key: cord-0781057-061ci686
authors: Silva, R. A. d.; Ferreira, L. P. d. S.; Sampaio Leite, J. M. R.; Tiraboschi, F. A.; Valente, T. M.; Roda, V. M. d. P.; Sanchez, J. J. D.
title: Statistical Modeling of deaths due to COVID-19 influenced by social isolation in Latin American countries
date: 2021-04-07
journal: nan
DOI: 10.1101/2021.04.05.21254941
sha: a2e1fda21db925b066f9b83f090a8ecbf6ef524e
doc_id: 781057
cord_uid: 061ci686

Social isolation is extremely important to minimize the effects of a pandemic. Latin American (LA) countries have similar socioeconomic characteristics and health system infrastructures. These countries face difficulties to deal with the COVID-19 pandemic and some of them had very high death rates. Government stringency index (GSI) of twelve LA countries was gathered from the Oxford COVID-19 Government Response Tracker (OxCGRT) project. GSI was calculated considering nine metrics such as school and work closures, stay-at-home requirements, among others types of social distancing and isolation measures. Population data from the United Nations Population Fund and number of deaths data was collected from the dashboard of the World Health Organization (WHO). We performed an analysis of the period March-December using a mixed linear model approach. Peru, Brazil, Chile, Bolivia, Colombia, Argentina and Ecuador had the highest death rates with an increasing trend over time, while Suriname, Venezuela, Uruguay, Paraguay and Guyana had the lowest ones, which remained steady. GSI in most countries followed the same pattern during the analyzed months. i.e., high indices at the beginning of the pandemic and lower ones in the last evaluated months, while the number of deaths increased over the whole period. Almost no country kept its GSI high for much time, especially from October to December. Time and GSI as well as their interaction were highly significant. As their interaction increases, death rate decreases. In conclusion, our statistical model explains and substantiates the need for maintaining social distancing and isolation measures over time during the pandemic.

The COVID-19 pandemic has affected healthcare systems and caused collapses across the 53 globe. In Latin America (LA), the first case of SARS-CoV-2 infection was recorded on 54 February 25th in the City of São Paulo. In less than a month after the first case, all LA countries 55 had confirmed cases of COVID-19 1,2 . 56

The LA region has several obstacles that make it difficult for countries to take action 57 against the spread of the virus. Precarious conditions, such as poverty, lack of hospital 58 infrastructure, low sanitary conditions, high prevalence of chronic diseases and government's 59 tardy responses are factors that make it difficult to prevent contamination by the virus, so that 60 they facilitate transmission and directly impact the hospital system 3-5 . Through predictive 61 models' studies, it has been suggested that the virus could spread aggressively through LA 6,7 . 62 Moreover, analyses of the initial cases of the COVID-19 pandemic in LA estimated an 63 unfavorable scenario for the countries, and also evidenced aggressive dynamics of the disease 64 outbreak in Brazil and Ecuador compared to Italy and Spain 7 . Above all, among the LA 65 countries, Brazil was considered a major epicenter of the disease 8 . 66

Although there are measures aimed at reducing the spread of the new coronavirus such 67 as social distancing, school closures, cancellation of public events and, sometimes, severe 68 methods such as lockdown, these measures have been relaxed, in addition to noncompliance by 69 the population and poor governmental management 9 . Considering that social distancing and 70

isolation are important protective measures for the containment of SARS-CoV-2 infection, and 71 that there is lack of studies demonstrating the relationship between the social isolation and death 72 rate due to COVID-19, based on the Government Stringency Index (GSI) from The Coronavirus 73

Government Response Tracker (OxCGRT) 10 , the objective of this work was to analyze the 74 relationship between GSI and time, and the death rate from COVID-19 in 12 LA countries, 75

using a mixed linear model approach. In the model, the variables time (month) and GSI were considered both as fixed. In 111 addition, the country was included as a random effect. All statistical analysis was performed 112 under the most commonly used significance levels (1%, 5% and 10%) using the RStudio 113 statistical software v.3.6. 114 115

Results 116 117

We analyzed data of deaths related to COVID-19 in twelve LA countries in order to evaluate the 118 relationship between death rates, government stringency index and time progression. In this 119

context, time and isolation index are useful to explain the dispersion of the data. 120 Figure 1a was much fluctuation in the GSI for most countries, but with a large decrease from October to 125

December. The only country whose GSI was maintained high for the whole period was 126

Venezuela. 127

Also, in Figure 1a , it is possible to observe the asymmetry of the data, so a skew-‫ݐ‬ 128 distribution was adopted for the model's error. Since the model presents a variable dispersion, 129

we used a linear regression model for the dispersion. 130 131 In Figure 2a , the QQ-plot envelope shows there is no evidence that the skew-‫ݐ‬ 145 distribution is inappropriate to explain the death rate for each million inhabitants. Other aspects 146 of the model were analyzed by the quantile residuals (Figure 2b ), such as the correct 147 specification of the model's dispersion and distribution. We can conclude from these graphs that 148 the model satisfies the assumptions so that the model specification is appropriate. 149 150

