We use data from the Brazilian Ministry of Health’s database, DATASUS, which is publicly available online (https://opendatasus.saude.gov.br/dataset). The Ministry of Health, through the Health Surveillance Secretariat (SVS), has been developing surveillance for specific respiratory diseases in Brazil, including, since 2009, for Serious Acute Respiratory Syndrome (SARS) due to the Influenza A (H1N1) pandemic. SARS was incorporated into the surveillance network for influenza and other respiratory viruses, and, recently (2020), COVID-19 was also included in the network.

We collected the information on April 10, 2021, when Brazil had registered 351,334 deaths from and 13,445,006 cases of COVID-19. In addition, we made a break in the data in order to understand and compare the age structure of deaths from COVID-19 in the two waves observed in the country. In this study, the first wave started on February 24, 2020 (first case registered in the country), reached its peak between the months of May and July 2020, and ended on October 31, 2020. The second wave began on November 01, 2020, and ended on April 10, 2021.

The original data are available at the individual level (case by case) and by municipality. The main limitation in using municipal-level data in Brazil is that the numbers of cases and deaths in each municipality may be small, given the limited number of people exposed to the risk of developing the disease in any given area, which may, in turn, lead to many random fluctuations in the estimates. To avoid such problems, while also pursuing our goal of analyzing and understanding regional variations, we aggregated municipalities into 137 comparable small areas (Annex 1) using the IBGE definition of geographic meso-regions. These geographical areas are statistical constructions that are aggregated and defined by regional and socioeconomic similarities, and that have not changed their boundaries over time. In addition, they have been used elsewhere (Baptista and Queiroz, 2019a,b; Baptista et al., 2021; Lima and de Queiroz, 2014; Lima et al., 2021a). We also produced estimates using standardized rates to enable us to compare COVID-19 mortality levels, thereby eliminating the effects of the population age structure (Dowd et al., 2020).

In this study, we analyze the regional disparities in COVID-19 mortality based on the spatial distribution of proportional deaths from this illness, disaggregated by age group and across 137 small areas of Brazil. We also compared the overall mortality levels, by age group and sex, in the two waves of the analysis to provide a descriptive view of the pandemic in Brazil over the course of the pandemic, and to contribute to the discussion on excess mortality (Lima et al., 2021b) and on the impact of COVID-19 on life expectancy (Castro et al., 2021).

One important measure of the dynamics of the pandemic is the severity of infection. In general, examining fatality rates can improve our understanding of the severity of the disease, help us identify the population at risk, and give us some idea of how the health care system is dealing with the pandemic. There are two main measures used to investigate the effects of a disease (Green et al., 2020; Kelly and Cowling, 2013; Spychalski et al., 2020). The first measure is the infection fatality ratio (IFR), which refers to the ratio of deaths among the entire infected population. In other words, the IFR can be defined as the number of COVID-19-associated deaths divided by the total number of infections. The second measure is the case fatality ratio (CFR), which is the ratio of deaths among confirmed cases (Dudel et al., 2020; Kelly and Cowling, 2013; Sánchez-Romero et al., 2021). However, a limitation of using the CFR to measure the risk of COVID-19 mortality is that the ratio is influenced by the number of people who receive the proper diagnosis. In other words, in countries with very few tests, we may observe higher CFRs because only people who have been admitted to hospitals or who have severe symptoms have been tested. In addition, we are counting deaths at a specific point in time, but some individuals with the disease might have a positive or a negative COVID-19 test outcome.

In contrast to the case fatality ratio (CFR), the IFR is not based only on the number of confirmed cases, and should therefore not be biased by potential shifts in testing policies, although it still has some limitations. One advantage of the IFR is that it incorporates asymptomatic and undiagnosed cases. However, the biggest problem that arises in determining these ratios is accurately establishing the numbers of cases (symptomatic and total) and deaths (Kelly and Cowling, 2013). As our current data do not provide enough information to enable us to estimate IFRs, we did not pursue analyses using this indicator. Unfortunately, Brazil’s COVID-19 testing rates are among the lowest in the world, albeit with considerable regional heterogeneity. While a few previous studies have estimated IFRs in the country based on seroprevalence surveys, they did not cover the areas that are the focus of this paper (Marra and Quartin, 2021).

