key: cord-1005717-milzvrjj
authors: Ferreira, C. P.; Marcondes, D.; Melo, M. P.; Oliva, s. M.; Peixoto, C.; Peixoto, P. S.
title: A snapshot of a pandemic : the interplay between social isolation and COVID-19 dynamics in Brazil
date: 2021-04-30
journal: nan
DOI: 10.1101/2021.04.29.21256267
sha: 803d67bc207a0ac792c7fd21c9f27e27393ea503
doc_id: 1005717
cord_uid: milzvrjj

In response to the COVID-19 pandemic, most governments around the world implemented some kind of social distancing policy in an attempt to block the spreading of the virus within a territory. In Brazil, this mitigation strategy was first implemented in March 2020 and mainly monitored by social isolation indicators built from mobile geolocation data. While it is well known that social isolation has been playing a crucial role in epidemic control, the precise connections between mobility data indicators and epidemic dynamic parameters have a complex interdependence. In this work, we investigate this dependence for several Brazilian cities, looking also at socioeconomic and demographic factors that influence it. As expected, the increase in the social isolation indicator was shown to be related to the decrease in the speed of transmission of the disease, but the relation was shown to depend on the urban hierarchy level of the city, the human development index and also the epidemic curve stage. Moreover, a high social isolation at the beginning of the epidemic relates to a strong positive impact on flattening the epidemic curve, while less efficacy of this mitigation strategy was observed when it has been implemented later. Mobility data plays an important role in epidemiological modeling and decision-making, however, we discuss in this work how a direct relationship between social isolation data and COVID-19 data is hard to be established. Understanding this interplay is a key factor to better modeling, for which we hope this study contributes.

importance of taking into account the socio-economical background of a region when implementing such 77 measures. Such a diverse response to the pandemic is this paper reason of being, since in a such heterogeneous 78 scenario a much needed assessment of the efficacy of the measures to stop the disease spread is not a 79 straightforward task, so a careful quantitative and qualitative analysis of the disease evolution and of measures 80 such as social isolation is needed to evaluate this efficacy. This paper aims to asses the effectiveness of human mobility control measures on reducing the COVID-82 19 spread in Brazil, taking into account many aspects of the locations and their population, such as urban 83 hierarchy, demography, and socio-economic profile. Our assessment focuses on the dataset containing daily 84 cases of Severe Acute Respiratory Illness, and a daily isolation index, of the most important cities in Brazil 85 in a period ranging from March 15 to October 30, 2020. We carefully dissect this dataset, coupling its 86 quantitative analysis with qualitative information about measures enforced to slow the disease spread and 87 characteristics of the population and locations around the country. With these analyses, we hope to give a 88 snapshot of how the pandemic evolved in Brazil, and study the interplay between social isolation and COVID-89 19 spread in the country. Here we discuss and quantify in detail the conditions under which social distancing, 90 measured through population local mobility, impact infection rates depending on different demography and 91 socio-economic profiles. As a result, besides providing on its own insightful knowledge of the behavior of the 92 pandemic in different conditions, our results allow models to be more precisely adjusted to take into account 93 local characteristics and provide more reliable epidemic scenarios. 94 In Section 2 we present the material and methods of the quantitative analysis, while in Section 3 we 95 present its results, and in Section 4 we discuss our findings.

