Factors Associated with the Odds, Duration, and Costs of Health-Related Absenteeism: A Population-Based Study in São Paulo City, Brazil

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for the opportunity to review this interesting and well‑written manuscript examining the associations of lifestyle behaviours, cardiometabolic risk factors, diagnosed NCDs and sociodemographic characteristics with the likelihood, duration, and costs of health‑related absenteeism in São Paulo. The paper addresses a relevant public health question and is based on a valuable population‑based dataset.

In brief, the study reports that cardiovascular disease, hypertension, type 2 diabetes and smoking are associated with increased sickness absence, and that adherence to recommended levels of leisure‑time physical activity is related to lower indirect costs. These findings are aligned with existing epidemiological and health economic evidence. While the conceptual novelty is limited—given that the core associations are well established in existing literature—the manuscript offers contextual value by providing updated estimates from a large middle‑income urban setting and by integrating economic outcomes alongside measures of absenteeism.

The manuscript is generally well written, clear, and stylistically strong. However, several aspects of the analytical rationale, model structure, and reporting need clarification before the paper is suitable for publication. Most importantly, the conceptual roles of lifestyle behaviours, cardiometabolic risk factors, and clinical diagnoses require clearer justification in relation to the modelling strategy, and the presentation of results would benefit from improved differentiation between primary predictors and adjustment variables.

For these reasons, the manuscript requires major revisions before it can be considered for publication.

Major comments

Conceptual clarity of predictor structure

A key conceptual challenge in the current manuscript concerns the simultaneous inclusion of lifestyle behaviours, cardiometabolic risk factors, and clinical diagnoses as parallel predictors, despite these variables occupying different positions on the same causal pathway. This modelling choice is not inherently inappropriate, but it requires a clearer justification than is currently provided. Specifically, the manuscript should explain whether the intention is to estimate mutually adjusted associations, direct associations with lifestyle behaviours, or broader risk profiles, and how this choice affects the interpretation of the coefficients. This conceptual framework should follow the argumentation throughout the manuscript. If such a justification cannot be articulated convincingly within the conceptual framework of the study, the authors may wish to consider alternative modelling strategies—such as hierarchical or block‑wise models, or presenting sensitivity analyses that separate lifestyle factors from disease diagnoses—to make the roles of different variables more transparent to readers.

Presentation of predictors vs. controls

In the Results section, all variables—lifestyle factors, cardiometabolic indicators, NCD diagnoses, and sociodemographic controls—are presented uniformly. This makes it difficult for readers to distinguish key predictors from adjustment covariates. The same issue persists in the Discussion. Please highlight which associations are central to the study aims.

The purpose and additional value of cost analysis

The rationale for analyzing absenteeism costs in addition to duration requires clearer justification. Because costs and sickness absence days are deeply interconnected, the manuscript should explain how the cost analysis provides additional interpretive value beyond the frequency and duration of sickness absence alone—whether, for example, the aim is to capture socioeconomic differences in productivity loss or to assess economic burden at the population level. Without such clarification, the distinct contribution of the cost outcomes remains difficult to interpret.

Description of the weighting procedure

The tables indicate ‘weighted frequencies’, which strongly implies that sampling weights were applied in the descriptive analyses.Additionally, the Methods section mentions ‘effects of complex survey sampling design’, which suggests that sampling weights and survey design features were used. However, the weighting procedure is not described explicitly. To ensure transparency and reproducibility, the manuscript should specify the type of weights applied, how they were derived, and whether they were incorporated in both descriptive and regression analyses. The potential use of weighted data should also be explicated in Tables 5 and 6 representing the results from the regression models.

Minor comments

1. In Table 1, all abbreviations—such as PPP—should be defined.
2. In Tables 2 and 3, “weighted frequencies” appear to represent weighted percentages of respondents. A brief clarification would improve transparency.
3. In Table 4, please specify units and denominators: prevalence as weighted % of respondents with ≥1 day of absence; duration in days; costs in 2015 PPP units.
4. Tables 5 and 6 include p‑values but no confidence intervals. Including 95% CIs could improve interpretability of effect sizes.
5. The placement of tables in the manuscript is somewhat inconsistent: several tables appear either too close to each other or far from the sections where they are first referenced. Improving the alignment between the text and the corresponding tables would enhance readability and help readers follow the results more easily.
6. All abbreviations in the supplementary tables are undefined. For clarity, please define all abbreviations within both tables.

