key: cord-1031974-79bhjyfn
authors: Mehl-Madrona, L. E.; Bricaire, F.; Cuyugan, A.; Barac, J.; Parvaiz, A.; Jamil, A. B.; Iqbal, S.; Sellier, M. K.
title: Understanding SARSCOV-2 propagation, impacting factors to derive possible scenarios and simulations
date: 2020-09-09
journal: nan
DOI: 10.1101/2020.09.07.20190066
sha: 5ad9ac47e6289d1dac5f9f77c45b9411052064a5
doc_id: 1031974
cord_uid: 79bhjyfn

We aimed to analyze factors impacting the Covid-19 epidemic on a macro level, comparing multiple countries across the world and verifying the occurrence at a micro level through cluster analysis. The severity of the epidemic was most strongly related to exposure to ultraviolet light and extrapolated levels of vitamin D and to the health of the population, especially with regards to obesity. We found no county with an obesity level < 8% with a severe epidemic. We also found that countries where the population benefited from sun exposure or vitamin D supplementation and spent time outside fared well. Factors related to increased propagation of the virus included the use of heating ventilation and air conditioning (HVAC), population density, poorly aerated gatherings, relative humidity, timely policies of closing clustering places until aeration was improved, and daily amount of ridership on public transportation, especially subways. Population lockdowns, masks, and blood type did not provide much explanatory power. Contact tracing was not analyzed as very few countries applied it for long enough. The excess mortality observed is within the ranges of severe past influenza epidemics of 2016/2017 or 1999/2000 and lower than older severe influenza epidemics of the 1940s or 1970s. We estimated that COVID mortality death counts in European countries is over estimated when taking into account excess mortality further confirming the important role of comorbidities independently of lockdown policy. A few countries observed an under-mortality despite of some deaths counts attributed to COVID-19. Treatments or vaccines should protect the fraction of the population that is not suffering from severe comorbidities. Our study suggested that prevention measures should be directed to improving aeration systems, enhancing diets and exercise, and ensuring adequate levels of vitamin D. Prevention measures and attention should be paid to anxiety resulting from this episode which may be associated with increases in obesity, addictions, vitamin D deficiency, depression, suicide, and hunger ultimately weakening population resiliency.

ridership on public transportation, especially subways. Population lockdowns, masks, and blood type did not provide much explanatory power. Contact tracing was not analyzed as very few countries applied it for long enough. The excess mortality observed is within the ranges of severe past influenza epidemics of 2016/2017 or 1999/2000 and lower than older severe influenza epidemics of the 1940s or 1970s. We estimated that COVID mortality death counts in European countries is over estimated when taking into account excess mortality further confirming the important role of comorbidities independently of lockdown policy. A few countries observed an under-mortality despite of some deaths counts attributed to COVID-19. Treatments or vaccines should protect the fraction of the population that is not suffering from severe comorbidities. Our study suggested that prevention measures should be directed to improving aeration systems, enhancing diets and exercise, and ensuring adequate levels of vitamin D. Prevention measures and attention should be paid to anxiety resulting from this episode which may be associated with increases in obesity, addictions, vitamin D deficiency, depression, suicide, and hunger ultimately weakening population resiliency.

The data needed to accurately track the transmission and impact of COVID-19 have been hard to collect. The available data do not provide the full picture of the epidemic, are not standardized among countries, and are not always standardized within regions of the same country. Understandably, most governments (and modelers) have focused their efforts on in-country tracking and predictions, making international comparisons difficult. Data are defined and collected differently from country to country, from period to period (and sometimes even within a country at the same period). In an ideal scenario, we would have an accurate daily measurement of the proportion of people who are not infected, who have just become infected, who have recovered from the infection, and who have died from the infection. In actuality, each of these measurements is sub-optimal; tests to distinguish between the non-infected and the infected have been rationed (testing only when there are symptoms), recovered/immune patients have been tracked only from the infected cohort (with occasional one-off immunity tests across a larger population), and COVID-19 deaths have been defined differently at different points in time and across countries.