. CC-BY-ND 4.0 International license It is made available under a 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 April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint whose notification is compulsory in Brazil. It has been a good thermometer to catch the disease spatio-136 temporal dynamics in the country, since it has struggled with an insufficient capacity of molecular diagnosis 137 and fast tests 21 . The information available does not distinguish between imported and autochthonous cases. 138 Although the first COVID-19 case occurred on February 25, the dataset used here ranges from March 15 until 139 October 30, 2020, since before that changes were not yet implemented on the surveillance system of SIVEP- 140 Gripe to identify COVID-19 cases among cases of other diseases such as Influenza, Respiratory Syncytial 141 virus, Adenovirus, and Parainfluenza 4 . This period comprises the first wave of the disease in Brazil during 142 which it is supposed the transmission of a unique variant of the virus in Brazilian territory. 143 A nowcasting procedure was performed, using the R package NobBS 30 , to correct delay in notifications 144 which in Brazil can take up to forty days. Since the nowcasted daily number of cases still presented weekly 145 variations, it was smoothed by taking a 7-day moving average. To compare the disease spread among the 146 cities, the daily number of cases was divided by 100,000 inhabitants. 147 The effective reproduction number R t was calculated using the nowcasted smoothed data of incidence. 148 For this, we considered the epidemiological model SEIR (susceptible-exposed-infected-recovered), and the 149 approach proposed by Wallinga and Lipsitch 36 . The parameters considered to calculate R t are the latent 150 period (η −1 ) of 3.0 days, the infectious period (τ −1 ) of 6.4 days, and the life expectancy in Brazil (µ −1 ) of 151 75 years. The rates of leaving the exposed and infectious classes are denoted by s 1 = η + µ and s 2 = τ + µ, 152 respectively. Therefore, the generation interval distribution g(t) is given by 1 153

After normalizing g(t) we can evaluate R t as

, and the daily transmission index is calculated as R t × τ . . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint is composed by: Belo Horizonte (MG), Curitiba (PR), Florianópolis (SC), Fortaleza (CE), Goiânia (GO), Florianópolis and Vitória the less populous with 1 and 1.8 million, respectively.

The second level, the Regional Capitals, are composed of 97 urban centers with regional influence, which 177 are divided in three groups: A, B and C. The Regional Capitals A is composed of 8 capital cities of states in the 178 Northeast and Midwest region, and the city of Ribeirão Preto (SP). These cities have a similar population 179 size, varying between 0.8 and 1.4 million inhabitants, and share a direct relationship with Metropolises.

There are 24 Regional Capitals B, ten of them in the South region, which have an average population of 181 530 thousand inhabitants, the most populous being São José dos Campos (SP) with 1.6 million inhabitants.

These cities are reference centers within their states, with the exception of state capitals Palmas (TO) and 183 Porto Velho (RO) which also exerts influence on other states. The Regional Capitals C are formed by 64 184 cities, 30 of them in the Southeast region, with average population size of 360 thousand inhabitants.

The Sub-regional Centers level is composed of 352 cities with less complex management activities than 186 Regional Capitals. The forth level, Zone Centers, is composed of 398 cities with less management activities 187 and, finally, the fifth level, the Local Centers, is composed of 4.037 cities, that is, 82.4% of urban centers of 188 Brazil with an average of 12.5 thousand inhabitants and whose influence is restricted to its own territorial 189 limits.

Socio-economic variables are also usually associated with disease spreading, and the Human Development In this section, we study the relationship between the daily social isolation index and the daily incidence of 209 the disease in the main cities that comprise the metropolises, which are São Paulo, Brasília, Rio de Janeiro, 210 Belo Horizonte, Curitiba, Florianópolis, Fortaleza, Goiânia, Manaus, Belém, Porto Alegre, Recife, Salvador, 211 Vitória, and Campinas. Similar to what was done for the daily incidence, the daily social isolation index 212 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint was smoothed by taking a 7-day moving average. Figure 2 shows the daily ratio between the social isolation 213 index, and the daily incidence, and their respective daily average among the cities considered. Spatial-214 temporal variations of both measures highlight regional and temporal differences resulting from geographic, 215 demographic, cultural and political characteristics of each main city of Brazil, whose behaviour affects other 216 cities in their region of influence. In general, the social isolation index was negatively related with the incidence 217 of the disease, but not linearly proportional. Besides, taking all cities together, the monthly average of the 218 social isolation index decreased with time: from March to October they were (average value and its standard 219 deviation), respectively, 0.21 ± 0.02, 0.22 ± 0.02, 0.18 ± 0.03, 0.13 ± 0.01, 0.13 ± 0.02, 0.11 ± 0.01, 0.09 ± 0.01, 220 and 0.08 ± 0.01.

