the 2011 respondents do these data apply, since according to the first three rows of the table there should be 79 undernourished, 4565 adequately nourished and 1419 overweight/obese respondents? And for which of these respondents do we have a distribution on DBM1-3, since there is data for 609 respondents?

Response

We thank the reviewer for raising this point. For Table 2, the prevalence of DBM (DBM1–3) was calculated among children with complete anthropometric and vitamin D data, and the resulting weighted prevalence is reported relative to the total study population (n = 6,063). Specifically:

• DBM1 (BMI-for-age z-score >1 and vitamin D <30 nmol/L) had 40 cases.
• DBM2 (BMI-for-age z-score >1 and vitamin D <37.5 nmol/L) had 188 cases.
• DBM3 (BMI-for-age z-score >1 and vitamin D <50 nmol/L) had 381 cases.

These counts are smaller than the total number of children classified as overweight or obese (n = 1,419) because not all children with overweight or obesity met the vitamin D deficiency criterion.

We have added clarification in the Methods (lines 219–229), as follows:

"We estimated the weighted prevalence of the double burden of malnutrition (DBM) among Colombian schoolchildren, defined as a BMI-for-age z-score >1 (over-weight/obesity) combined with vitamin D deficiency, using three thresholds: <30 nmol/L (DBM1), <37.5 nmol/L (DBM2), and <50 nmol/L (DBM3). Weighted prevalence was calculated as the proportion of children meeting both criteria among those with complete anthropometric and vitamin D data, relative to the total study population (n = 6,063). For DBM3, we additionally calculated prevalence separately for children classified as overweight only and for those classified as obese only; for DBM1 and DBM2, the number of cases was too small to allow this stratification. All analyses accounted for the complex survey design—including primary sampling units (PSUs), strata, and sampling weights—to produce nationally representative estimates."

This clarification should make it clear that the prevalence estimates for DBM1–3 are based on children who meet both conditions, and that the counts differ from the total overweight/obese population due to the additional requirement of vitamin D deficiency.

Comment 4:

- The tables are comprehensive but lengthy. Consider moving some detailed tables or parts of them to supplementary material to improve readability.

Response

We thank the reviewer for this suggestion. To enhance clarity and readability, we organized the tables related to DBM3 (Tables 3–8) to include both prevalence estimates and regression models (adjusted and unadjusted) at the individual, interpersonal, and community levels. Each table is accompanied by a concise description in the main text, emphasizing the most relevant findings, which is why these tables remain in the main text to support transparency and interpretation of key results.

For DBM1 and DBM2, all tables related to prevalence and regression analyses (Tables S1, S2, S5–S8), as well as sensitivity analyses for DBM3 (Tables S3, S4, S9), are provided in the Supplementary Material. The main text includes descriptive summaries of these supplementary tables to highlight important patterns and findings for the reader.

This structure balances readability with comprehensive reporting, ensuring that significant predictors are emphasized while maintaining access to complete analytical details in either the main text or supplementary material.

Comment 5:

- The decision to categorise DBM into three categories is understandable given the lack of consensus on the thresholds. However, this approach makes interpretation somewhat difficult, especially as DBM1 and DBM2 are based on very rare events and lead to unstable estimates. The manuscript would perhaps benefit from a greater emphasis on DBM3 (<50 nmol/L) as the most widely accepted threshold for public health purposes, while DBM1 and DBM2 are presented in a complementary analysis

Response

We thank the reviewer for this comment. Our study aimed to evaluate DBM1, DBM2, and DBM3; however, we gave greater emphasis to DBM3, as it represents the most widely accepted public health threshold. Accordingly, all tables and results related to DBM3 are presented in the main text, except for sensitivity analyses, which are provided in the Supplementary Material. For DBM1 and DBM2, we describe the main findings in the text, but all corresponding tables—including prevalence, regression analyses, and sensitivity analyses—are included in the Supplementary Material. This approach allows us to maintain comprehensive reporting of all three DBM thresholds while highlighting DBM3 for clarity and relevance to public health interpretation.

Comment 6:

Conclusion of the paper is very general.

Response

We thank the reviewer for this suggestion. Although DBM prevalence is relatively low, our findings highlight distinct patterns at individual, household, and community levels. DBM3 (<50 nmol/L) reflects the most policy-relevant threshold, while stricter thresholds (DBM1 and DBM2) capture severe deficiency and specific behavioral or household characteristics. Overall, the results underscore the need for monitoring and interventions addressing both overweight/obesity and vitamin D deficiency in Colombian schoolchildren.

