Results

The officially reported COVID-19 cases and deaths timeline are shown in Figure 1. Considering the total of inhabitants, the number of cases and deaths can be considered small. However, after saturation of daily cases in the third week of April, a new rise in the curve of infected cases at the end of April can be observed. The same behaviour was observed in the curve of deaths about 25 days later. Possibly, these growths could be attributed to the Social Distance Index, shown in Figure 1 (c). During this period infections and deaths increased, meanwhile a decline in social isolation was observed. A sum of two Gonpertz functions was necessary to appropriately fit both datasets over the whole timeline. The fitted curves have been extended to show the tendency to saturation of the number of infection cases and deaths by the end of the year. This model predicts a total number of 1.182.657 infected people and 15.703 deaths. The daily projected cases and death curves (first derivatives of the fit curves) clearly show the initial saturation of cases and deaths curves, followed by a strong and broad peak. The advantage of using the daily projected cases and death curves is to avoid strong oscillations (with approximately a 7-day period) present in the daily reported new deaths and infected cases data. This model projects a peak of 102 daily deaths on September 8th, while the infected cases peak of 6047 daily cases will be reached on September 30th. The case fatality rate (CFR) was calculated as the ratio between the cumulative number of deaths and the cumulative number of infected cases. The evolution of social distancing (InLoco, 2020) is shown in Figure 1(c). Our model describes well the pandemic evolution, nowcasts and forecasts. However, due to its simplicity, the total number of projected deaths and infected cases should be considered cautiously and depends on the evolution of social distancing. The values of the fitted parameters are shown in the supplementary information.
Another scenario arises when the excess of deaths during the COVID-19 pandemic is considered. The standard approach to calculate excess mortality uses the previous year’s historical all-deaths data as reference (Nogueira et al., 2005). However, in the current pandemic, with social distancing and closing of non-essential economic activities, the situation is not comparable with any previously historical mortality series reported so far. The total deaths in Minas Gerais between 2008 and 2014 increased 21% and then varied significantly between 2015 and 2018 (18% of variation), making the use of any kind of historical series unapproachable. In addition, transport accidents and deaths by assaults two of the most important mortality causes in Minas Gerais, should heavily decrease during this period because of the reasons mentioned above.
Figure 2 shows a comparison between the different causes of death between 2019 (dashed lines) and 2020 (solid lines) in Minas Gerais (Cartorio, 2020). It is worthy to note that until the 14th epidemiologic week of 2020, when the epidemic of COVID-19 started in Minas Gerais, the weekly number of deaths for all causes in 2020 oscillated similarly to the ones observed in 2019. However, beyond that week most of the 2020 causes of deaths, except for SARS, fall systematically below the values of 2019, probably as an effect of social distancing measurements. We interpret the sharp drop after week 25th, marked by a grey area, as a four to five weeks delay to update the RO database. Therefore, to calculate the excess deaths in 2020 regarding data from 2019, we should consider only until the 25th epidemiologic week of 2020 and only COVID-19 and SARS death causes. This excess is calculated as follows: from the total number of deaths caused by COVID-19 plus SARS in 2020, the number of deaths caused by SARS in 2019 is subtracted. We assumed that all other causes of death were affected by social distancing and decreased economic activities and were therefore not considered in the excess deaths calculation.
The excess deaths (open black squares) calculated according to the above developed methodology is presented in Figure 3. Since data have been harvested from 2 different public platforms (Cartorio, 2020; SUS, 2020), the reported deaths by COVID-19 in both platforms have also been plotted. Both platforms use different approaches to compile the number of deaths caused by COVID-19. Whereas the governmental data (open red circles) is collected by reports from health surveillance, RO collects the number of deaths (open blue triangles and open black squares) from registered death certificates. Therefore, not only the number of deaths attributed to COVID-19 will be different, but also the time delay in updating both databases. The sum of two Gompertz functions was used to well fit all datasets. The values of the fitted parameters can be found in the supplementary information. The excess deaths curve modelling predicts a peak of 135 daily deaths on September 22nd, two weeks later, and 32 % higher than that predicted by the governmental data. The difference between the RO and the governmental data is due to COVID-19 deaths sub-notification. The upper bound of the sub-notification can be estimated by the difference between excess deaths and the governmental reported deaths, while the lower bound of the sub-notification can be estimated by the difference between RO and governmental reported deaths. The time evolution of the sub-notification is shown in Figure 3(b).