In order to mitigate the effects of the disease on the healthcare system, lockdown strategies were imple-222 mented on some cities in the North and Northeast regions, causing the sharp increase on the social isolation 223 index measured in Belém, Fortaleza and Recife (see Figure 2 around May 10). In Belém, differently from 224 the other two cities, the lockdown can be clearly seen in the data since its social isolation index was below 225 average in the period leading up to the lockdown and then increased way over the average. For these cities, 226 the epidemic peak was from 2.5 to 4.2 times the observed average incidence on their respective days.

Among the cities of the Northeast region, there is Salvador, which seems to have kept a more efficient 228 control of the disease transmission, since it endured a higher isolation rate, and had a smaller peak, compared 229 to other cities of this region. On the other hand, there is Fortaleza, which has the lowest HDI among the 

In the Southeast region, although an increase on the social isolation index was observed in Belo Horizonte 239 at the end of the study period, it had a similar transmission pattern as Campinas, while Vitória, which is 240 among the cities with the highest HDI and lacks an international airport, had the best performance on disease 241 control transmission when compared to the other Southeast cities, what can be due to these important factors.

Among the South cities, despite Curitiba not having an international airport, its has worse performance on 243 the social isolation index when compared to Porto Alegre and Florianópolis, which may be behind its larger 244 number of reported cases. Still in the South, it is worth mentioning that not only Florianópolis has a high 245 HDI, but also a major part of its territory is an island, which could have contributed to its high performance 246 on controlling the disease spreading when compared to the other South cities.

When comparing both cities of the Midwest region, one sees that Goiânia had a social isolation index 248 pattern similar to Brasília, but its performance to contain the spread of the disease was worse. The fact that 249 Brasília has a higher HDI and lower population density may explain the low disease incidence compared to 250 the other cities. Another important factor is that the Midwest region annually records the lowest number of 251 influenza and other respiratory viruses cases inside the national territory 13 , so these cities might be prone 252 to have lower incidence of a respiratory disease. Lastly, Rio de Janeiro and São Paulo, both Southeast cities, CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint having an outbreak of COVID-19 always above the average. Table 1 shows different measures on the data 272 calculated before, during and after the lockdown period. This analysis was done considering a lag of seven 273 days (≈ incubation period) between the social isolation index and the daily incidence.

In all cases, the lockdown was effective on reducing the incidence (and R t values). This control method 275 was clearly efficient in São Luís, which was presenting an increase on the rate of transmission before lockdown 276 that changed to a decreasing pattern after it. Fortaleza, Belém and Recife were already diminishing the rate 277 of transmission before lockdown occurred. Among them, Fortaleza was able to keep it longer with a high 278 social isolation index. Although having a huge increase on social isolation index as a result of lockdown, 279 Recife was not able to keep it for long. As a consequence, the R t value after lockdown was similar to the one 280 measured before it. It is interesting to note that in Recife the social isolation index was already increasing 281 before lockdown.

For São Luís, the maximum incidence occurred six days before lockdown, and after lockdown disease 283 incidence still decreased until fifteen days after it ended. For Belém, the maximum incidence occurred sixteen 284 days before lockdown, and after lockdown the incidence still decreased until five days after it ended. For 285 Fortaleza, the maximum incidence occurred five days before lockdown, and after lockdown the incidence still 286 decreased until ten days after it ended. For Recife, the maximum incidence occurred eighteen days before 287 lockdown, and after lockdown the incidence still decreased until fifteen days after it ended. The highest 288 percentage on the reduction of the infection observed in São Luís can be related to its lower position on 289 the urban hierarchy when compared to the other three, and to its spatial geography (it is an island). All 290 these cities have public laboratories for the molecular diagnosis of SARS-CoV-2, but we can not compare the 291 diagnostic capacity of each one because this data is not publicly available. The lack of diagnostic capacity 292 can jeopardize control efforts by causing delay on decision making.