Author Response

Reviewer 1

Comment: Dear Authors,

Thank you for the opportunity to review this interesting and well‑written manuscript examining the associations of lifestyle behaviours, cardiometabolic risk factors, diagnosed NCDs and sociodemographic characteristics with the likelihood, duration, and costs of health‑related absenteeism in São Paulo. The paper addresses a relevant public health question and is based on a valuable population‑based dataset.

In brief, the study reports that cardiovascular disease, hypertension, type 2 diabetes and smoking are associated with increased sickness absence, and that adherence to recommended levels of leisure‑time physical activity is related to lower indirect costs. These findings are aligned with existing epidemiological and health economic evidence. While the conceptual novelty is limited—given that the core associations are well established in existing literature—the manuscript offers contextual value by providing updated estimates from a large middle‑income urban setting and by integrating economic outcomes alongside measures of absenteeism.

The manuscript is generally well written, clear, and stylistically strong. However, several aspects of the analytical rationale, model structure, and reporting need clarification before the paper is suitable for publication. Most importantly, the conceptual roles of lifestyle behaviours, cardiometabolic risk factors, and clinical diagnoses require clearer justification in relation to the modelling strategy, and the presentation of results would benefit from improved differentiation between primary predictors and adjustment variables.

For these reasons, the manuscript requires major revisions before it can be considered for publication.

 

Response: Dear reviewer,

Thank you very much for your time reviewing our manuscript and for the valuable suggestions that have substantially improved the revised version of our paper. Briefly, we have added a sensitivity analysis (block-wise models) to the supplementary materials, revised the Results and Discussion sections to focus more clearly on the main predictors of our study, and provided more details regarding both the cost analysis, and the complex survey weighting of the ISA-Capital data. We appreciate your constructive feedback and remain open to any additional suggestions.

Sincerely,

The Authors.

 

Major comments:

1. Conceptual clarity of predictor structure

A key conceptual challenge in the current manuscript concerns the simultaneous inclusion of lifestyle behaviours, cardiometabolic risk factors, and clinical diagnoses as parallel predictors, despite these variables occupying different positions on the same causal pathway. This modelling choice is not inherently inappropriate, but it requires a clearer justification than is currently provided. Specifically, the manuscript should explain whether the intention is to estimate mutually adjusted associations, direct associations with lifestyle behaviours, or broader risk profiles, and how this choice affects the interpretation of the coefficients. This conceptual framework should follow the argumentation throughout the manuscript. If such a justification cannot be articulated convincingly within the conceptual framework of the study, the authors may wish to consider alternative modelling strategies—such as hierarchical or block‑wise models, or presenting sensitivity analyses that separate lifestyle factors from disease diagnoses—to make the roles of different variables more transparent to readers.

Response: We adopted a block-wise modeling strategy to clarify the structural relationships among our predictors, guided by established epidemiological frameworks (Victora et al., 1997). We constructed three nested models: Block 1 adjusts for distal confounders (socioeconomic and demographic variables, and survey years); Block 2 introduces intermediate lifestyle behaviors to estimate their total effects; and Block 3 (our main, fully adjusted model) adds proximal health characteristics to estimate direct, mutually adjusted associations.

We have included the full step-by-step block-wise models (Blocks 1, 2, and 3 side-by-side) for all outcomes (logistic, negative binomial, and two-part model) in the Supplementary Material (Tables S1 to S3c). This allows readers to directly observe how the coefficients for lifestyle behaviors are attenuated once health mediators are introduced. Importantly, this sensitivity analysis revealed that while some attenuation occurs, key lifestyle factors (e.g., recommended leisure time physical activity and smoking) retain independent associations with health-related absenteeism, reinforcing the value of the fully adjusted approach.

To ensure this conceptual framework follows the argumentation throughout the manuscript, we have updated both the Materials and Methods and the Discussion sections.

Changes made in the manuscript:

In the Statistical Analysis section, paragraphs 5 and 6, we updated the text to read:

“… To address this complex interplay, our primary analytical strategy aimed to estimate mutually adjusted associations. Although these variables may occupy different positions on the same inferential pathway (e.g., lifestyle influencing cardiometabolic risks, which in turn lead to clinical diagnoses), including them simultaneously allows us to assess the independent, direct contribution of each factor to health-related absenteeism outcomes. This approach models the broader risk profile commonly seen in clinical and occupational practice, where individuals present with a co-occurrence of these conditions [3, 7].”

