Introduction
Tuberculosis (TB) remains one of the leading cause of death from infectious diseases worldwide, posing a significant public health concern [1]. In 2022, there was a notable 28% increase in the global number of newly diagnosed cases compared to 2020 [1]. Nevertheless, the global targets set in 2018 regarding treatment, prevention, and funding have not been met, and efforts to reduce this burden remain insufficient [Reference Fukunaga2]. To reverse this trend, it is crucial for each country to intensify the identification and proper treatment of TB cases, aiming to achieve the global goal of ending the epidemic by 2035 [1,Reference Velleca3].
TB contact investigation is a crucial and cost-effective strategy [Reference Lung4], aimed at enhancing disease detection [Reference Fox5] and improving treatment outcome [Reference da Silva and Diaz-Quijano6]. Its primary goal is to promptly identify and treat any secondary cases of the disease, as well as to detect contacts with latent TB infection (LTBI) eligible for preventive treatment [7]. Additionally, it plays a pivotal role in tracing the source case, particularly for children under 5 years old diagnosed with TB, facilitating the implementation of appropriate control measures [7]. The effectiveness of the investigation is assessed by its yield, which is the percentage of screened contacts found to have TB [Reference Velleca3].
In a meta-analysis of 181 studies, the combined global prevalence of TB from contact investigations was 3.6%. This proportion was 5.0% in low-income countries, decreasing to 4.4% in middle-income countries and 1.8% in high-income countries [Reference Velen8]. In Brazil, a country with high middle-income status and a high TB burden, the incidence of TB among household contacts is estimated at 427.8 per 100000 person-years at risk, approximately 16 times the incidence in the general population [Reference Pinto9], with prevalence potentially reaching 5.7% [Reference Acuña-Villaorduña10]. Since 2009, it has been recommended that all close contacts of a smear-positive pulmonary TB case, regardless of symptoms, age, and HIV status, undergo investigation for active TB or LTBI [11].
However, in 2023, only 53.9% of the identified contacts of laboratory-confirmed new pulmonary TB cases were examined [12]. Moreover, there is a lack of comprehensive information on the yield of this strategy in routine programmatic settings, as well as whether individual characteristics of index cases are associated with a higher likelihood of TB infection among contacts.
Understanding these factors can assist national programmes adapt their contact investigation strategies to improve their effectiveness and efficiency, especially in high-incidence settings [Reference Velen8,Reference Baluku13]. Therefore, our aims were to evaluate the yield of community-based contact investigations for new TB cases, estimate disease prevalence among contacts, and identify which characteristics of index cases are associated to infection among contacts.
Methods
Study design and setting
A community-based cross-sectional study was conducted in the state of São Paulo, Brazil, from January 2010 to December 2020, using routinely collected data from the State Tuberculosis Control Program. São Paulo state is located in the Southeast region of Brazil and is the most populous and developed in the country. It leads the nation in TB cases, accounting for 24.5% of the total, with an estimated incidence above the national average of 42 cases per 100000 person-years [12].
Participants
The study included all contacts of new TB cases (index cases), defined as individuals who had never received TB treatment or had taken anti-TB medications for 1 month or less [1,11]. Our choice was based on the predominance of new cases (83%) and our consideration that the other cases belong to a distinct population, particularly regarding the exposure times between contacts and index cases, which consequently leads to a higher yield [Reference Baluku13]. In this analysis, index cases were defined as patients diagnosed with TB according to national guidelines [11], and contacts were any individuals who had been exposed to an index case [7]. Screened contacts were those referred by index cases in the notification form. Examined contacts included all individuals who underwent clinical evaluation [7]. During this assessment, it was expected that all steps proposed by the contact investigation algorithm would be completed.
Data source
All information was obtained through the electronic Notification and Monitoring System for Tuberculosis Cases in the State of São Paulo (TBWEB). This system encompasses all TB cases reported by state residents and, in addition to the individual and clinical characteristics of index cases, includes three specific fields related to contact investigation, specifying the number of contacts screened, examined, and diagnosed with active TB per index patient. These recorded counts of contacts were used to determine the study outcomes.
Contact investigation procedure
Following the diagnosis of TB in the index case, regardless of clinical presentation, healthcare professionals conduct an in-person interview with the patient to gather information about all their contacts, including names, ages, and risk assessment, in order to prioritize clinical examination. Furthermore, they educate the patient on the importance of contact investigation. Subsequently, they request that contacts visit the designated health facility for evaluation or be contacted to schedule a visit, as needed [7,11].