Every epidemic, every cluster starts and propagates through an infection. For the epidemic to be understood, we must understand environmental factors (which can include hours of daylight, temperature, humidity, ease of absorption of vitamin D, diet, and the presence of other pathogens, host factors (lifestyle, past and present exposure to other pathogens, obesity, co-morbid conditions, age, hygiene, mask wearing, and treatments), transmission factors (aerosol, hands, droplets, intermediary agent), and context (urban or rural, population density, types of heating and cooling, types of transportation, and gatherings, shared toilets) for example. The susceptibility rate is a function of the infection rate, the rate of recovery, and the death rate. The examination of large datasets and the comparisons of countries with each other can provide important clues regarding the behavior of the virus and this epidemic and can give clues regarding optimum management strategies. Any given factor can make the virus more or less infective, more or less benign, shorten or lengthen the time of contagion, or impact host resistance and recovery. These factors work together so that their influence may sometimes only be seen in large datasets. promote contagion and superspreading 18.7 times more often than open environments (Nishiura et al., 2020) .

Epidemic clusters have been identified involving closed places with people gathering with little air circulation such as transportation centers, places of worship, slaughterhouses, companies, hospitals, care facilities, prisons, jails, and police stations. On a more general level, major cities including London, New York, Paris, Madrid, and Milan, with factories, airconditioned offices, dense public transportation, and regular crowed social and religious events offer higher superspreading opportunities than medium cities or rural areas where little superspreading happened. Endo et al. showed that super-spreaders contribute more to infectivity than most people and recommend concentrating efforts on identifying those 10% of the infected population. (Endo et al., 2020) Lloyd-Smith, et al. explained the dissemination of sarscov-1 and remind that for a dissemination k value around 0.1, containment might be obtainable at ((1-1/R0)/2). This means that herd immunity could be obtained, depending on an R 0 value between 14 % and 39 % for the sensitive population prone to an outbreak (Lloyd-Smith, Schreiber, Kopp, & Getz, 2005) .

A shenzen study confirms such overdispersion by observing that 8.9 % of cases caused 80 % of infections . Successfully controlling superspreading may bring such epidemics to extinction as with SARSCOV-1. The 10 % of people who are super-spreaders are often highly socially connected individuals. Their many connections make them more exposed and thus more likely to be infected early and more likely to spread early causing a higher spread rate in the first phase of the epidemic during which time they become infected, spread, and become immune (Gomes et al., 2020) . For example, a bus driver who sells tickets, may be more exposed, and, once infected, may contaminate dozens of people. After he recovers, he no longer infects anyone. Once these kinds of superspreaders have become immune, the epidemic slows down. This "natural selection" of super-spreaders would mean that herd immunity could be as low as 5 % to 35 % depending on the R rate which is dependent upon factors that lead to propagation and severity (Gomes et al., 2020) . Lewis tests susceptibility notion for Stockholm and reaches a herd immunity estimate between 7 % and 24 % depending mostly on correlation between susceptiblity and infectivity of individuals based on an R0 of 2.4 (Lewis, 2020) . Careful analysis across countries and taking into account that the virus was present in France since early December 2019 and possibly weeks or months before, it does seem that R value and severity of this virus may vary significantly. It may actually have a very low natural Rn that has maintained it active for a long time without being visible. This R value may vary between seasons. In the past, the rare severe cases may have been confused with flu, pneumonia or other infections. Its mildness may have kept it invisible until a set of conditions united for repeated consecutive super-spreading leading to this epidemic and consequent world panic.

Although there was a case in France early December, the epidemic was invisible in terms of over-mortality until March 16 th . Hospitalizations began to peak in terms of death in week 14 between March 30th and April 5th as per Insee Data and EuroMomo. This is a combination of a very slow epidemic with an R that could be as low as 1.4 that picks up suddenly with and an R that may have tripled when the right conditions of clustering accumulated.

Changes in R and severity maybe associated to changes in populations density, activation of heating and air conditioning (HVAC), multiplications of reunions in closed spaces (ski vacation, occasional holiday gatherings, changes in commuting), presence or absence of competing/multiplying other pathogens, ultraviolet light, vitamin D levels, relative humidity, mucosa, seasonal diet and other factors that have not yet been identified. Such factors may impact the virus itself or the host in terms of propagation, receptivity, course, symptoms, or severity, thus in turn impacting contagion and R value.