We used the approach proposed by Kashnitsky and Schöley (2018), later detailed by Schöley (2021), to investigate the spatial variation in deaths from COVID-19 in Brazilian meso-regions. We map the deaths from COVID-19 by age group (25–59, 60–79 and >80 years) using ternary color-coding. This visualization technique maximizes the amount of information conveyed by colors. It works by expressing the relative shares among three parts – in our case, among three age groups – as the mixture of three primary colors (we define yellow as the primary color for the 25–59 age group, cyan as the primary color for the 60–79 age group and magenta as the primary color for the 80+ age group). In other words, ternary color-coding is designed to visualize proportions of a whole; that is, anything that splits into three non-negative parts that add up to a unified whole. This approach is perhaps its biggest limitation. A second problem can occur when the data are unbalanced; that is, if the observations are concentrated in one specific age group.

In this study, we find that COVID-19 deaths are concentrated at ages 40 and older; that is, that there is little variation with regard to the visual reference point, which is the threshold that marks perfectly balanced proportions (Baptista et al., 2021; Kashnitsky and Aburto, 2019; Schöley, 2021). In addition, we have also limited the lower bound age group, which starts at age 25 instead of at zero years old. Therefore, we have changed the age point of reference to the location of the average structure of COVID-19 mortality in Brazil, and have thereby visualized the direction and the magnitude of the deviations from that average.

Figure 1 shows the age-specific mortality rates for COVID-19. The goal here is to show the mortality differences by age and sex in Brazil. We opt to present this measure instead of the more traditional CFR, because, as we mentioned above, the latter is not an appropriate measure of COVID-19 mortality risk, since it is influenced by the number of people who receive the proper disease diagnosis. In the case of Brazil, we find that the age-specific mortality rates for males are higher than those for females in all age groups, in line with patterns observed in many other countries. However, for younger adults, the mortality risks in Brazil are shown to be greater than those observed in other countries.

Other studies have also investigated the impact of the pandemic using other measures. According to Lima et al. (2021b), Brazil had excess mortality of around 19% during the pandemic compared to previous years, but with large regional variation, with the less developed Northern part of the country experiencing much higher excess mortality than the more industrialized Southern and Southeastern parts of Brazil. In the same vein, Castro et al. (2021) estimated the impact of the pandemic on life expectancy at birth and at age 65 in Brazil. The results showed a reduction of about 1.3 years in life expectancy between 2019 and 2020, but with greater declines in the Northern than in the Southern states of the country.

Table 1 shows summary results for the country and its main regions. The distribution of COVID-19 deaths by age in Brazil follows a pattern similar to that observed in other countries. For males, we find that around 28% of deaths are of individuals aged 25 to 60, about 50% are of individuals aged 60 to 79 and 22% are of individuals aged 80 or older. For females, these values are 24%, 48% and 28%, respectively. In addition, there is some variation across regions of the country, which we will highlight later in the spatial analysis. One interesting result is that the distribution of deaths differs slightly by sex. For example, more of the female than the male deaths are concentrated at older ages, while slightly more men than women died at young adult ages.

Despite its limitations, in Table 1, we also show estimates of case fatality ratios by region of the country. The overall CFR in Brazil is 0.0280, ranging from 0.0345 in the Southeast to 0.0223 in the South. Recent data indicate that compared to all other countries in South America, Brazil has the highest CFR. Moreover, when we compare age-standardized measures for Brazil with those for 178 other countries in the world with available data, we find that the mortality levels in Brazil are greater than in 90% of these countries.

