key: cord-0950909-b37uluhc authors: Hajar, F. N.; Fernandes-Silva, M. M.; Pereira da Cunha, G. S.; Herrera, G.; Hamud, A.; Azevedo, V. F. title: Epidemiological and clinical characteristics of COVID-19 in Brazil using digital technology date: 2020-10-02 journal: nan DOI: 10.1101/2020.09.30.20204917 sha: 6372799809c9c6f16415a2788940f9be0b69a6fa doc_id: 950909 cord_uid: b37uluhc Background: Brazil has the third-highest number of Coronavirus disease 2019 (COVID-19) cases worldwide. Understanding the epidemiology of COVID-19 from reported cases is challenging due to heterogeneous testing rates. We estimated the number of COVID-19 cases in Brazil on a national and regional level using digital technology. Methods: We used a web-based application to perform a population-based survey from March 21st to August 29th, 2020 in Brazil. We obtained responses from 243 461 individuals across all federative units, who answered questions on COVID-19-related symptoms, chronic diseases and address of residence. COVID-19 was defined as at least one of the following: fever, cough, dyspnea and nasal flaring, associated with a history of close contact with a suspect or confirmed COVID-19 case in the previous 14 days. A stratified two-stage weighted survey analysis was performed to estimate the population level prevalence of COVID-19 cases. Results: After calibration weighing, we estimated that 10 339 461 cases of COVID-19 occurred, yielding a 2.75 estimated infection per officially reported case. Estimated/reported ratios varied across Brazilian states and were higher in states with lower human development indexes. Areas with lower income levels displayed higher rates of COVID-19 cases (66 vs 38 cases/1000 people in the lowest and highest income strata respectively, p<0.001), but presented lower rates of COVID-19 testing. Conclusion: In this population-based survey using digital technology in Brazil, we estimated that the COVID-19 case rates were 2.75 times higher than officially reported. The estimated per reported case ratios were higher in areas with worse socioeconomic status. The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is the causal agent of Although COVID-19 case reporting requires accurate testing using RT-PCR or serology, widespread testing has shown to be impractical, with highly heterogeneous per-capita testing rates across different regions in the world, depending on test production inputs, resources and allocation constraints. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint We, therefore, described the results from the largest web-based collaborative COVID-19 application in Brazil -called Together Against COVID. We used the Together Against COVID database to estimate the cumulative incidence rates of COVID-19 cases in each Brazilian federative unit, and compared them with the respective rates of reported cases. We also evaluated the COVID-19 case rates according to neighborhood socioeconomic characteristics. These results will help provide insights on the epidemiology of the COVID-19 pandemic in Brazil, the world's third most-hit country and the current epicenter of the pandemic. This was a nationwide, large survey-based study including all brazilian federative units. On March 21st, 2020, we started a non-governmental web-based application called Juntos Contra o COVID -Together Against COVID in Portuguese, available in juntoscontraocovid.org. After consenting, each participant filled in a form answering questions on current symptoms, chronic diseases and address of residence. When they completed the form, they had access to a map showing the location of suspected COVID-19 cases. Participants received a weekly email to update their symptom status. The survey had no pre-specified eligibility criteria, ranging from all ages. No symptoms nor prior known contact with the coronavirus was used as an inclusion criteria. Furthermore, no exclusion criteria were predefined. Participants from all over the country were able to contribute in a collaborative fashion. For this study, we presented data collected up to August 29th, 2020. The primary recruitment method was the snowball strategy -where previous participants invite known potential participants to enter the survey. This organic sampling method was All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint followed by social media engagement and news portals sharing. However, the latter served, primarily, as a focus for the snowball recruitment strategy. Two main strategies were used to enhance patient recruitment. The first relied on social media and message platforms. Predetermined messages and images were created to publicize the platform as a free, collaborative and useful tool -where people could navigate through a map and check if their surroundings had suspected COVID-19 cases. This strategy worked as a mainstream method to enhance individuals' access to the web-platform, where they were invited to participate in the research project, sharing their symptoms and health information. The second strategy leaned on publicizing the platform in nationwide news channels. The overall reach of the platform was over 5 million Brazilians. We collected data on self-reported age, gender, COVID- 19 Accordingly, there are three scenarios where a patient is a suspected COVID-19 case: All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint I -Patient evolves with fever and at least one of the following signs and symptoms: cough, dyspnea, nasal flaring; and travel history to local transmission area, according to WHO, up to 14 days before the first symptoms; or II -Patient evolves with fever and at least one of the following signs and symptoms: cough, dyspnea, nasal flaring; and close contact with a suspected COVID-19 case, up to 14 days before the first symptoms; or III -Patient evolves with fever or at least one of the following signs and symptoms: cough, dyspnea, nasal flaring; and close contact with confirmed COVID-19 case, up to 14 days before the first symptoms; A participant who filled in any of the above rules was defined as a suspected COVID-19 case. All data were maintained on an AWS hosted Structured Query Language (SQL) database. Participants were identified by an anonymous hash value created upon form completion. This value was used to track multiple responses from the same user given a pre-established email pattern. We performed a stratified two-stage weighted survey analysis to estimate the population level