Escrever um comentário

General assessment

This manuscript examines the association between municipal political context and COVID-19 vaccination in Brazil by linking individual vaccination information from EPICOVID-19 II with municipality-level results from the second round of the 2022 presidential election. The topic is relevant to epidemiology and public health because understanding contextual correlates of vaccine uptake may help identify populations and settings in which vaccination coverage differs.

The study has several notable strengths. EPICOVID 2.0 is a large probability-based population survey covering 133 sentinel cities across all Brazilian states and the Federal District, and the use of official electoral data provides an independently measured contextual exposure. The use of individual vaccination information rather than aggregate municipal vaccination counts also represents an advance over purely ecological studies. OUP Academic

However, several aspects of the analysis and its interpretation require clarification. Most importantly, the manuscript sometimes moves from a contextual association to causal interpretations that cannot be established by this design. There are also important issues concerning temporality, adjustment for confounding, the definition of “complete vaccination,” the influenza analysis, the treatment of the hierarchical data structure, and the interpretation of the odds ratio. Addressing these points would substantially strengthen the paper.

Major comments

1. The temporal relationship between the political exposure and vaccination needs much more careful treatment. The exposure is the municipal vote share in the second round of the October 2022 presidential election, whereas COVID-19 vaccination in Brazil began in January 2021. Thus, a substantial part of the outcome defined as “two or more doses” must have occurred before the electoral exposure was measured. The 2024 survey subsequently ascertained cumulative vaccination history. Serviços e Informações do Brasil

   This does not invalidate the observed association, because 2022 voting patterns may serve as a proxy for a pre-existing political environment. However, it substantially changes its interpretation. The analysis cannot establish that the municipal voting pattern preceded or caused the vaccination behavior. I recommend explicitly describing the vote share as a proxy for municipal political context and discussing the temporal ambiguity as a central limitation. Statements suggesting that the measured political environment “influenced” vaccination should consequently be qualified. If feasible, a sensitivity analysis using an indicator of political context measured before the vaccination campaign would strengthen the temporal argument.

2. Individual-level confounding should be addressed more fully. The outcome is measured at the individual level, but the adjustment described in the manuscript appears to rely primarily on municipality-level demographic composition and mean income. SciELO Preprints This is an important limitation because EPICOVID 2.0 itself demonstrates pronounced differences in COVID-19 vaccination by age, sex, race/skin colour, education, wealth and geographic region. For example, vaccine coverage increased markedly with age in the parent study. OUP Academic

   Municipal age composition or municipal income cannot necessarily substitute for an individual's age, education, socioeconomic position or other characteristics. The principal regression should therefore, if the required variables are available, adjust for relevant individual-level characteristics in addition to contextual municipal variables. At minimum, age, sex, race/skin colour, education and individual/household socioeconomic position should be considered. Potential contextual confounders such as geographic region or state should also be examined. A useful presentation would show a sequence of models: unadjusted; adjusted for individual characteristics; and additionally adjusted for municipal characteristics. This would make it possible to determine how much of the association changes after accounting for compositional differences between municipalities.

3. Geographic and spatial confounding deserves specific investigation. Both electoral patterns and vaccination coverage vary geographically in Brazil. Consequently, an association between municipal vote share and vaccination may partly reflect broader regional differences in population composition, healthcare access, vaccine delivery, urban development or other contextual factors. The manuscript currently provides insufficient information to determine how this possibility was handled.

   I suggest including state or macroregion adjustment as a sensitivity analysis and assessing whether the estimated association persists within broader geographic areas. The authors should also consider whether residuals display spatial autocorrelation. If substantial spatial dependence is present, standard survey-based variance estimation alone may not adequately account for it. A multilevel or spatial sensitivity analysis would help establish whether the association is robust to geographic clustering.

4. The description of non-linearity and the use of the logit transformation are conceptually unclear. The manuscript states that because the association was not linear, a logit transformation of vaccine coverage was used. SciELO Preprints However, logistic regression already models a binary outcome on the log-odds scale. A nonlinear relationship between vote share and raw probability is expected under a logistic model and does not by itself demonstrate that the exposure has a linear relationship with the log odds.