Clinical observations hint that infections have become milder in Europe. Since there is no indication of any mutation, this may be because of natural selection mechanisms such as a minor evolution in the virus or that the most vulnerable were infected earlier. This is often observed in epidemics towards the end. It may also be because spring or summer make the hosts transmit less viral load or resist better.

Any association between susceptibility and super-spreaders would have been very strong if most transmission happened because of social contacts. This would have taken the bus driver, cashier, teacher, waiter, trainer, physician, nurse…. out of the equation very early on.

Much of the superspreading seems to be related to places: discotheques, hospitals, schools, dorms, slaughterhouses, malls, transport,s supermarkets…..Workers of such places are more susceptible and more likely to superspread as they spend more time and have more contacts with others in such places. Yet, it could be anyone who spends given enough time in such a place. HVAC may make dissemination more complex as it may become random within the space. This reduces the impact of such an association.

Much of contagion seems to require prolonged exposure and/or significant viral load. A cashier in an air-conditioned supermarket would be more exposed and would catch it earlier to pass it on earlier as it would stay longer in the supermarket. Once most nonresistant cashiers of supermarkets have become immune, they no longer contribute to superspreading. Superspreading becomes rarer in supermarkets. A customer may occasionally be at the origin of a cluster but as he spends less time, his impact, in most cases, would be relatively low compared to the cashier infecting customers all through several days, then infecting other cashiers whom in turn would infect other customers.

A very similar situation would happen with a bus driver, transport salesperson, restaurant waiter, slaughterhouse employee, dorm inhabitant, mall employee…… Sensitivity and resistance Another significant factor in contagion is Population resistance, that is the presence of a population whom, under normal circumstances would not get infected because of past immunity or other reasons as observed in the past.

Children are less likely to contract the virus and less likely to transmit it. In the French aircraft carrier Charles De Gaulle case study, out of a population of 1760 sharing common dorms, common corridors, and common cantines for a month, 1043 (59%) were infected, but the other 41% tested negative in spite of having been exposed to significant viral loads repeatedly. This is mostly a male and rather young population. It is possible this resistance may be lower in an older population and potentially higher in women and children. It is also possible that a higher fraction of the adult population that may be resistant to virus under normal exposure. Part of that protection may be due to genetics, cross-immunity, past immunity or other factors. For example, SARS-CoV-2 reactive CD4+ T cells were detected in ∼ 40%-60% of unexposed individuals, suggesting cross-reactive T cell recognition between circulating "common cold" coronaviruses and SARS-CoV-2 (Grifoni et al., 2020) .

Recent manifestations across the world in countries with tens of thousands of people gathering with close contacts in a very restricted open air space with few masks did not lead to cluster formation in spite of massive testing further confirming contamination happens indoors. Melbourne manifestation is particularly interesting as Melbourne is entering winter and Australia is tracing regularly and has to this date managed to keep virus under control.

Health Care facilities or Hospitals Slaughter houses Schools Hotels . 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 September 9, 2020. . 

We formed a multi-disciplinary team that included an infectious diseases physician, data scientists, and software developers. We continuously challenged our projections with reality and numbers from countries around the world, allowing us to refine our models and better understand the progression of the epidemic. All our predictions and findings were discussed and validated from a clinical viewpoint.

For computing, we used a MacBook Pro ( 

The following are the sources of data that were used in this paper:

The Japan Refrigeraton and Air Conditioning Industry Association: "Word Air Conditioner Demand by Region" https://www.jraia.or.jp/english/World_AC_Demand.pdf Imperial College London YouGov Covid 19 Behaviour Tracker, Jones, Sarah P., Imperial College London Big Data Analytical Unit and YouGov Plc. 2020, Imperial College London YouGov Covid Data Hub, v1.0, YouGov Plc, April 2020 https://github.com/YouGov-Data/covid-19-tracker, Accessed: July 2020

Euromomo "Excess Mortality" https://www.euromomo.eu/graphs-and-maps/. Accessed: May 2020 . 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 September 9, 2020. . Sunshine Exposure and Weather Temperatures: Weather Atlas "Weather forecast and Climate information for cities all over the Globe. Global Weather and Climate data" https://www.weatheratlas.com/ . Accessed: May 2020 Accuweather Inc. https://www.accuweather.com/ Accessed: May 2020 TWC Product and Technology LLC Weather.com: https://weather.com Accessed: May 2020. When the data was not located in only one source, we used manual data collection and aggregation.