Additionally, we observe an interesting gender pattern of COVID-19 mortality in Brazil. During the first pandemic wave, the male-to-female ratio of deaths was 1.36. This means that there were 136 male deaths to every 100 female deaths. We also observe variations in this measure across regions. This ratio ranges from 1.43 in the South to 1.31 in the Center-West and Southeast. When we compare the first and the second waves, we see a decline in the gender differences in COVID-19 mortality for all regions of the country. For Brazil as a whole, the ratio went from 1.36 to 1.25 between the first and the second pandemic waves. Despite this reduction, males still had higher mortality levels than females, as shown in Figure 1. This gender differential in COVID-19 mortality has also been reported in other studies (Souza et al., 2020). Moreover, it can be argued that pre-existing gender gaps in mortality, such as those caused by the higher risks of external causes of death in males, were present even before the virus affected the population. In addition, in Brazil during the pandemic, males have accounted for almost 60% of deaths and 53% of hospitalizations related to COVID-19, and for over 70% of deaths and 40% of hospitalizations related to respiratory diseases besides COVID-19 (SARI) (Souza et al., 2020). In general, males in Brazil have higher mortality rates than females, and they have significantly higher rates of death from external causes. In the first stages of the pandemic, it was observed that mortality rates for younger adults declined due to changes in external causes of death (Santos et al., 2021). In other countries and regions, for example, the risk of dying was found to be higher for males than for females, but with a decline above age 80 (Ahrenfeldt et al., 2021). Aburto et al. (2021) also investigated the impact of the pandemic on life expectancy at birth and at age 60 for a series of countries. The study showed that the impact of the pandemic on life expectancy was greatest for males. This cause of death has been largely responsible for the elevated mortality risk among males during the pandemic, and explains a considerable share of the gender differentials in mortality in Brazil (Aburto et al., 2021). However, other empirical evidence indicates that that the restrictions imposed to limit the spread of COVID-19 have reduced external causes of death during the pandemic (Santos et al., 2021). Hence, it is possible that the new disease was responsible for causing another mortality gender gap.

Figure 2 (females) and Figure 3 (males) show the proportional distribution of COVID-19 deaths by age group and pandemic wave across Brazilian meso-regions. Before presenting the main findings, we show an example of how to interpret the ternary color-coding that uses females (Figure 2) for the first wave (left) as an example. Each point within the triangle represents a meso-region. The reading on the percentage of deaths from COVID-19 observed in each age group and meso-region occurs in a clockwise direction. Therefore, in this study, the percentage of deaths in the 25–59 age group can be read on the left side of the triangle; the percentage of deaths in the 60–79 age group can be read on the right side of the triangle; and the percentage of deaths in the 80+ age group can be read at the bottom of the triangle. Taking the visual reference point (point of intersection of the three lines within the triangle) as an example, which is the average structure of COVID-19 mortality in Brazil, the observed percentages are 24.5, 46.4 and 29.2 for ages 25–59, 60–79 and 80+, respectively. The colors represent the direction and the magnitude of the deviation from the average distribution of COVID-19 mortality in Brazil by age group. Yellow, cyan and magenta represent, respectively, a higher-than-average share of COVID-19 deaths in the 25–59, 60–79 and 80+ age groups. The saturation of the colors expresses the amplitude of the deviation, with perfect gray indicating a region that has an age distribution of mortality composition equal to the Brazilian average (Kashnitsky and Schöley, 2018).

The overall results indicate that, for both sexes, the percentage of deaths in the 60–79 age group is higher in almost all meso-regions in both waves. When we compare males and females, we see that in the 25–59 and 60–79 age groups, there are more meso-regions in which the number of male deaths from COVID-19 is higher than the national average than in which the number of females deaths from COVID-19 is higher than the national average; while in the 80+ age group, the opposite pattern is observed.

Another important finding is that the gap has narrowed between the two waves studied. That is, whereas in the first wave, the number of deaths from COVID-19 among men in the 25–59 age group was higher in approximately 83% of the meso-regions; in the second wave, this share had declined to ∼67%. In addition, it is not possible to observe a clear spatial pattern (or cluster) between the meso-regions for both sexes, although the results confirm that the ternary compositions are more spread out for men, especially in the first wave, since the data are more balanced. These overall results provide us with some clues that will be explored in more detail below.

A more specific analysis of the spatial variation by sex shows that there is no clear trend in deaths from COVID-19 for females (Figure 2). Roughly speaking, when we compare the two waves, we see that in the most developed region, the Southeast, the proportion of deaths from COVID-19 increased in the 60–79 age group; and, consequently, the proportion of deaths decreased in the 25–59 and 80+ age groups. In the Center-West region, mortality increased in the 60+ age groups and decreased in the younger age groups (25–59). On the other hand, when we look at the proportion of deaths from COVID-19 in the meso-regions in the North, Northeastern and Southern regions, we see no clearly defined patterns, as the spatial variation appears to be quite heterogeneous.

For males (Figure 3), we also observe no clear trend in deaths from COVID-19. In the Southern and Southeastern regions, the proportion of deaths from COVID-19 in the 60–79 age group increased, and the proportions of deaths in the 25–59 and 80+ age groups decreased. In the latter case, this may already be a reflection of vaccinations, which, in Brazil, followed age criteria, and occurred more quickly in precisely these regions. In the other regions, the spatial variation observed is again quite heterogeneous.