Contacts are then assessed for the presence of persistent cough of any duration or other symptoms such as persistent fever, weight loss, anorexia, and night sweats, among others. Regardless of symptoms presented, a chest X-ray is requested. Contacts under 10 years old undergo tuberculin skin testing or interferon-gamma release assay (IFN-γ) to check for LTBI. Those over 10 years old who are capable of producing a sputum sample are investigated using sputum smear microscopy or GeneXpert MTB/RIF®. Cases positive on these tests are diagnosed with active TB and immediately start treatment, tailored to drug resistance patterns [11].
Contacts unable to produce sputum or those with negative sputum results but abnormal radiographic findings are referred for additional medical clinical evaluation. Also, asymptomatic contacts are screened for LTBI and, if necessary, referred for treatment [11]. However, we did not have access to this information; therefore, our assessment was limited to cases of active TB. It is important to note that all tests for the diagnosis and treatment of TB are fully covered by the Brazilian Unified Health System (SUS) [11]. Supplementary materials provide flowcharts for contact investigation based on the age of contacts (Supplementary Figures S1 and S2).
Variables
The primary outcome was the detection of TB among contacts of TB index patients. We interpreted ‘positive yield’ as the proportion of this outcome, i.e., the percentage of screened contacts diagnosed with active TB as a result of TB contact investigation strategy [7,Reference Baluku13]. The following indicators were also assessed: proportion of index cases for which contacts were registered; proportion of screened contacts who were examined; number needed to screen (NNS), and number needed to test (NNT), to identify one new TB case [7].
Due to the absence of individual information on contacts, the independent variables considered in the analyses pertained to the characteristics of the index cases. These included sociodemographic information, health behaviours, medical history, and TB-related characteristics. A full description of all variables used in the study can be found in the Supplementary Table S1.
Statistical analyses
The characteristics of index cases and information about TB contact investigation were presented descriptively.
Predictive model for contact examination
We conducted various predictive modelling to estimate the likelihood of index cases having their contacts examined based on their characteristics. The final model was selected based on multiple criteria (Supplementary Figure S4 and Table S2). The zero-inflated Poisson (ZIP) regression model, with which we obtained a pseudo-R 2 of 51.6% (Supplementary Table S2), showed the best fit compared to the other models evaluated and was used to obtain probability estimates (Figure 1).
Figure 1. Comparisons among observed versus predicted probabilities among count models.
Abbreviations: PRM – Poisson Regression Model. NBRM – Negative Binomial Regression Model. ZIP – Zero-inflated Poisson.
Prevalence and factors associated with infection among contacts
We calculated sampling weights as the inverse probability of contacts being examined using the predictions from the previous model. This approach allowed us to derive two prevalence estimates: one unweighted, representing the examined contacts, and one weighted, representing the screened contacts, both with their respective 95% confidence intervals (CIs). By comparing these two values, we estimated underreporting, indicating the likely number of undetected cases among all screened contacts. It is important to note that there is no record explaining why not all screened contacts were examined. Therefore, we assumed that unexamined contacts do not have a lower prevalence of TB, conditioning for the known characteristics of the index cases.
After, we also investigated factors associated with TB among contacts based on index case characteristics using multilevel mixed-effects Poisson regression model. Additionally, we incorporated random effects at the municipal level to address variability not explained by fixed predictors and employed robust standard error estimates. We obtained adjusted models both unweighted and weighted (based on the previously described weights), allowing estimates for examined contacts and the total screened contacts, respectively. Moreover, the weighting approach avoided introducing collider bias in the inference for screened contacts (Supplementary Figure S5).
The adjusted models were built using a hierarchical analysis, structured based on a conceptual framework (Supplementary Figure S6). This framework includes: first, temporal and geographical characteristics of index cases, as distal variables; second, sociodemographic and health characteristics, as intermediate I and II variables; and third, case detection strategies and clinical characteristics of index patients, as proximal variables. We interpreted the results in terms of prevalence ratio with their 95% CI, adopting a significance level of 5%. To address the missing values in the age variable, we performed simple data imputation (0.1% of index cases). For variables with more than 5% missing values, we included these cases as an additional category in the analysis.