Tropospheric Emission Monitoring Internet Service, European Space Agency "UV station data based on operational TEMIS satellite ozone data" http://www.temis.nl/uvradiation/UVarchive/stations_uv.html. Accessed: June 2020

Google LLC "Google COVID-19 Community Mobility Reports" https://www.google.com/covid19/mobility. Accessed: June 2020

Worldometers "COVID-19 Coronavirus Pandemic": https://www.worldometers.info/coronavirus/ Central Intelligence Agency, The World Factbook: "Country Comparison: Obesity -Adult Prevalence Rate," https://www.cia.gov/library/publications/the-worldfactbook/rankorder/2228rank.html. Accessed: July 2020

The Atlantic Monthly Group "The COVID Tracking Project: https://covidtracking.com/. Accessed: July 2020

Data was transformed to fit parametric distributions prior to statistical analyses, which were performed with Student's t-test, linear regression and post-hoc tests. Especially for ordinary least squares regression, natural logarithmic transformations were done to remediate normality violations in the standardized residuals. Interpretation on the final model of the regression and analysis of variance were adjusted because of logarithmic transformations.

For correlational studies, whether continuous or count data were used, both Pearson's and Spearman's rank correlational tests were done according to the data used. Normality assumptions were not made as we used non-parametric correlation tests to overcome non-linearity. Most of these normality violations related to bimodal distribution, so running Pearson's correlation tests provided a complete description of the association. See On the Effects of Non-Normality on the Distribution of the Sample Product-Moment Correlation Coefficient (Kowalski, 1975) . Also, power in significance testing were also done to ensure that for some data points that have very minimal number of samples, this was considered that the interpretation of significance is practically taken into consideration.

For cross-correlation of stochastic processes proper diagnostics were done to ensure the absence of autocorrelation processes that could signal the delayed copy of itself from its own function of lag. All significance levels are set at 0.05, otherwise specifically stated.

Our initial Euro-centric approach led us to predict an end to the epidemic by late spring/ early summer in most of Europe which we published on March 20, 2020 (Karim, 2020) .

Subsequently, we reviewed contemporary CoVid-19 studies for descriptions of the epidemic dynamics and the factors impacting spread or severity. We expanded our search for and use of worldwide data to help us calibrate models internationally. We applied our study to a variety of countries and continents in order to validate the approach and understand a wider range of situations.

We chose to focus on countries' published mortality rates rather than their published infection rates, as the former seemed more reliable. We confirmed that in Europe, communicated data on Covid-19 lethality matched the official mortality rates (in spite of many press articles indicating otherwise). We adjusted mortality numbers to account for methodology changes (e.g., France included retirement home deaths only in mid-February). We adjusted calculations by exploring case counts from the few reliable environments where enough tests were done to provide some validity. Finally, we included serology antibodies tests in our calculations, taking into account their error margins.

. 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint

Data from the Project for European Mortality Monitoring in Denmark showed that the overmortality peak was caused by Covid and not lockdown. Some countries were caught off guard by a virus that is mild enough to be invisible with a changing R value, a fairly long incubation period, and a sudden acceleration in infectivity. The countries of Spain, Italy, France, Belgium, UK, Netherlands, Switzerland, Sweden, and possibly Portugal, Malta, and Greece peaked before any lockdown took place.

We classified lockdowns into 7 major groups:

1-Stop large gatherings, encourage social distancing --→ Sweden 2-Places lockdown: Close places of indoor gatherings especially restaurants, keep people in their own separate regions, and strongly encourage working from home (--→ Switzerland, Germany, Denmark, Finland, Malta, Greece, Austria, Ireland, Netherlands, Australia, New Zealand).