. CC-BY-ND 4.0 International license It is made available under a 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 April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint

The history of SARS-Cov-2 spreading on the State of São Paulo is characterized by different time and spatial 295 scales. Starting from the regional centers, the disease displaced to municipalities with major connections, after 296 to municipalities with minor connections, and lastly to rural municipalities (spatial pattern well explained 297 by the gravity model). Locally, the spread was by contiguity (diffusion model) 16 . As the disease spread, 298 following the example of the metropolitan region of São Paulo, the inner state adopted strong mitigation 299 measures closing schools, universities, and all its trade, keeping only essential services such as pharmacies, 300 supermarkets, and hospitals open. This delayed the arrival of the virus to the inner state by at least one 301 month, letting the cities prepare their healthcare systems, which are more fragile in this region.

On May 27, São Paulo State started to move out from a restrictive quarantine with a flag system that 303 classifies, based on several indicators, the disease transmission risk, and the probability of break-down of the 304 healthcare system. Five colors were adopted: (i) red (phase 1) is considered a contamination phase and only 305 essential services are permitted; (ii) orange (phase 2) is considered an attention phase, with the possibility of 306 some services opening (limit of 20% of its capacity and maintaining all specific hygiene protocols); (iii) yellow 307 (phase 3) is considered a controlled phase, with some flexibilization (limit of 40% of its capacity and main-308 taining all specific hygiene protocols); (iv) green (phase 4) is a partial opening phase, in which all services are The variance observed on the transmission rate can be related to the population density in each city which 327 is respectively 7398.3, 1359.6, and 201.2 inhab/km 2 at São Paulo, Campinas, and Votuporanga. Although 328 the incidence was higher at Votuporanga, the average absolute number of cases was 3.4 (from 1 to 11), 36.3 329 (from 9 to 95), and 422.8 (from 161 to 986) in Votuporanga, Campinas, and São Paulo, respectively. In this section, we study the relationship between social isolation index and how fast the disease spread over 332 different geographic locations around the country. The enrolled cities satisfy one of the following criteria: (i) 333 . CC-BY-ND 4.0 International license It is made available under a 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 April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint is the capital of a state, (ii) is the first or second city of a state with a higher number of cases, (iii) belongs 334 to São Paulo state and is among its cities with highest number of cases. 335 For each municipality, the smoothed curve of daily cases in the upward phase, that is, from March 15 336 until the day when the maximum number of cases occurred, was considered. Then, the upward phase was 337 divided into seven stages in the following way. First, we calculated the total number of infected people in this 338 period (upward phase), divided it in half and marked the day in which such a number of cases was reached 339 as the start of stage 7, which ends when the peak of cases is reached. Therefore, stage 7 represents the 340 period contemplating the second half of the cases until the peak. Analogously, new successive divisions were 341 made, minding the day when a corresponding half of accumulated cases was reached. In total, six divisions 342 were made, creating 7 stages. In this way, stage 1 refers to period between March 15 and the occurrence of 343 1.5625% of the cases in the upward phase; stage 2 refers to period between 1.5625% and 3.125% of it; and 344 stages 3 to 6 correspond to periods between 3.125%-6.25%, 6.25%-12.5%, 12.5%-25%, 25%-50%, respectively, 345 of cases in the upward phase (see Figure 5 ).

The length of these stages is an estimate of the speed of disease spreading. When efforts to stop or 347 diminish the disease spread are implemented in one stage, one expects an increase in the length of later 348 stages. Hence, finding a relationship between the efforts to control the disease spread, such as the social 349 isolation index, and the length of these stages could be an efficient way of assessing the efficacy of the control 350 measures.