We also estimated the intraclass correlation coefficient for each multilevel Poisson regression model to assess the proportion of total variation in TB prevalence among contacts attributable to differences between municipalities. Furthermore, we examined the effect of each municipality on TB prevalence among contacts and generated a caterpillar plot that organizes predicted proportions in ascending order along with their respective 95% CIs.
All analyses were performed using Stata version 16.1 (StataCorp LP, College Station, Texas, USA). This study was reported according to the recommendations of the RECORD statement.
Results
Characteristics of index cases
Between 1 January 2010 and 31 December 2020, a total of 186446 new TB cases were reported to the TBWEB system. Of these index cases, 70.7% were male (n = 131777) with a median age of 35 years (IQR: 25–49), and 84.6% (n = 157830) with pulmonary anatomical classification. Table 1 shows the remaining characteristics of the index cases and compares them with the screening of at least one contact.
Table 1. Characteristics of tuberculosis index cases, yield of tuberculosis contact investigations, number needed to screen, and number needed to treat in São Paulo, Brazil, 2010–2020
Note: Yield was defined as the total number of active TB cases divided by the total number of contacts screened. NNS was expressed as the total number of contacts screened divided by the number of active TB cases needed to detect one new TB case. NNT was expressed as the total number of contacts examined divided by the number of active TB cases needed to detect one new TB case.
Abbreviations: NNS – number needed to screen. NNT – number needed to test. PCF – passive case-finding. ACF – active case-finding.
The yield of contact investigation
Among index cases, 131055 (70.3%) had at least one contact registered in TBWEB, totalling 652286 contacts screened (5 per index case). The median number of contacts screened per index case was 3 (IQR: 2–5), ranging from 1 to 300. Regarding contact investigation, 451704 (69.2%) underwent examinations to detect the presence of the disease. The median number of contacts examined per index case was 3 (IQR: 2–5), ranging from 1 to 272. In total, 12243 new TB cases were diagnosed, representing an overall yield of contact investigation of 1.9% (Figure 2), resulting in an NNS of 53 and NNT of 37 (Table 1). The yield was higher among contacts of index cases with pulmonary TB (1.9%; NNS = 51) compared to those with extrapulmonary TB (1.2%; NNS = 86), and varied from 0.6% (NNS = 172) among contacts of index cases in correctional facilities to 12.1% (NNS = 8) among contacts of index cases under 5 years old (Table 1).
Figure 2. Flowchart of screening and yield from tuberculosis contact investigation in the State of São Paulo, Brazil, 2010–2020.
The prevalence of TB among contacts
We found that the unweighted prevalence among examined contacts was 2.7% (95% CI: 2.6%–2.8%). The weighted prevalence representing the total screened contacts was 2.8% (95% CI: 2.7%–2.9%), resulting in 18264 cases (95% CI: 17612–18916) among all screened contacts. These results suggest that 6021 cases (95% CI: 5269–6673) of undetected infections among contacts referred by index cases. Table 2 shows additional prevalence values disaggregated according to the characteristics of the index cases.
Table 2. Prevalences and multilevel Poisson regression analysis adjusted for characteristics of tuberculosis index cases associated with the presence of tuberculosis diagnosis among their contacts. São Paulo, Brazil, 2010–2020
Note: Model 1: unweighted, representing examined contacts. Model 2: weighted, representing the total screened. Bold values indicate statistically significant associations.
Abbreviations: PR – prevalence ratio. 95% CI – 95% confidence interval. PCF – passive case-finding. ACF – active case-finding. SE – standard error. ICC – intraclass correlation coefficient.
* Prevalence weighted by the inverse probability of being examined.
a Distal model, RP adjusted for years of diagnose and metropolitan area of residence.
b Intermediate model, PR adjusted for sex, age group, self-reported or race or ethnicity, country of birth, education, homelessness, HIV status and comorbidities, plus distal variable.
c Proximal model, RP adjusted for case detection strategies, chest X-ray, anatomical classification and drug-resistant tuberculosis, plus distal and intermediate variables.
Factors associated with TB diagnosis among contacts
Index patients from the metropolitan areas of Campinas, Baixada Santista, Vale do Paraíba e Litoral Norte, and Sorocaba were associated with a higher likelihood of TB diagnosis among their contacts compared to those in other metropolitan areas. Similarly, index cases of Black or Brown races/ethnicities showed a greater probability of TB compared to White race/ethnicity patients. This trend was also observed among contacts of female index patients, foreigners, and individuals experiencing homelessness. However, an inverse relationship was noted between younger age and lower years of schooling among index patients and the likelihood of TB diagnosis among their contacts, compared to index patients aged 60 years and older and those with more than 12 years of education, respectively (Table 2).