Comment 7:

The manuscript requires a clearer interpretation of the results and a refinement of the presentation. Some of the tables (at least parts of them) are superfluous and detract from the text of the paper itself.

Response

We thank the reviewer for this suggestion. To improve clarity and readability, we prioritized DBM3 results in the main text, including prevalence estimates and regression models (adjusted and unadjusted) at the individual, household, and community levels. All related tables for DBM1 and DBM2, as well as sensitivity analyses for DBM3, were moved to the Supplementary Material. Descriptive summaries of these supplementary tables are fully described in the main text to guide the reader through the most relevant patterns and findings. This approach ensures that the manuscript emphasizes key results while maintaining transparency and completeness of reporting. Additionally, we refined the presentation of results to provide a clearer interpretation of patterns and factors associated with DBM across all thresholds, highlighting distinctions between DBM1, DBM2, and DBM3.

Reviewer 3 Report

Comments and Suggestions for Authors

Manuscript: The Prevalence and Correlates of Vitamin D Deficiency and Overweight/Obesity of School-Age Children in Colombia Findings on the Double Burden of Malnutrition from Nationally-Representative Data

Novelty and significance: This work provides new, policy-relevant evidence by delivering the first nationally representative estimates of DBM in Colombian schoolchildren using laboratory-standardized 25(OH)D and multiple clinical thresholds, while situating DBM within urban/wealth gradients, an angle rarely examined in Latin America.

Thank you for this valuable contribution; however, I recommend minor revision to strengthen coherence and internal validity. Specifically:

(a) the abstract mentions “multilevel analyses,” yet the Methods describe survey-weighted logistic regression; please correct the terminology or implement genuine hierarchical models with random effects;

(b) there are table inconsistencies (e.g., identical figures between DBM2 and DBM3 and zero standard errors for Bogotá) that require verification and rerunning;

(c) DBM1 shows sparse events, so consider penalized (Firth) or exact logistic regression, reduce covariates, and report events-per-variable and stability (e.g., bootstrap);

(d) fully specify the survey design used in the software (strata, PSU, weights, and any finite population correction) to ensure reproducibility;

(e) incorporate key determinants of 25(OH)D, season/month of sampling, latitude/altitude or UV/pollution proxies, diet/supplements, and skin pigmentation or acknowledge them explicitly as limitations;

(f) standardize terminology for 25(OH)D (deficiency vs. insufficiency) and add sensitivity analyses using obesity (z>2) and robust Poisson models for prevalence ratios; and

(g) temper causal language (e.g., “determinants”) given the cross-sectional design. Addressing these points will substantially improve clarity, consistency, and policy relevance.

Author Response

Comments and Suggestions for Authors

Manuscript: The Prevalence and Correlates of Vitamin D Deficiency and Overweight/Obesity of School-Age Children in Colombia Findings on the Double Burden of Malnutrition from Nationally-Representative Data

Novelty and significance: This work provides new, policy-relevant evidence by delivering the first nationally representative estimates of DBM in Colombian schoolchildren using laboratory-standardized 25(OH)D and multiple clinical thresholds, while situating DBM within urban/wealth gradients, an angle rarely examined in Latin America.

Thank you for this valuable contribution; however, I recommend minor revision to strengthen coherence and internal validity. Specifically:

Comment 1:

the abstract mentions “multilevel analyses,” yet the Methods describe survey-weighted logistic regression; please correct the terminology or implement genuine hierarchical models with random effects;

Response

We thank the reviewer for this observation. We have revised the abstract to remove the term “multilevel analyses,” as the methods actually involved survey-weighted logistic regression rather than hierarchical models with random effects. The text now accurately reflects the statistical approach used, ensuring consistency between the abstract and Methods section.

Comment 2:

there are table inconsistencies (e.g., identical figures between DBM2 and DBM3 and zero standard errors for Bogotá) that require verification and rerunning;

Response:

We thank the reviewer for highlighting apparent inconsistencies in the tables—such as identical values for DBM2 and DBM3 and zero standard errors for Bogota. Following a detailed verification, we carefully reviewed the data and reran all analyses to ensure the results are accurate, reliable, and robust.

Comment 3:

DBM1 shows sparse events, so consider penalized (Firth) or exact logistic regression, reduce covariates, and report events-per-variable and stability (e.g., bootstrap);

Response

We thank the reviewer for highlighting the potential issue of sparse data bias, particularly for outcomes with low prevalence. To address this, we implemented multivariable models using penalized logistic regression with Firth’s correction for all DBM outcomes, including DBM3, which—despite its higher prevalence—remained below 10%. This method improves the stability and reliability of coefficient estimates when events are sparse.