In Table 2 we present, for each municipality (a total of 31 analyzed), the median of the social isolation 352 index and the length of the disease spreading stages. The incidence was associated to the social isolation 

Comparing the length of stages 5, 6, and 7 with their respective medians, we see that 24 municipalities 358 (77.4%) maintain the same behavior, that is, they have the length of stages 5, 6, and 7 above the median or 359 below the median in all three stages. Also, the correlation coefficients between the length of these stages are 360 0.773 between stages 5 and 6, 0.691 between stages 5 and 7, and 0.849 between stages 6 and 7. Due to the 361 similarity of these stages, plus the fact that they correspond to 87.5% of all cases in the upward phase, our 362 analysis will be restricted to stage 7. 363 Table 3 shows that a high social isolation index in stages 3 and 4, can slow the transmission of the disease 364 in stage 7. From these 16 cities with length at stage 7 above the median, 12 (75%) of them had isolation 365 above Q3 in stages 3 or 4. The four cities that did not had isolation above Q3 in these two stages were: 366 Belo Horizonte, Guarulhos, Jundiaí and São Paulo. In the case of São Paulo, the length of the initial stages 367 of disease spreading in this city was extremely short, indicating a fast initial dispersion of the disease. A 368 social isolation index above Q3 was only observed in stages 5 and 6. This late high social isolation index may 369 decrease the speed of disease, leading to the length of the final stages above of median. The city of Jundiaí 370 had social isolation index above Q3 only in stages 1, 6 and 7, but presented a social isolation index above 371 median in stages 2, 3 and 5. Therefore, except for stage 4, which was short (7 days), Jundiaí presented a 372 social isolation index above the median in all the stages, which may have contributed to the reduction of the 373 speed of disease transmission. On the other hand, we have the city of Guarulhos, that had a social isolation 374 index above or equal the median in the stages 4, 5 and 6, and Belo Horizonte that had social isolation index 375 . CC-BY-ND 4.0 International license It is made available under a 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 April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint above of median in stage 4 and really close to the median in stage 5, which may have contributed to a 376 duration superior than the median in the stage 7. value is equivalent to 1000 accumulated cases in this city. In this analysis, we consider all cities not under 400 the direct influence of São Paulo † , with ∆ 7 above or equal to the median, and social isolation index above 401 Q3 in stages 3 or 4 (see Table 3 ), which were the cities with earlier high isolation, and a slower stage 7, which 402 could be caused in part by this high isolation.

As an example, we show in Figure 6 the daily incidence for the city of Porto Alegre where the days in The dispersion plot between the daily incidence and social isolation, colored according to the increase or 415 decrease in cases, is presented in Figures 7 and 8 . Five of those cities are Metropolises, three are Regional 416 † Hence we excluded Jundiaí and São Paulo from this analysis.

. CC-BY-ND 4.0 International license It is made available under a 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 April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint Capitals A and four are Regional Capitals B. A similar behavior was observed for the cities considered which 417 have implemented some control measures in stages 3 or 4 (section 3.4) and are not under the influence of 418 a Metropolis that did not do some mitigation control in these stages. For all these cities there seems to be 419 a threshold, which may vary from city to city, where a decrease on the isolation index triggers an abrupt 420 increase on the daily incidence. This phenomenon is characterized by the L-shape of the dispersion between 421 the isolation index and daily incidence, where the horizontal part of the L refers to a period when higher 422 isolation may have held the incidence low, and the vertical part of the L refers to a period when either the 423 variation on the isolation does not affect the incidence or the decrease on the isolation below the threshold 424 was followed by an explosion of cases. Analyzing the results of each hierarchy, it seems that the number of intervals increases as the HDI 452 increases (correlation coefficients 0.90, 0.98 and 0.76 for Regional Capitals A, Regional Capitals B and 453 Metropolises, respectively). Although there seems to exists a threshold of the social isolation index below 454 which an explosion of cases occurs, its value vary among the cities. Therefore, besides human mobility, other 455 factors are certainly influencing the velocity of the disease's spread.