Additionally, index cases who were smokers, illicit drug users, had a pulmonary TB diagnosis, or drug resistance also showed an increased probability of TB diagnosis among their contacts compared to index patients without these characteristics. Furthermore, index cases with unknown HIV status, identified through active case-finding strategies, and with abnormal chest X-rays were associated with a higher probability of TB diagnosis among their contacts, compared to HIV-negative index patients, identified through passive case-finding, and with normal chest X-rays, respectively (Table 2).
Examining the effect of municipalities on TB prevalence among contacts
We found that 6.6% (95% CI: 5.0%–8.7%) and 8.6% (95% CI: 6.7%–10.9%) of the variation in TB prevalence among examined and screened contacts, respectively, was attributed to variation between municipalities (Table 2). For 21 municipalities, the 95% CIs were below the zero line, indicating lower predicted TB prevalence among contacts compared to the average. In contrast, 43 municipalities had 95% CIs above the zero line, suggesting higher TB prevalence among contacts than the average. For 90% of the municipalities included, it was not possible to distinguish from the overall average due to overlapping 95% CIs with the zero line (Figure 3). For more information, please consult Supplementary Table S3.
Figure 3. Caterpillar plot showing the effect of municipalities on tuberculosis prevalence among contacts and their respective 95% confidence intervals (n = 639). São Paulo, Brazil, 2010–2020.
Discussion
In this study, we evaluated the yield of TB contact investigation strategies among index cases in the state of São Paulo, Brazil, using available surveillance data. We estimated the prevalence of TB among contacts and identified the characteristics of index cases associated with active TB diagnosed among their contacts. The yield was 1.9% among screened contacts, increasing to 2.7% among those examined. These results align with estimates from other studies in Brazil across different populations, ranging from 1.9% to 3.0% [Reference Cubillos-Angulo14,Reference Cailleaux-Cezar15], although they remain slightly lower than the global range of 2.87%–3.60% reported in previous studies [Reference Velleca3,Reference Fox5,Reference Velen8,Reference Deya16]. Nevertheless, these numbers are comparable to those observed in other countries in the Americas (2.68%) [Reference Velleca3] and in similar income and incidence settings (2.22% and 1.9%, respectively) [Reference Velleca3,Reference Velen8]. They are also consistent with the yield reported in another study that used data from national TB surveillance program data (1.8%) [Reference Blok17]. Additionally, our findings reveal that the weighted prevalence inferred for all screened contacts was 2.8% (95% CI: 2.7%–2.9%), representing underreporting of nearly one-third of all cases among contacts. This result aligns with previous studies that employed different approaches to determine underreporting [Reference de Oliveira18].
Our study shows that, on average, nine contacts are diagnosed with TB for every 100 index cases screened. This estimate likely underestimates the true proportion of cases per index patient, as 30% of them did not report any contacts. However, it is crucial to note that this non-screening rate is lower than that found in other high TB-burden countries in Africa, Asia, and the Middle East [Reference Blok17]. Another significant finding is the proportion of contacts that were actually examined, which accounts for nearly 70% of the total contacts screened. We considered that this gap between screened and examined contacts may explain the lower yield compared with previous studies.
It is essential to emphasize that the issue of underdiagnosis of the disease requires further research, especially due to previous findings that highlighted the stigma associated with TB and HIV as a significant barrier to contact investigation [Reference Faccini19,Reference Kolte20]. Furthermore, other studies indicate that difficulties in reporting contacts by index cases may be attributed not only to stigma but also to the complexity arising from the number and identity of potentially exposed contacts, the strategies used for contact tracing, limited knowledge about the disease among contacts, challenges in accessing health services, and inadequate follow-up by health teams after contact identification, among others [Reference Fenta21,Reference Ayakaka22]. These factors, whether alone or in combination, may contribute to a scenario where TB cases among contacts are not promptly diagnosed and treated, thereby increasing disease transmission [Reference Kolte20].