“Regarding potential mediation effects, we conducted a sensitivity analysis using a block-wise modeling approach [31], sequentially introducing lifestyle factors and health characteristics. The results are presented in the supplementary material (Tables S1 to S3c). …”

 

In the discussion section, we added two paragraphs (19 and 20):

“Another potential limitation of this study is the simultaneous inclusion of lifestyle behaviors and health characteristics in the final multiple regression models. From an inferential perspective, this may result in overadjustment, as health characteristics can act as mediators in the pathway between lifestyle and absenteeism. Consequently, the coefficients for lifestyle factors in our final models represent their direct, mutually adjusted associations rather than their total effects.”

“However, our supplementary block-wise analysis provides a clear view of these relationships (Tables S1-S3c). Notably, while the effects of lifestyle behaviors are partially attenuated when cardiometabolic conditions are considered, factors such as leisure time physical activity and smoking retain independent associations with absenteeism outcomes. This suggests that their effects may operate through direct pathways beyond just cardiometabolic disease mediation, reinforcing the importance of the fully adjusted analytical approach.”

 

2. Presentation of predictors vs. controls

In the Results section, all variables—lifestyle factors, cardiometabolic indicators, NCD diagnoses, and sociodemographic controls—are presented uniformly. This makes it difficult for readers to distinguish key predictors from adjustment covariates. The same issue persists in the Discussion. Please highlight which associations are central to the study aims.

Response: We agree that giving equal prominence to central predictors and adjustment covariates in the results and discussion sections diluted the main aims of our study. Therefore, to improve the readability of the manuscript, we have made the following changes:

• Formatting regression tables: We have added internal subheadings (“Variables of interest” and “Control variables”) to Tables 5, 6, and 7, visually separating the core predictors (cardiometabolic indicators, NCD diagnoses, and lifestyle factors) from the sociodemographic adjustment covariates.
• Refining the results text: In sections 3.3. and 3.4., we condensed the description of the control variables. Instead of reporting the exact coefficients for sociodemographic factors in the text, we now provide a brief summary of their general behavior.
• Focusing the discussion: We reviewed the discussion section to ensure that the argumentation focuses on the variables of interest. Minor mentions of sociodemographic controls were either removed or condensed to briefly state that they behaved as expected according to the literature.

 

3. The purpose and additional value of cost analysis

The rationale for analyzing absenteeism costs in addition to duration requires clearer justification. Because costs and sickness absence days are deeply interconnected, the manuscript should explain how the cost analysis provides additional interpretive value beyond the frequency and duration of sickness absence alone—whether, for example, the aim is to capture socioeconomic differences in productivity loss or to assess economic burden at the population level. Without such clarification, the distinct contribution of the cost outcomes remains difficult to interpret.

Response: We agree that the distinct contribution of the cost analysis needed a clearer justification to highlight its interpretive value. While the duration of health-related absenteeism primarily reflects the functional and clinical impact of an illness on the individual, estimating costs using the human-capital approach incorporates socioeconomic heterogeneities (e.g., wage disparities). Consequently, the exact same duration of sick leave can result in a vastly different economic burden depending on the individual’s position in the labor market. Capturing these socioeconomic disparities in productivity loss at the population level is the primary additional value of our cost analysis. To ensure this rationale is explicit to the readers, we have revised the Introduction and the Materials and Methods (Section 2.3.1) to clarify this distinction.

 

Changes made in the manuscript:

In the Introduction, paragraph 7, we updated the text to read:

“...while clinical conditions and lifestyle may dominate in driving absenteeism likelihood and duration, the analysis of costs incorporates socioeconomic heterogeneities, providing a broader quantification of the economic burden and productivity loss at the population level. Thus, examining these interrelated yet distinct dimensions may provide evidence to help policymakers tailor interventions to address absenteeism holistically [15].”

In Section 2.3.1 (Outcome Variables), we added the following paragraph:

“It is important to emphasize that, beyond analyzing the likelihood and duration of health-related absenteeism, estimating its costs through the human-capital approach integrates the socioeconomic and health aspects of the individuals; therefore, the same duration of sick leave may result in a different economic burden. The cost analysis is essential to quantify productivity losses and to understand factors associated with the economic burden of health-related absenteeism in a population context, providing evidence for decision-making in public health policies.”