As shown above, untested COVID-19 deaths can be erroneously diagnosed and reported as SARS deaths. We suppose the same problem is happening for the diagnosis of moderate infection cases. We consider as moderate infection cases those that were erroneously diagnosed as SARS cases but did not result in death. Therefore, the same methodology applied to estimate the sub-notification of COVID-19 deaths will be used to estimate the sub-notification of COVID-19 moderate cases. Figure 4 (a) shows the evolution of cumulative weekly SARS cases (Infogripe, 2020) in 2020 (black solid squares) and 2019 (red solid circles). Interestingly, during the first 9 weeks of the year, the number of cases reported in 2020 and 2019 were similar. However, as soon as the COVID-19 epidemic started in Minas Gerais, end of week 9th, the number of SARS in 2020 cases heavily increased compared to 2019. Since there were no reported additional epidemics related to respiratory symptoms apart from COVID-19 as well as those of every year, we interpret that difference as non-diagnosed cases of COVID-19.
Figure 4(b) shows the evolution of COVID-19 excess cases (green open stars), calculated as the difference between SARS and COVID-19 cases in 2020 related to 2019 SARS cases. For comparison, the cumulative weekly number of reported COVID-19 cases (blue open triangle) is also plotted. The excess COVID-19 cases dataset has been well fitted by the sum of two Gompertz functions (red solid line). The officially reported COVID-19 cases data have also been well fitted by a sum of two Gompertz functions (blue solid line). The fitted parameter values are shown in the supplementary information.
The sub-notification of COVID-19 cases upper bound can be calculated now as the difference between the excess cases and governmental reported cases and is shown in Figure 4(c). An estimation of CFR upper bound can now be obtained considering the excess of deaths and the excess of infected cases, as shown in Figure 4(c).
Ethnic discrepancies impact health care access in Minas Gerais where self-declared black and brown-skinned people have lower incomes and less access to health assistance than self-declared white people (IBGE, 2020). Herein, we include the impact of COVID-19 in both ethnics separating the data presented in Figure 1 by self-declared skin colour or ethnicity of the patients. We have considered as black patients the sum of all self-declared black and brown-skinned patients, while white patients correspond to self-declared white only. To calculate the percentage of cases and deaths from black persons we have also considered patients that were not classified in both groups. In this case the availability of data separated by self-declared skin colour was restricted up to June 24th.
Figure 5(a) shows the cumulative number of infected cases evolution, separated by black (open black squares) and white patients (open blue circles). A small increase of cases of black patients can be observed. The modelling of the datasets (solid lines) predicts an increase of such difference towards the end of the year. The percentage progression of infected black patients (regarding the overall patients), as calculated from the modelling results, is shown by the dashed red line. This number oscillates between 60% and 55% from the beginning of the epidemic towards the end of 2020.
The cumulative deaths progression is shown in Figure 5(b). A small increase of deaths amongst black patients (open black squares) can be observed in recent days. The modelling of the datasets (solid lines) also predicts an increase of such difference towards the end of the year. The percentage change in deaths in black patients (regarding the overall number of patients), as calculated from the modelling results, is shown by the dashed red line. This number rises from around 27% at the beginning of the epidemic to a saturation of about 54% at the end of 2020.
The CFR of both populations can be calculated from these modelled results. Figure 5(c) shows the timeline of CFR for each population. As can be seen, a saturation of about 4.5%, for black patients was predicted while for white patients an increase of 8.2% was calculated towards the end of 2020.
Demography has been also impacting COVID-19 pandemic in Minas Gerais. Then, in figure 6 we the number of cases and deaths (same data as in Figure 5) per 100.000 inhabitants separated by municipality and self-declared skin colour.
Interestingly, the distribution of cases and deaths is non-uniform across the state. There are municipalities with a higher number of cases and a lower number of deaths. Most cases and deaths per 100.000 inhabitants are not only concentrated in the capital, but in cities closer to the state borders with interstate connexions. There are also strong differences in locations with more cases and deaths in black and white patients. The number of cases and deaths in black patients is somehow better distributed along the territory, with a large incidence in the northern half. However, the cases and deaths in white patients are concentrated in the southern half, which is economically more developed and populated.