This result further underscores the need to ensure the completion of the entire contact tracing cascade for all eligible index cases, thereby avoiding selection bias towards individuals who self-identify as symptomatic [Reference Subbaraman23]. This process is crucial, particularly in resource-limited settings where contacts are encouraged to seek medical assistance only when symptoms appear [Reference Deya16]. Previous evidence supports this finding, demonstrating that locations testing all contacts, regardless of symptoms, achieved a more significant detection of TB cases compared to those applying more restrictive criteria [7,Reference van’t Hoog24]. Moreover, integrating laboratory tests into contact tracing activities can be a valuable investment, given the limitations of relying solely on symptoms to guide TB case screening [Reference Deya16,Reference Habte25]. It is also crucial to highlight the importance of identifying and initiating early treatment for LTBI among eligible contacts of index cases, which helps prevent future reactivations of TB [Reference Souza26]. Unfortunately, these data on LTBI were not available for analysis in our study.
In this investigation, on average, 53 contacts need to be screened and 37 examined to identify a case of active TB, a result similar to other studies [Reference Blok17,Reference Bohlbro27,Reference Loredo28]. This implies conducting approximately eleven home visits, considering the average number of contacts per index case (53/5), which can be a significant expenditure of resources and a burden on healthcare professionals, depending on the location. Notably, we observed that this number is significantly lower in certain groups, which supports World Health Organization (WHO) and national guidelines to target screening towards specific groups at higher risk of the disease [7,11]. This targeted approach suggests that additional screening efforts in these groups would be an effective way to enhance the detection and control of TB [Reference Deya16,Reference Bohlbro27].
Our analyses indicated that contacts of index cases with drug resistance were more likely to be diagnosed with TB compared to contacts of drug-susceptible index cases [Reference Shah29], although the literature remains inconsistent [Reference Grandjean30]. We also observed that contacts of index cases with pulmonary TB and abnormalities on chest X-rays were more likely to be diagnosed with TB compared to cases with extrapulmonary TB and normal chest X-rays. The high risk of infection in contacts of patients with these characteristics is plausible because pulmonary lesions release large quantities of bacilli, including drug-resistant strains, which significantly enhances the transmission of the disease to close contacts [Reference Pinto9,Reference Cubillos-Angulo14,Reference Shah29]. In addition, the strategy used to identify the TB index case was associated with the presence of the disease among contacts. This finding add to others studies that emphasize the effectiveness of active case-finding strategies in early detection of new cases and LTBI [Reference Fox5,Reference Mhimbira31], and in reducing mortality and unfavourable treatment outcomes [Reference da Silva and Diaz-Quijano6].
Studies have shown that smoking, illicit drug use, and other immunosuppressive conditions are known risk factors for TB infection, due to their negative impact on lung function and the immune system’s ability to fight infections [Reference Deiss32,Reference Mathema33]. We also identified an association between TB index cases with such characteristics and the disease prevalence among their contacts. This relationship is likely a result of increased exposure of contacts to infectious droplets from index cases, due to the sharing of objects in close environments with low air circulation, no ultraviolet light, and inadequate protection [Reference Arriaga34,Reference Reichler35]. We also noted an intriguing finding, contacts of TB index cases with diabetes mellitus (DM) showed a lower likelihood of TB diagnosis compared to those without DM, which concurs with previous studies [Reference Grandjean30]. However, this result contrasts with two other studies conducted in Brazil [Reference Arriaga34,Reference Rajan36]. This discrepancy reveals the complexity of the relationship between DM and TB and suggests the need for more comprehensive cohort studies to clarify these associations between TB index cases and their contacts.
Despite strong recommendations in national [11] and international guidelines [7] for screening and testing all contacts of people living with HIV, less than half of them had their contacts screened in our study, with only 58.8% (n = 18403) undergoing examination. These findings indicate the urgency of implementing tailored strategies to ensure an inclusive and effective approach in controlling TB and HIV among their contacts. The association identified in our study between sex and TB diagnosis can be attributed to the central role women play in family activities, particularly in looking after others. This leads to extended exposure time for their contacts [Reference Jia37], reflecting a pattern observed in other studies [Reference Pinto9].
Our study shows a higher prevalence of TB among contacts of index patients belonging to groups with historical characteristics of social vulnerability, such as those with low education, black and mixed-race ethnicity, homelessness, foreigners, and residents in overcrowded populated areas. These groups, likely due to their precarious housing conditions, restricted access to healthcare services, low income, and unemployment, are more likely to come into contact with patients with active TB, thereby facilitating the spread of the disease among contacts [Reference Hamilton38,Reference Liu39].