 

4. Description of the weighting procedure

The tables indicate ‘weighted frequencies’, which strongly implies that sampling weights were applied in the descriptive analyses. Additionally, the Methods section mentions ‘effects of complex survey sampling design’, which suggests that sampling weights and survey design features were used. However, the weighting procedure is not described explicitly. To ensure transparency and reproducibility, the manuscript should specify the type of weights applied, how they were derived, and whether they were incorporated in both descriptive and regression analyses. The potential use of weighted data should also be explicated in Tables 5 and 6 representing the results from the regression models.

Response: We agree that a more explicit description of the weighting procedure is essential for transparency. The sampling weights applied in this study are the design weights derived from the ISA-Capital complex sampling frame. These weights account for the inverse of the probability of selection at each sampling stage and were further adjusted for non-response. While full technical details of these calculations are provided in the cited literature [reference number 17: Alves et al., 2018.], we have now expanded the “Study Design”, “Dataset”, and “Statistical Analysis” subsections to elucidate how these weights were derived and handled during the pooling of the three survey waves.

Furthermore, we confirm that all analyses presented in this study, both descriptive statistics and multiple regression models (logistic, negative binomial, and two-part models), were performed considering the sampling weights, strata, and primary sampling units. As requested, we have added explanatory footnotes to Tables 5, 6, and 7 to explicitly state that the reported estimates account for the complex survey design and sampling weights.

 

Changes made in the manuscript:

In section 2.1 (Study Design), paragraph 3, we updated the text to read:

“… Subsequently, sampling weights, regarding the inverse probability of inclusion in the sample, and further adjusted for non-response, were applied to the interviewed individuals to ensure population-level representativeness; details regarding the calculation of the sampling weights used in the ISA-Capital have been previously published [17].”

 

In section 2.2 (Dataset), paragraph 4, we updated the text to read:

“… The three waves of the ISA-Capital survey were pooled while preserving the complexity of the sampling design. Indicators for primary sampling units and sampling strata were individualized for each survey edition, and sampling weights were rescaled to ensure population-level representativeness for the city of São Paulo [20]. This enabled the conduction of statistical analyses on the study’s analytical database, covering the 2003–2015 period.”

 

In section 2.4 (Statistical Analysis), paragraph 7, we updated the text to read:

“All statistical analyses, including descriptive statistics, logistic and negative binomial regression, and the two-part model, were conducted using Stata® software (StataCorp., College Station, TX, USA), version 18.0. The analysis incorporated the ISA-Capital’s complex survey design features (weights, strata and primary sample units), to ensure representativeness at the population level, and it included adjustments for potential correlations between subgroups of the sample, with a statistical significance level of 5% (p-value < 0.05).”

 

Minor comments:

5. In Table 1, all abbreviations—such as PPP—should be defined.

Response: We have added the definitions of all abbreviations to the table footnotes.

 

6. In Tables 2 and 3, “weighted frequencies” appear to represent weighted percentages of respondents. A brief clarification would improve transparency.

Response: We changed the term “weighed frequencies” to “weighted percentages of respondents”.

 

7. In Table 4, please specify units and denominators: prevalence as weighted % of respondents with ≥1 day of absence; duration in days; costs in 2015 PPP units.

Response: We added the specifications of the units and denominators of the variables described in the table.

 

8. Tables 5 and 6 include p‑values but no confidence intervals. Including 95% CIs could improve interpretability of effect sizes.

Response: We have added 95% CIs to Tables 5, 6, and 7.

 

9. The placement of tables in the manuscript is somewhat inconsistent: several tables appear either too close to each other or far from the sections where they are first referenced. Improving the alignment between the text and the corresponding tables would enhance readability and help readers follow the results more easily.

Response: We have relocated the tables so that they appear as close as possible to the text describing and interpreting the results. We also included subsections in the Results section, to organize the presentation of the results.

 

10. All abbreviations in the supplementary tables are undefined. For clarity, please define all abbreviations within both tables.

Response: We have defined all abbreviations in the footnotes of the supplementary tables.