. CC-BY-ND 4.0 International license It is made available under a 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 April 30, 2021. In this work, we explored some variables that can be responsible for the failure or success of the mitigation 458 strategies to halt the spreading of the COVID-19 epidemic in the Brazilian territory. In particular, we 459 focused on social isolation, since during this epidemic period it was one of the few strategies available, 460 besides prophylactic measures such as the use of masks and personal hand-hygiene. Our analysis sought to 461 associate two sets of temporal series, the social isolation index, and the incidence, exploring the relationship 462 between these two data sets in different cities in Brazil looking for patterns and what can explain them. 463 The first thing that we could notice is that there is no direct and simple relationship between the social 464 isolation index and the incidence. In fact, the evolution of the disease is driven by multiple factors. Besides sharply increased in such a way that high isolation measures did not work anymore.

There are some limitations related to the data used here. The bias of the daily isolation index is not 474 controlled. On top of that, the daily incidence data cannot capture the asymptomatic cases. Despite these 475 limitations, we successfully pointed out some patterns related to these data. We also could hypothesize that Tables  Table 1 Percentage variation of the social isolation index and the transmission index per period. From the first to the fourth, period means week. In the case of Fortaleza and Recife, the fifth period is also a week. For all of them, the second week corresponds to the lockdown. In all cases, the last period is counted until the first minimum of the incidence curve after the lockdown period is finished. A lag of seven days between the isolation index and the transmission index was considered. Table 2 Median of social isolation index and length of the stages (∆ i with i = 1, 2, 3, 4, 5, 6, 7) of the upward phase for each municipality. The stage 1 refers to period between March 15 and the occurrence of 1.5625 % of the cases in the upward phase and the stage 2 refers to period between 1.5625% and 3.125% of it. The stages 3 to 7 correspond to periods between 3.125%-6.25%, 6.25%-12.5%, 12.5%-25%, 25%-50% and 50%-100% of cases in the upward phase, respectively. . CC-BY-ND 4.0 International license It is made available under a 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 April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint Table 3 Distribution of municipalities according to social isolation index and disease transmission speed. The stage 1 refers to period between March 15 and the occurrence of 1.5625 % of the cases in the upward phase and the stage 2 refers to period between 1.5625% and 3.125% of it. The stages 3 to 7 correspond to periods between 3.125%-6.25%, 6.25%-12.5%, 12.5%-25%, 25%-50% and 50%-100% of cases in the upward phase, respectively. ∆ 7 is the lenght of stage 7 and Q3 is the third quartile. Fig. 2 In (a-c), (g-i) the daily ratio between the social isolation index and its respectively daily average. In (d-f) and (j-l) the daily incidence and its respectively daily average. The average was calculated considering the main cities that represent each

Metropolis. The date corresponds to the day of the first symptoms. The dashed horizontal line shows when this ratio is equal to 1.

. CC-BY-ND 4.0 International license It is made available under a 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 April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint Fig. 5 Temporal evolution of the number of cases for the municipality of Aracaju. The date corresponds to the day of the first symptoms, measured from (t=0) March, 15. The stage 1 refers to period between March 15 and the occurrence of 1.5625 % of the cases in the upward phase and the stage 2 refers to period between 1.5625% and 3.125% of it. The stages 3 to 7 correspond to periods between 3.125%-6.25%, 6.25%-12.5%, 12.5%-25%, 25%-50% and 50%-100% of cases in the upward phase, respectively. Fig. 6 (a) Temporal evolution of the daily incidence and (b) the dispersion between the daily incidence and the isolation index for the city of Porto Alegre. The colors refer to periods of days where the incidence increases or decreases for at least three in a row. The downward phase of the incidence curve is in black.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 30, 2021. ; https://doi.org/10.1101/2021.04.29.21256267 doi: medRxiv preprint

On the convolution of exponential distributions