Finally, we found an inverse association between the age of index cases and the prevalence of TB among their contacts, consistent with other studies on the subject [Reference Pinto9,Reference Saunders40]. This finding is understandable, as children are prioritized in contact tracing policies due to the high likelihood of disease transmission occurring within the family environment [11]. Therefore, identifying index patients under 5 years old might trigger more intensive case-finding efforts within healthcare services [Reference Pinto9]. Hence, we suggest that improving contact investigation across all age groups of index patients could have a significant impact on TB prevention and treatment within the state and the country as a whole.
Limitations and strengths
Our study also had some limitations. The unavailability of data on the characteristics of contacts, including their relationship with the index case and genotypic matching, limited our ability to thoroughly assess whether these factors could influence the likelihood of being examined and the potential determinants of TB among them. This information is not routinely recorded in TBWEB. In addition, we assumed that the lack of examination for some contacts was not related to variables other than those assessed in the index cases. We cannot rule out the possibility of underreporting of contacts by index cases, which may suggest that our underreporting estimates are conservative (i.e., may be higher). Therefore, it is essential to enhance surveillance and recording of contacts in TBWEB to obtain more accurate prevalence estimates in future studies. Due to the cross-sectional design of the study, we were unable to establish the time interval between the initial TB report of the index case and the development of the disease among their contacts. This limitation prevents us from determining whether the cases are co-prevalent or incident. However, we believe they are likely co-prevalent, given that the study focused exclusively on new TB cases and employed an investigation algorithm that typically prioritizes recent cases to identify active disease outbreaks. Moreover, we cannot confirm whether the contacts acquired TB through direct transmission from the index patient, external exposure, or reactivation of LTBI. These considerations underscore the importance of including such information in future longitudinal studies to better elucidate transmission patterns, disease risk under different circumstances, and the cost-effectiveness of contact investigations. This would provide more robust evidence. Furthermore, such studies could determine whether conducting contact screening for all cases would increase yield compared to current symptom-based recommendations, which focus on individuals with bacteriologically confirmed pulmonary TB, children aged 5 years or younger, and people living with HIV.
Despite these limitations, the study has several strengths that deserve highlighting. First, the data for this evaluation were routinely collected by the state TB control programme and, therefore, accurately reflect programmatic conditions in low- and middle-income settings with a high TB burden. Consequently, our results are likely generalizable to similar settings where the WHO currently recommends contact investigations. Second, our analysis contributes to a growing body of literature that assesses the effectiveness of TB contact investigation strategies and their performance across different groups and predisposing factors. Third, we employed weighting strategies to ensure the representativeness of the number of contacts screened in the prevalence results, allowing us to estimate the probable number of underreported cases. Fourth, given that the weights were derived from a well-performing model, we believe this contributed to generating more accurate estimates for the entire screened population, thereby mitigating potential selection biases related to factors influencing the examination of contacts. Finally, the study’s strength lies in the large number of contacts screened and examined compared to previous studies.
Consequently, we recommend strengthening and expanding contact investigations for all TB index cases to facilitate early detection and appropriate treatment of new cases. On the other hand, in resource-limited settings, priority should be given to investigating contacts of specific index cases, such as those in socially vulnerable groups, including women, children, and cases indicating more severe disease. These findings have significant implications for public health policies related to TB control, not only in the state of São Paulo but also in other regions with similar contexts.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0950268824001675.
Data availability statement
Due to the ethical reason, data sharing is not applicable.
Acknowledgements
We would like to thank all staff of the Tuberculosis Control Division at the Epidemiological Surveillance Center ‘Prof Alexandre Vranjac’ of the São Paulo State Department of Health, Brazil.
Author contribution
JMNS and FADQ conceptualized the study. JMNS conducted data collection, organization, and analysis. FADQ provided supervision and contributed to the planning of the analyses. All authors participated in interpreting the results. JMNS drafted the initial version of the manuscript. FADQ made substantial contributions to the revision, and both JMNS and FADQ reviewed and approved the final version.
Funding statement
This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brazil (CAPES) – Finance Code 001 as a Brazilian CAPES scholarship to JMNS. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interest
The authors declare none.
Ethical standard
The study was approved by the Research Ethics Committee of the School of Public Health at the University of São Paulo (protocol number: 4285870).
Open access