Reviewer 2 Report

Comments and Suggestions for Authors

Recommendation: Minor Revisions

This is a well-structured and relevant manuscript addressing an important public health and health economics topic in a middle-income country setting. The study has several strengths, including the use of a population-based dataset, a clear analytical strategy, and the simultaneous examination of three related dimensions of health-related absenteeism: likelihood, duration, and costs. The manuscript is generally well written, the tables are informative, and the discussion is adequately grounded in the literature. The main findings are coherent throughout the abstract, results, and conclusions, particularly regarding the role of tobacco use, cardiometabolic conditions, and leisure-time physical activity .

Overall, I believe the manuscript has merit and is suitable for publication after minor revisions. My recommendation is not based on major methodological concerns, but rather on a few points that require clarification and refinement to improve interpretability and internal consistency.

Main comments

1. Clarification of the absenteeism outcome measure

The most important point concerns the operationalization of the absenteeism outcome. The manuscript defines absenteeism as “self-reported absences from work due to health-related reasons,” but the survey question used was: “How many days did you have to be absent from work, school, or routine activities?” . This wording appears broader than work-related absenteeism alone and may include absences from non-work activities. Although the sample was restricted to individuals in formal employment or paid work, this issue still deserves a clearer justification.

I recommend that the authors explain more explicitly why this survey item can be interpreted as a valid proxy for work-related absenteeism in the selected sample. At minimum, this limitation should be acknowledged more directly in the Discussion/Limitations section.

1. Interpretation of the 2015 survey-year effect

The discussion attributes the higher duration and costs of absenteeism in 2015 to the concurrent dengue, Zika, and chikungunya outbreaks in Brazil, suggesting that the survey-year effect “likely reflects this exogenous epidemiological shock” . This interpretation is plausible and interesting, but it is somewhat stronger than what the data can directly support, especially given the cross-sectional design and the absence of disease-specific absenteeism information.

I suggest softening this interpretation and presenting it more clearly as a plausible contextual hypothesis rather than a direct explanation demonstrated by the study. This would make the discussion more cautious and methodologically aligned with the evidence presented.

1. Consistency in the description of income

The manuscript describes per capita household income as a “continuous variable” in the controls section, but then states that it was categorized into low-, medium-, and high-income tertiles . In the results tables, the models clearly use income categories with low income as the reference group .

This is a minor issue, but the text should be revised so that the description of the income variable is fully consistent across Methods, Table 1, and the regression tables.

Minor/editorial comments

1. The manuscript uses terms such as “likelihood” in several places when discussing logistic regression results. Since the models report odds ratios, I recommend revising the wording in some passages for greater epidemiological precision, distinguishing more carefully between odds, probability, and likelihood where appropriate .
2. The text reports that variance inflation factors (VIFs) were estimated and presented in the supplementary material . Please ensure that the abbreviation is used consistently throughout the manuscript and supplementary material.
3. A final careful language and style revision would be helpful. The manuscript is already in good shape, but a few sentences could be slightly tightened for clarity and precision.

The issues identified above are limited in scope and can be addressed without substantial reanalysis. For this reason, I recommend acceptance after minor revisions.

Author Response

Reviewer 2

Comment: Recommendation: Minor Revisions

This is a well-structured and relevant manuscript addressing an important public health and health economics topic in a middle-income country setting. The study has several strengths, including the use of a population-based dataset, a clear analytical strategy, and the simultaneous examination of three related dimensions of health-related absenteeism: likelihood, duration, and costs. The manuscript is generally well written, the tables are informative, and the discussion is adequately grounded in the literature. The main findings are coherent throughout the abstract, results, and conclusions, particularly regarding the role of tobacco use, cardiometabolic conditions, and leisure-time physical activity.

Overall, I believe the manuscript has merit and is suitable for publication after minor revisions. My recommendation is not based on major methodological concerns, but rather on a few points that require clarification and refinement to improve interpretability and internal consistency.

 

Response: Dear reviewer,

Thank you very much for taking the time to review our manuscript and for the valuable suggestions that have substantially improved the revised version of our paper. We appreciate your constructive feedback and remain open to any additional suggestions.

Sincerely,

The Authors.

 

Main comments:

1. Clarification of the absenteeism outcome measure

The most important point concerns the operationalization of the absenteeism outcome. The manuscript defines absenteeism as “self-reported absences from work due to health-related reasons,” but the survey question used was: “How many days did you have to be absent from work, school, or routine activities?”. This wording appears broader than work-related absenteeism alone and may include absences from non-work activities. Although the sample was restricted to individuals in formal employment or paid work, this issue still deserves a clearer justification.