prevalence of COVID-19 cases. A Federative unit was used as stratum, the census block as the first stage (primary) sampling unit, and individuals as the second-stage sampling unit. Sampling weights were defined by the inverse of the probability of at least one census block response within each federative unit and the inverse of the probability of the individual response within the respective census block. To account for selection bias from survey responses, sampling weights were adjusted using the general regression method for All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint calibration with age, sex and census block average income level (< ½, ½ to <1, 1 to <2, 2 to <3, 3 to <5, and 5 and above minimum wage) as auxiliary variables. Age-and sex-specific population totals were obtained from the 2020 estimated population size for each federative unit according to the IBGE. [8] Income level category-specific population totals were obtained from the 2010 Census data. The model performance was checked by comparing the final estimated average age, sex and income levels from the survey analysis with official population data from IBGE Census. From this model, we estimated the cumulative incidence rates of COVID-19 cases by federative unit and compared them with the reported cases. Finally, we performed a logistic regression model using the same survey-based approach to evaluate the association of income level and number of individuals per household (< 2, 2 to < 3, and 3 and above) with the rates of COVID-19 cases. The analyses were performed using Stata version 15.1 (Stata Corp, College Station, TX). Supplementary table 1) . Among participants with suspected COVID-19, the most common symptom was fever (76%) followed by cough (75%), anosmia (40%), ageusia (40%) and dyspnea (38%), and 44% of them reported contact with a confirmed COVID-19 case in the previous 14 days (Supplemental figure 1). Table 1 shows the descriptive characteristics of the sample, along with data after survey weighting and calibration methods, and the population demographics according to Brazilian census. [11] Compared with Brazilian population, our sample had an over-representation of women between 20 and 60 years old, and an under-representation of individuals younger than 20 years old and older than 80 years old. Moreover, it was over-represented by individuals living in regions with higher income per capita. After weighing and calibration, the All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. By August 29, 2020, the Ministry of Health reported 3,761,391 COVID-19 cases, as displayed in Figure 1A . Nevertheless, we estimated that 10,339,461 cases occurred in the same period, with different relative distributions among the federative units ( Figure 1B , table 2 ). Overall, the ratio between the estimated and reported COVID-19 cases was 2.75:1 in Brazil, but this ratio highly varied across states, with higher ratios in the states of Pará (PA), Rio Grande do Norte (RN) and Pernambuco (PE) ( Figure 1C ). In Mato Grosso (MT), Mato (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. The COVID-19 case rates in Brazil were significantly associated with socioeconomic factors, showing a "J" shaped association with income, with a down trend from the lower to the All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint upper-middle income regions and an uptrend from upper-middle (3 to < 5 minimum wage) to the upper income stratum (5 and over minimum wage, figure 3A ). The estimated COVID-19 rates in the lower income stratum were approximately three times higher than the rates in the upper-middle income stratum (68 vs 20 per 1000 people) and twice the one in the upper income stratum (68 vs 39 per 1000 people, Figure 3A ). On the other hand, self-reported COVID-19 testing was the most common in the highest income stratum (123 tests per 1000 people) compared to other strata (p<0.001, Figure 3B ). The COVID-19 rates varied across age strata, with the highest rates in the population between 20 and 39 years and the lowest ones between 60 and 79 years old. Noteworthy, COVID-19 was more common among individuals with comorbidities, particularly those with self-reported cardiovascular diseases, than those with no comorbidities (Table 2) . To our knowledge, this is the first study to estimate the cumulative incidence rates of COVID-19 using a web-based collaborative tool in Brazil. In this population-based survey, the estimated COVID-19 case rate was 2.75 times higher than it has been reported. Underreporting of COVID-19 varied across states, as reflected by different estimated/reported COVID-19 ratios, and it appears to be higher in states with lower HDI. The rates of COVID-All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. Brazil from March to May, 2020 that were 2.3-fold higher than the SARI due to COVID-19. [18] In the Metropolitan Region of São Paulo in the southern of Brazil, confirmed COVID-All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint 19 diagnosis among SARI cases was less likely in areas with lower per-capita income. Similarly, our study found that COVID-19 case rates from self-reported symptoms were 2.75fold higher than the reported cases, with higher estimated/reported ratios in states with lower HDI and lower testing rates. [18] This may imply non-equitable access to testing and diagnosis, but the reasons for under-reporting of COVID-19 are complex and multifactorial, involving both healthcare delivery, cultural aspects and socioeconomic characteristics. We showed a significant association between COVID-19 case rates and socioeconomic factors. Areas with lower income and more people living in the same house were associated with higher rates of COVID-19. Social distancing may be seemingly impossible in small, overcrowded and poorly ventilated houses with a single room, enhancing the transmission of respiratory viruses whenever a family member develops symptoms. [19] [20] [21] Although there is a lack of data on the transmission of SARS-Cov-2 in poor communities, a study suggested that a hypothetical pandemic of a new strain of influenza would have greater impact in a lowincome country, such as Papua New Guinea, than in a developed country, mainly due to a larger number of individuals per household. [19] Consistent with this hypothesis, the prevalence of COVID-19 was not associated with income strata and number of individuals per household in the seroprevalence study in