   Moreover, Figure 1B appears almost perfectly linear because a model containing a linear vote-share term will necessarily generate a linear relationship on the logit scale. The figure therefore cannot by itself demonstrate that linearity on this scale is empirically appropriate. The authors should clarify how functional form was assessed. A restricted cubic spline, fractional polynomial, or categories of municipal vote share could be used to test this assumption. If a linear term adequately describes the data on the logit scale, this should simply be stated.

5. The definition of “complete COVID-19 vaccination” as ≥2 doses requires reconsideration. The manuscript repeatedly describes receipt of two or more doses as a “complete vaccination scheme.” SciELO Preprints This terminology is potentially misleading in a survey conducted in 2024. Brazilian recommendations had changed over time and differed according to age and risk group; booster doses were recommended for several priority populations, while specific childhood and immunocompromised schedules could involve different numbers of doses. Serviços e Informações do Brasil

   The underlying EPICOVID vaccination analysis itself reports coverage for ≥1, ≥2 and ≥4 COVID-19 doses; only 33.0% of respondents had received four or more doses, whereas 84.6% had received two or more. OUP Academic I recommend referring to the principal outcome simply as “receipt of at least two COVID-19 vaccine doses”, rather than “complete vaccination,” unless completeness is defined according to contemporaneous age- and risk-specific recommendations. Sensitivity analyses using ≥1 and ≥4 doses would be particularly informative because they could show whether the political-context association is specific to initial vaccine uptake or also evident for subsequent doses.

6. The influenza analysis does not support the claim of a “spillover effect.” The manuscript states that the weaker association with influenza vaccination suggests that “COVID-19 vaccine hesitancy had a spillover effect on other immunizations.” SciELO Preprints This conclusion is substantially stronger than the analysis allows. The study does not appear to model COVID-19 vaccine hesitancy as an exposure, nor does it establish that COVID-19 vaccination attitudes caused influenza vaccination behavior. Both outcomes may instead be influenced by common demographic, socioeconomic, healthcare-access, cultural or contextual determinants.

   There is an additional measurement issue that should be explicitly addressed. In the published EPICOVID 2.0 vaccination analysis, the influenza variable is described as whether the participant had ever been vaccinated against influenza, whereas the COVID-19 variables concern numbers of COVID-19 doses. Scribd These outcomes therefore have different temporal meanings. Some influenza vaccinations could have occurred long before either the pandemic or the 2022 election. The influenza result can reasonably be presented as a comparison showing that municipal vote share is also associated with a broader vaccination-history indicator, but it does not demonstrate spillover. I recommend replacing the causal wording with a descriptive interpretation and defining the influenza outcome explicitly in the Methods.

7. The multilevel nature of the data and the survey analysis need more detailed reporting. Participants are nested within census tracts and municipalities, whereas the principal exposure is constant for everyone within the same municipality. The paper states that svy logistic regression was used and that the design accounted for the hierarchical structure, but this description is insufficient to reproduce the analysis. SciELO Preprints In the EPICOVID 2.0 methodology, city was used as the stratification variable and census sector as the primary sampling unit. PubMed Central (PMC)

   The authors should report the complete svyset specification, weights, strata and clustering units used in this particular analysis and explain how uncertainty associated with a municipality-level exposure was estimated. It would also be valuable to present a sensitivity analysis that explicitly treats municipality as the contextual level—for example, a multilevel logistic model or another approach that permits residual correlation among participants sharing the same municipal environment.

8. Effect sizes should be presented in a form that is easier to interpret. The reported OR of 0.76 per 10-percentage-point increase in PL vote share is statistically precise, but the outcome is common: approximately 85% of the EPICOVID 2.0 sample had received at least two doses. OUP Academic With a common outcome, an odds ratio should not be interpreted as an equivalent proportional change in vaccination probability.

   I suggest supplementing the OR with marginal predicted probabilities of ≥2-dose vaccination at meaningful levels of municipal vote share—for example, 20%, 40%, 60% and 80%—with confidence intervals. Absolute percentage-point differences would make the magnitude of the association much clearer to public-health readers. A prevalence-ratio analysis could also be included as a sensitivity analysis.