Discussion 151 152

In the COVID-19 pandemic, prolonged periods of social isolation were adopted across 153 the globe, as recommended by the WHO. Social isolation can have a dual impact. It has been 154 observed that, in this period, there is an increase in the rate of suicides, mental disorders, and 155 depression, which are explained by human hyper sociability. In addition, it can trigger physical 156 effects that impact children, young people and the elderly 11,12 . In contrast, we know that social 157 . CC-BY-NC-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 7, 2021. ; isolation is extremely important to decrease the spread of the SARS-CoV-2 virus. It is important 158 to note that, during the pandemic caused by Influenza A (H1N1), social isolation was also 159 adopted 13 . Mortality from influenza and pneumonia during the 1918-1919 pandemic was lower 160 in civilians in rural areas when compared to those in urban areas. These observations have led to 161 the planning of strategies for pandemics, suggesting that social distancing interventions have a 162 potential effect on mortality by reducing the number of deaths 14 . In addition to social distancing 163

and isolation, large-scale testing is fundamental to fight against the pandemic. However, 164

addressing the influence of this factor on death rates remains a big challenge because countries 165 publish their testing data at different time points: some provide daily updates, while others 166 provide only on a weekly-basis, and some only publish figures on an ad-hoc basis at longer 167 intervals. 168

Based on GSI data extracted from the OxCGRT project 10 , it is possible to propose 169 statistical models to evaluate how closely these variables are related to time. Herein, the model 170

shows that the relationship between time and GSI is highly significant. When analyzing time 171

and GSI together, it was observed that, as the interaction of these two variables increases, a drop 172

in the death rate is detected. CC-BY-NC-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 7, 2021. ; maintained even in December (an atypical month because of Christmas and holiday season), and 213 its death rates were low and remained unchanged over time. 214

According to the present analysis, Uruguay followed a relatively lower GSI than other 215

LA countries but showed low death rates. Uruguay was a country that acted quickly, closing its 216 borders and schools, with insertion of screening tests, reducing SARS-CoV-2 infections and 217

controlling the outbreak very efficiently 20 . In contrast, Ecuador started with high social 218

isolation, but a decrease in the isolation rate was observed later. On the other hand, Ecuador had 219 a high mortality rate, which is accentuated over time even with the adoption of lockdown. In 220 addition, it should be noted that this country had poor conditions of public health infrastructure 221 at the beginning of the pandemic 2 . At the beginning of the COVID-19 pandemic, it had been 222 already suggested that closing public transportation, work places and schools is particularly 223 effective in reducing COVID-19 transmission 21 . 224

The rapidly evolving pandemic in LA countries is worthy of especial attention, 225

considering their often weak and low stringency responses to the current sanitary crisis. In this 226 study, GSI varies considerably in all LA countries over time. This variation can partially explain 227 why these countries have been differently impacted by COVID-19. In spite of not specifically 228 addressing and discussing the government policies adopted by each country, in this 229 investigation, we successfully show that social distancing and isolation measured by GSI 230 influences death rates from COVID-19 over time. For instance, the interaction between GSI and 231 time can decrease the number of deaths, which demonstrates the importance of maintaining 232 social distancing and isolation measures for longer periods, as opposed to what most LA 233 countries did. Almost no country kept its GSI high for much time, especially from October to 234

December. 235

Our results have significant implications; however, some limiting aspects must be 236 considered. 1) The GSI was extracted from the OxCGRT project. The curators of this database 237 emphasized how challenging the collection of information on the exact data was due to the 238 nature and extent of the policies of the different governments. This complex data set can 239 obscure the qualitative differences in each of the nine metrics GSI measures across countries. In 240 addition, many local and cultural factors can affect the implementation of interventions. 2) Our 241 data provide a general interpretation of the influence of time and GSI on death rates in LA. 242

Therefore, future studies can deepen the search for more specific interpretations for each 243 country, taking into account local aspects and other metrics not covered here.

3) The numbers of 244 deaths from COVID-19 can be easily underreported 22-23 , this is due to limited testing, problems 245

in determining the cause of death and the way in which COVID-19 deaths are recorded. Hence, 246

we cannot define the real impact of the GSI on death rates with perfect precision. 4) We know 247 that the differences in population size between countries are often large, and the COVID-19 248 death count in more populous countries tends to be greater. Thus, in order to perform a more 249 truthful comparison, we used the cumulative death data and calculated the death rate adjusted by 250 the population of each country. 251 252

Conclusions 253 254

We conclude that, in combination, time and GSI have beneficial effects on the decrease of death 255 rates from COVID-19 in LA countries. . CC-BY-NC-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 7, 2021. ; https://doi.org/10.1101/2021.04.05.21254941 doi: medRxiv preprint . CC-BY-NC-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 7, 2021. ; https://doi.org/10.1101/2021.04.05.21254941 doi: medRxiv preprint . CC-BY-NC-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 7, 2021. ; Table 1 . Estimates of the dispersion model and mixed linear model for death rates from COVID-19 in 2020 in Latin American countries.

Legend: GSI -Government stringency index. Significance levels: "***" 0.001, "**" 0.01, "*" 0.05, "." 0.1. The relationship between the predictors and the original response variable is inversely proportional .i.e. a negative sign indicates an increase of death rates while the positive one indicates a decrease. . CC-BY-NC-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 7, 2021. ; https://doi.org/10.1101/2021.04.05.21254941 doi: medRxiv preprint

Isolation Due to COVID-19 on Health in Older People: Mental and Physical Effects and 322 Recommendations

Effectiveness of workplace social distancing 325 measures in reducing influenza transmission: A systematic review

Host and environmental factors reducing mortality 328 during the 1918-1919 influenza pandemic

Description of Covid-19 Cases in Brazil 333 and Italy