I recommend that the authors explain more explicitly why this survey item can be interpreted as a valid proxy for work-related absenteeism in the selected sample. At minimum, this limitation should be acknowledged more directly in the Discussion/Limitations section.

Response: We agree that the original wording of the survey question is broader than strictly occupational absenteeism, which is a common challenge when using generalized population-based health surveys (such as the ISA-Capital) rather than corporate administrative datasets.

Our rationale for utilizing this variable as a proxy for work-related absenteeism relies heavily on the strict inclusion criteria of our study design. We reasoned that by restricting the analytical sample exclusively to adults engaged in formal employment or paid work, the primary “routine activity” disrupted by illness for this demographic is highly likely to be their work, translating into a lost workday.

Nevertheless, we acknowledge that this broad wording could capture days where an employed individual missed a non-work activity, which could introduce noise or slightly overestimate occupational absenteeism. To ensure transparency, we have explicitly added this justification to the Methods section and acknowledged it as a formal limitation in the Discussion section.

 

Changes made in the manuscript:

In Section 2.3.1 (Outcome Variables), we added:

“Although the wording of this question encompasses school and routine activities, restricting our analytical sample exclusively to economically active adults (i.e., those engaged in formal employment or paid work) allowed us to use this variable as a proxy for work-related absenteeism and productivity loss.”

 

In the Discussion section (Limitations paragraph), we added:

“… Third, the operationalization of the absenteeism variable relies on a broad survey question that includes absences from “school or routine activities” alongside work. While restricting the sample to employed individuals ensures that work is their primary routine activity, this formulation may capture days missed from non-occupational obligations, potentially introducing noise or slightly overestimating the strict number of work days missed.”

 

2. Interpretation of the 2015 survey-year effect

The discussion attributes the higher duration and costs of absenteeism in 2015 to the concurrent dengue, Zika, and chikungunya outbreaks in Brazil, suggesting that the survey-year effect “likely reflects this exogenous epidemiological shock”. This interpretation is plausible and interesting, but it is somewhat stronger than what the data can directly support, especially given the cross-sectional design and the absence of disease-specific absenteeism information.

I suggest softening this interpretation and presenting it more clearly as a plausible contextual hypothesis rather than a direct explanation demonstrated by the study. This would make the discussion more cautious and methodologically aligned with the evidence presented.

Response: We agree that our original wording was too strong given the cross-sectional nature of the data. We have updated the text to soften this interpretation, presenting it clearly as a contextual hypothesis rather than a direct explanation.

 

Changes made in the manuscript (Discussion section):

“Furthermore, the substantial increase in absenteeism duration and costs observed in the 2015 survey year may reflect an exogenous epidemiological shock. This period coincided with concurrent outbreaks of Dengue, Zika, and Chikungunya in Brazil [76,77], which cause acute systemic symptoms requiring recovery periods consistent with the observed median absence for that year. However, given the lack of disease-specific absenteeism data, this interpretation remains a plausible contextual hypothesis rather than a direct explanation demonstrated by our study.”

 

3. Consistency in the description of income

The manuscript describes per capita household income as a “continuous variable” in the controls section, but then states that it was categorized into low-, medium-, and high-income tertiles. In the results tables, the models clearly use income categories with low income as the reference group.

This is a minor issue, but the text should be revised so that the description of the income variable is fully consistent across Methods, Table 1, and the regression tables.

Response: We agree that the description was ambiguous. The continuous measure of per capita household income was used strictly for the baseline descriptive characterization of the sample (reporting the overall mean). However, for the comparative descriptive analyses and all multiple regression models, we used the categorized tertiles (low, middle, and high income) to better capture potential non-linear socioeconomic effects.

To ensure consistency, we have revised the Methods section to explicitly clarify the distinct uses of the continuous and categorical forms of the income variable.

 

Changes made in the manuscript:

In Section 2.3.3 (Variables of Control), we updated the text to read:

“Socioeconomic characteristics: per capita household income (used as a continuous variable solely for descriptive sample characterization, and categorized into low-, middle-, and high-income tertiles for the regression models); …”

 

And further down in the same section, we added:

“… To preserve non-linear socioeconomic effects in the multiple regression models, per capita household income was categorized into low, middle, and high-income levels based on sample tertiles …”

 

Minor/editorial comments:

4. The manuscript uses terms such as “likelihood” in several places when discussing logistic regression results. Since the models report odds ratios, I recommend revising the wording in some passages for greater epidemiological precision, distinguishing more carefully between odds, probability, and likelihood where appropriate.