9. The ecological/cross-level interpretation needs to be made consistent throughout the manuscript. This study is not a conventional ecological analysis in which both exposure and outcome are aggregated: vaccination is measured for individuals, while political context is measured for municipalities. It is therefore more precisely described as a cross-level contextual analysis using individual outcomes and a municipal exposure. Nevertheless, the study cannot infer an individual's political preference from the municipal vote share.

   The manuscript recognizes this limitation in one passage, but other wording risks conflating municipality-level electoral context with individual “political affiliation,” “political alignment,” or beliefs. These should be kept conceptually separate. The municipal vote-share variable is an attribute of the respondent's place of residence, not a measure of that respondent's ideology or vote.

10. Claims concerning mechanisms should be separated from empirical findings. The discussion mentions misinformation, distrust in institutions, social norms and political/informational environments as possible mechanisms. These are plausible hypotheses, but they were not measured in this analysis. SciELO Preprints The distinction between observed findings and proposed explanations should be made explicit. Phrases such as “may be consistent with” or “potential mechanisms include” would be preferable to language implying that these mechanisms were demonstrated.

Minor comments

The manuscript would benefit from a conventional Methods section specifying the study dates, eligible population, age range, sampling process, number of municipalities, construction and source of the electoral variable, outcome questions, missing-data handling, exact covariate definitions, weighting strategy and statistical software. EPICOVID 2.0 was conducted between March and June 2024 in the urban areas of 133 sentinel cities, and this information should be stated directly rather than requiring the reader to consult another publication. epidemio-ufpel.org.br

There also appears to be a sample-size discrepancy requiring clarification. The present manuscript reports 33,320 participants, whereas the main EPICOVID 2.0 methodological/publication reports 33,250 individuals. SciELO Preprints The authors should state whether 33,320 is a typographical error or represents a different analytic dataset and provide a participant flow or analytic-sample description.

For Figure 1A, the manuscript should specify whether the municipal correlation of r = −0.37 is weighted or unweighted and what each municipal vaccination estimate represents. Because each point corresponds to one municipality while city population sizes differ substantially, an unweighted municipal correlation answers a different question from the survey-weighted individual regression. It would be useful to state this explicitly. The truncated axes should also remain clearly identified, because they can visually magnify differences in coverage. SciELO Preprints

The term “political alignment” should generally be replaced by “municipal electoral context” or “municipal vote share,” unless individual political affiliation is actually measured. For reproducibility, the manuscript should also identify the candidate and the official electoral data source explicitly. The second-round presidential election occurred on 30 October 2022, with the PL candidate receiving 49.10% of valid votes nationally. Justiça Eleitoral

Finally, given that the data are stated to be publicly available, providing the analytic code and the municipality-level electoral/IBGE merge file would substantially improve reproducibility. A STROBE-style description of the observational study would also improve transparency.

Concluding assessment

The manuscript reports a clear and potentially informative contextual association between municipal electoral patterns and individual COVID-19 vaccination in a large Brazilian population survey. The principal empirical result is worth reporting, but its interpretation should remain at the level supported by the design: people residing in municipalities with different 2022 electoral profiles differed in their probability of having received at least two COVID-19 vaccine doses in the 2024 survey. The present data alone do not establish that municipal political support caused lower vaccination, that the association reflects an individual's political affiliation, or that COVID-19 vaccine hesitancy produced a spillover effect on influenza vaccination.

The most informative revision would therefore combine stronger individual-level and geographic adjustment, sensitivity analyses using alternative COVID-19 dose thresholds, clearer modelling of the municipal-level exposure, absolute predicted probabilities, explicit consideration of the exposure–outcome timeline, and a more cautious interpretation of the influenza comparison. These changes would make the distinction between the observed association and the hypothesized mechanisms considerably clearer.

Competing interests

The authors declare that they have no competing interests.

Use of Artificial Intelligence (AI)

The authors declare that they did not use generative AI to come up with new ideas for their review.