Spain. [14] This suggests that the higher transmissibility in crowded households may particularly affect lower-income countries. Policy strategies that address crowded households when an individual is tested positive can be important tactics to mitigate the spread of COVID-19 in low-income settings. On the other hand, our data showed that the association between COVID-19 rates and income remained almost unchanged after adjusting for the number of individuals per household, suggesting that other factors play an important role in this association. We found that the COVID-19 case rates were higher, but testing rates were lower, among census blocks with lower average income. This underscores the heterogeneous access to healthcare even when there is universal health coverage. Individuals in the lower income strata, who were at higher risk of COVID-19, were less likely to receive testing and, therefore, All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint to self-isolate. [22] In addition to the disease itself, poor communities face difficulties indirectly caused by the pandemic, such as a drastic reduction in income and a loss of community targeted support. [23] Besides that, low-income people may be more reluctant to social isolation due to: family responsibilities as a food and wage provider; less flexible jobs; fear of losing their position and lack of formal jobs. [24] [25] [26] Digital technology can be a powerful tool to quickly provide valuable epidemiological information at a low cost. Disease tracking, contact tracing, diagnosis support, clinical status monitoring and telemedicine are among applications that have been implemented in different countries to fight against this pandemic. [4] The impact of the COVID-19 pandemic will be worse in low income settings with weaker healthcare systems, which will increase the world health disparities due to the effect of the negative social determinants. Despite the concerns on limited access to the internet, we demonstrated that low-cost digital initiatives coupled with effective social engagement can be extremely useful to provide important data on disease activity in a developing country with continental proportions. Our study has limitations that deserve attention. First, we used a web-based application that relies on spontaneous survey responses. Although access to the internet has been reported in 79% of households in Brazil, our sample was under-represented by elderly and lowincome individuals. [30] Appropriate weighing and calibration helped mitigate this limitation, as estimated demographics became similar to census-based population data. Second, most COVID-19 cases were defined by self-reported symptoms and history of contact with a suspected/confirmed case, which may overestimate the number of symptomatic infections and does not account for asymptomatic infections. Nevertheless, this strategy is a feasible approach to evaluate large population-based surveys, particularly in settings with insufficient tests for screening every suspected case, such as most regions in the world during this pandemic. Finally, a noteworthy aspect of this initiative was its collaborative and non-profit nature. Over 30 professionals and over 10 companies from various fields gathered resources and expertise to help brazilians stay aware of their surroundings while also contributing to All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint science. This study proves that it is possible to conduct nationwide high-level research in developing countries with limited resources. -Developed countries have successfully used digital technologies to gain insights about epidemiological and clinical data regarding the COVID-19 pandemic -There is limited evidence on the use of such technologies in developing countries. -More than 240.000 patients were assessed for COVID-19 symptoms and epidemiological data in Brazil using a low cost digital web-application. -It proved to be a safe and viable method in developing countries, such as Brazil, enabling insights on under notification and the impacts of wage on infection and testing. FUNDING: None. The study protocol was approved by the Ethics Committee from the Federal University of Paraná (CEP CHC/UFPR #35028620.0.0000.0096) as a safe design, offering no harm to any participant. All participants provided consent for non-commercial use of their data. We would like to thank Unimed Curitiba for the donation to support the web-based application. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. Figure 2A shows the relation between estimated/reported COVID-19 cases ratio and COVID-19 tests. Figure 2B shows the relation between estimated/reported COVID-19 cases ratio and Human Development Index. Figure 2C (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint income level. While the COVID-19 rates were inversely associated with income level, self-reported COVID-19 testing was considerably higher in the highest income stratum, as compared with all other income strata. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204917 doi: medRxiv preprint A B C D All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 2, 2020. Less than 1/2 1/2 to <1 1 to <2 2 to <3 3 to < 5 5 and over Seroprevalence of Antibodies to SARS-CoV-2 in 10 Sites in the United States Repeated population-based surveys of antibodies against SARS-CoV-2 in Southern Brazil. medRxiv Epidemiological and clinical characteristics of the COVID-19 epidemic in Brazil Pandemic influenza in Papua New Guinea: a modelling study comparison with pandemic spread in a developed country The COVID-19 Pandemic Effects on Low and Middle-Income Countries Impacts of COVID-19 on vulnerable children in temporary accommodation in the UK Geographic access to United States SARS-CoV-2 testing sites highlights healthcare disparities and may bias transmission estimates Redefining vulnerability in the era of COVID-19 Pandemic Influenza Preparedness and Response Among Public-Housing Residents, Single-Parent Families, and Low-Income Populations Effective health risk communication about pandemic influenza for vulnerable populations The data underlying this article are available in the article and in its online supplementary material. https://www.washingtonpost.com/opinions/2020/04/10/how-digital-data-All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.All rights reserved. No reuse allowed without permission.(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.