Response: We agree that the term “likelihood” was used imprecisely in the context of our regression models. To ensure epidemiological precision, we have systematically reviewed the manuscript and replaced “likelihood” with the appropriate statistical terms. Specifically, when discussing the results of the logistic regression (which estimates odds ratios), we updated the text to refer to the “odds” of health-related absenteeism.

 

5. The text reports that variance inflation factors (VIFs) were estimated and presented in the supplementary material. Please ensure that the abbreviation is used consistently throughout the manuscript and supplementary material.

Response: We have corrected the abbreviation.

 

6. A final careful language and style revision would be helpful. The manuscript is already in good shape, but a few sentences could be slightly tightened for clarity and precision.

Response: We have reviewed the manuscript and improved the text for greater clarity and precision.

 

The issues identified above are limited in scope and can be addressed without substantial reanalysis. For this reason, I recommend acceptance after minor revisions.

Reviewer 3 Report

Comments and Suggestions for Authors

The study's strengths lie in its robust statistical approach (two-part models, marginal effects) and its focus on a middle-income country context, where such evidence is scarce. However, a major revision is required to address a significant concern regarding the timeliness and generalizability of the data, along with several minor points to sharpen the manuscript's impact.

The data is from 2003, 2008, and 2015. While combining them creates a decent sample size, the most recent data is now a decade old. São Paulo, like the rest of the world, has undergone profound changes since 2015 (e.g., economic crises, the Zika epidemic mentioned by the authors, and the seismic shift in work patterns due to the COVID-19 pandemic). The paper needs a much more robust discussion of how these factors might limit the applicability of these findings to the post-pandemic, hybrid-work world of 2026. A simple acknowledgment in the limitations is no longer sufficient; this needs to be a central point of the discussion.

The authors correctly note the use of the Human Capital approach as a limitation. However, this point is a bit underdeveloped. The difference between this and the Friction Cost method isn't just academic; it can lead to wildly different policy implications. A more critical exploration of this choice—perhaps with a back-of-the-envelope calculation or a more detailed theoretical discussion of how the "friction period" might function in Brazil's unique labor market—would significantly strengthen the paper's economic credibility.

 

 

 

Author Response

Reviewer 3

Comment: The study’s strengths lie in its robust statistical approach (two-part models, marginal effects) and its focus on a middle-income country context, where such evidence is scarce. However, a major revision is required to address a significant concern regarding the timeliness and generalizability of the data, along with several minor points to sharpen the manuscript’s impact.

 

Response: Dear reviewer,

Thank you very much for taking the time to review our manuscript and for the valuable suggestions that have substantially improved the revised version of our paper. We appreciate your constructive feedback and remain open to any additional suggestions.

Sincerely,

The authors.

 

1. The data is from 2003, 2008, and 2015. While combining them creates a decent sample size, the most recent data is now a decade old. São Paulo, like the rest of the world, has undergone profound changes since 2015 (e.g., economic crises, the Zika epidemic mentioned by the authors, and the seismic shift in work patterns due to the COVID-19 pandemic). The paper needs a much more robust discussion of how these factors might limit the applicability of these findings to the post-pandemic, hybrid-work world of 2026. A simple acknowledgment in the limitations is no longer sufficient; this needs to be a central point of the discussion.

Response: We agree that the profound paradigm shift in work arrangements (particularly remote and hybrid work) brought about by the COVID-19 pandemic fundamentally changed how sickness absence is experienced and recorded today. A simple acknowledgment of data age as a limitation would indeed fail to capture the complexity of this new occupational reality.

To address this appropriately, we have added a discussion (two new paragraphs) detailing how the post-pandemic landscape affects our findings. Specifically, we explore two main fronts: (1) how hybrid work likely increases sickness presenteeism (working from home while ill), thereby changing the threshold for formal absenteeism; and (2) how the pandemic exacerbated the very cardiometabolic risk factors and sedentary behaviors identified in our models.