As detailed in the Materials and Methods (lines 253–262):

"Survey-weighted logistic regression models were fitted to estimate both crude and adjusted associations between DBM outcomes and explanatory variables. Variables with a bivariate association of p < 0.20 and no evidence of problematic multicollinearity (VIF ≤ 5) were eligible for multivariable modeling. Penalized regression with Firth’s correction was applied to DBM1, DBM2, and DBM3, with stratified analyses for overweight and obesity considered separately and in combination. Effect estimates were interpreted with attention to their magnitude and precision, emphasizing confidence intervals rather than relying solely on p-values."

This approach ensures that our results are robust, reliable, and comparable across all DBM outcomes, even in the presence of low event prevalence.

Comment 4:

Fully specify the survey design used in the software (strata, PSU, weights, and any finite population correction) to ensure reproducibility;

Response

We thank the reviewer for highlighting the need to fully specify the complex survey design to ensure reproducibility. In response, we have expanded the description of the survey design in the Materials and Methods section (2.1 Study Population and Data, lines 111–121):

"We used data from the Colombian National Nutrition Survey 2015–2016 [Encuesta Nacional de Situación Nutricional (ENSIN)] [41], a nationally representative survey covering 99% of the Colombian population through a multi-stage stratified sampling design. Non-institutionalized civilian residents were selected to represent the population. The survey included 44,202 households, organized into 177 strata. The sampling procedure followed the official ENSIN hierarchical design, with 238 primary sampling units (PSUs) comprising groupings of municipalities across all 32 departments and Bogotá. Secondary sampling units (SSUs) were groups of contiguous city blocks within each sector and census section, each including at least 96 households. Tertiary sampling units (TSUs) consisted of 5,000 segments averaging 12 contiguous households (range 6–17), of which 4,962 were effectively selected and 4,813 contained at least one completed household interview."

Additionally, in section 2.2 (Statistical Analysis), we specify how these complex survey components—including strata, PSUs, and sampling weights—were incorporated into all analyses using the survey package in R. This ensures that clustering, stratification, and weighting were fully accounted for in variance estimation and all inferential procedures, guaranteeing reproducibility.

Comment 5:

Incorporate key determinants of 25(OH)D, season/month of sampling, latitude/altitude or UV/pollution proxies, diet/supplements, and skin pigmentation or acknowledge them explicitly as limitations;

Response

We thank the reviewer for emphasizing the importance of key determinants of serum 25(OH)D that may influence the observed double burden of malnutrition (DBM). In response, we have revised the Strengths and Limitations section (lines 702–707) to explicitly acknowledge that potential residual confounding may persist due to unmeasured factors, including dietary vitamin D intake, supplement use, seasonal or monthly variation in sampling, latitude- and altitude-related differences in ultraviolet exposure, air pollution, and skin pigmentation.

While our analyses accounted for multiple individual-, household-, and community-level factors, these additional determinants could contribute to variability in vitamin D status and partially explain the observed patterns of DBM. This clarification strengthens the discussion of limitations and ensures readers understand that the reported DBM prevalence and associations are robust yet may be influenced by these unmeasured environmental and behavioral factors.

Comment 6:

Temper causal language (e.g., “determinants”) given the cross-sectional design. Addressing these points will substantially improve clarity, consistency, and policy relevance.

Response

We thank the reviewer for highlighting the importance of tempering causal language in light of the cross-sectional design. In response, we have carefully revised the manuscript to ensure that statements regarding relationships between variables are appropriately phrased. Specifically:

1. Terms such as “determinants” have been replaced with more neutral language, including “factors associated with,” “correlates of,” or “characteristics linked to,” throughout the text.
2. In the Results and Discussion sections, we clarified that observed associations between individual-, household-, and community-level characteristics and DBM outcomes indicate correlation rather than causation.
3. Sentences describing potential mechanisms or pathways (e.g., associations between physical activity, outdoor time, and vitamin D status) have been reworded to emphasize plausible links or hypotheses rather than causal effects.

These revisions improve clarity and consistency, helping readers correctly interpret the findings within the limitations of the cross-sectional study design. They also strengthen the policy relevance of the manuscript by presenting evidence in a transparent, non-causal framework.

 

 

 

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Accepted in present form