By framing our results as a possible pre-pandemic baseline, we highlight the contribution of the present study for future researchers aiming to measure the economic delta caused by the shift to hybrid work. We believe these additions substantially elevate the relevance and depth of our Discussion section.

 

Changes made in the manuscript:

In the Discussion section, we added the following paragraphs:

“A critical consideration when interpreting our findings is the temporal scope of our data, which concludes in 2015. Over the past decade, the global and Brazilian socioeconomic landscapes have undergone profound transformations, most notably precipitated by economic crises and the COVID-19 pandemic. These events triggered a seismic shift in occupational dynamics, particularly the widespread adoption of remote and hybrid work models [78]. In this current post-pandemic context, the traditional operationalization of health-related absenteeism has become increasingly complex [78]. With flexible work arrangements, employees may be more likely to engage in sickness presenteeism [79] rather than formally reporting an absence. Consequently, the threshold for formalizing a sick day may be higher today than it was in 2015, which could alter the frequency and the formal economic valuation of short-term absenteeism [78-80].”

“Furthermore, the COVID-19 pandemic profoundly impacted the very lifestyle behaviors and cardiometabolic risk factors analyzed in this study. Recent evidence suggests a global exacerbation of sedentary behaviors, weight gain, and mental health challenges following the pandemic lockdowns and the transition to screen-heavy remote work [81, 82]. Therefore, while the behavioral patterns of taking sick leave may have evolved with hybrid work, the underlying health mechanisms driving productivity losses have likely intensified. In this light, our findings provide a population-level pre-pandemic baseline. Establishing this historical benchmark is essential for future longitudinal studies aiming to quantify how the post-pandemic restructuring of work and the worsening of population cardiometabolic profiles have fundamentally altered the dynamics of health-related absenteeism in middle-income countries.”

 

2. The authors correctly note the use of the Human Capital approach as a limitation. However, this point is a bit underdeveloped. The difference between this and the Friction Cost method isn’t just academic; it can lead to wildly different policy implications. A more critical exploration of this choice—perhaps with a back-of-the-envelope calculation or a more detailed theoretical discussion of how the “friction period” might function in Brazil’s unique labor market—would significantly strengthen the paper’s economic credibility.

Response: We agree that there are profound policy implications regarding the method of indirect cost estimation, and the contrast between the Human Capital Approach (HCA) and the Friction Cost Method (FCM) deserves proper theoretical exploration in our context.

To address this, we have expanded the limitations section to critically examine how the friction period operates in the Brazilian formal labor market. We theorized that because the absences in our sample are predominantly short-term (e.g., 5 to 7 days), the friction period dynamics dictate that employers are unlikely to hire replacements. Instead, the workload is absorbed by colleagues or postponed. Therefore, under an FCM perspective, the actual short-term financial losses to employers might be negligible, whereas HCA presents the “upper bound” of the intrinsic value of lost health.

 

Changes made in the manuscript:

In the Discussion section (Limitations paragraph), we expanded the text to read:

“The present study has certain limitations. First, estimating absenteeism costs using the Human Capital Approach (HCA) based on an individual’s wage fundamentally measures the potential value of lost time, which often overestimates actual productivity losses compared to the Friction Cost Method (FCM) [10, 83]. The distinction between these approaches carries profound policy implications; under an FCM framework, actual productivity losses are confined to the friction period, and because the vast majority of health-related absences in our sample are short-term, employers plausibly do not hire temporary replacements. Instead, short-term productivity gaps are typically mitigated by organizational slack: colleagues absorb the workload, or the employee catches up upon returning. Consequently, a theoretical application of the FCM would suggest that the actual financial loss to employers for these short absences might be substantially lower, or even negligible, compared to our HCA estimates. Nevertheless, the HCA was retained in this study because it captures the intrinsic value of the workers’ lost health capacity and allows for comparability with the broader international literature. Policymakers should therefore interpret our cost estimates as the upper bound of potential economic value to be recovered through health promotion, rather than strictly as direct financial losses to the labor market.”

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for the revised manuscript. The authors have addressed the previous comments thoroughly, and the revisions have clearly improved the clarity and rigor of the paper. I appreciate the careful responses to the methodological suggestions, including the updates to the analyses. I have no further substantive concerns and consider the manuscript suitable for publication in its current form.

 

Reviewer 3 Report

Comments and Suggestions for Authors

Good Job. Needful has been done..