3-Peoples' lockdown: = Public places are closed as are parks and peoples' movements within the city were restricted to absolute necessity under police control (--→ Italy, Spain, France, Belgium, UK) 4-Night lockdown = Some indoor gathering places are closed. Free movements are allowed in the daytime with restricted movement at night (Senegal, Algeria, Egypt) 5-Full contact tracing = South Korea, Taiwan, Singapore 6-3Cs lockdown = Instruct population to avoid Closed spaces, Crowded places, Close Contact settings (-→ Japan) 7-Different policies across a same country: Mostly large Federal countries (-→ USA, Brazil, China)

Understanding the natural course of the virus gives insight as to how much collective immunity was acquired in a particular context. To assess this, a few sample countries were analyzed, working backwards from mortality peak to probable infection date.

Given that the median delay between infection and death lies between 21 days and 25 days (mean of 5.1 days

incubation to symptoms (Lauer et al., 2020 ) and 17.8 days first symptoms to death) (Lauer et al., 2020) , that would place the peak of . 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 September 9, 2020. . the infection in week 11 (March 9 to March 15) before lockdown of people happened and after lockdown of cluster risk places had taken place. Locking down non-essential clustering places may have been sufficient to reverse the curve that started becoming aggressive around week 8 to 9. For at least 12 weeks the epidemic was moving at a very slow pace, becoming aggressive when conditions permitted.

Temperature levels. We found no correlation for temperature, even though the epidemic did not pick up significantly where temperatures were below freezing. That may also be explained by lower population density in areas with lower temperatures. We could not find any correlations with temperature outside of an observation that infectivity did not increase when temperatures were below freezing. Wang et al. found that high temperature and high humidity reduced infectivity (Wang et al., 2020) , though we found correlations only with relative humidity, hours of daylight, and levels of ultraviolet radiation. We estimated the mean incubation period to be 6.4 days (95% confidence interval: 5.6-7.7).

We confirmed this by checking EuroMomo's data and dates and by comparing their curves with data extracted from INSEE in France and Statbel in Belgium which were consistent with EuroMoMo's data. We traced back with a fairly reasonable degree of confidence, infection peak time for countries by placing it 3 weeks before mortality peaks to evaluate how much of its natural course the disease had in different countries. Most peoples' lockdowns happened concurrently or after the peak had and when the curve was already diminishing.

Belgium hit a plateau of 83 deaths on April 4th with consecutively 70, 75 and 86 to start dropping on April 8th, placing the probable peak and turnaround infection period around March 16th. That places the turnaround right after public places lockdown and before individual people lockdowned.

France hit its plateau of excess rmortality on April 1st which places its probable peak around March 12th and March 14 th , after the ban on large gatherings, right before public places locked down, and 4 days before individual peoples were locked down. This applies also to Belgium, France, Italy, and the UK. No clear trend existed for lockdown in terms of mortality. Some public places were locked down early, some awaited outcomes and were locked down late, but we saw no clear trends to outcome for mortality in terms of when the lockdown occurred, how severe the lockdowns were, and how long they lasted. Most countries that locked down early avoided a severe peak, but Greece, Portugal, Malta, and Switzerland locked down late and had a mild epidemic.

. 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 September 9, 2020. . Countries deciding to lockdown their citizens did so as they were heading towards a high peak of infection and at the time without knowledge as to where they were in the curve except for predictions from mostly faulty epidemic models. Around April 10th (in Europe) consecutive drops began to occur each day in mortality. Studies from the Far East give us some sense as to what measures may have contributed to lowering of the curve.

From the above data, it seems reasonable to assume that the epidemic followed much of its natural course in at least 17 of 21 countries mentioned. This is further confirmed by the absence of rebounds anywhere except for localized clusters in locations/populations that had not been exposed. Regions that were exposed and where the epidemic took its course benefit from a fair level of protection under spring and summer conditions with some risk of occasional clusters but a limited risk of a full national epidemic.

Looking at a different part of the world, Japan had no lockdown, performed relatively few tests (less that most European countries), has very dense cities and one of the oldest populations, yet suffered a mild epidemic that did not seem to rebound. Japan simply instructed its population very early to avoid closed places, large crowds, and unnecessary physical contact with strangers as a policy that could be applied in the long term. Could that have sufficed? Perhaps their relatively superior diet and health helped. Tadjikistan had no lockdown and 6 deaths per million compared to Kazakstan's (with a lockdown) of 20 deaths per million. Japan's death rates were 8 deaths per million people; UK, 650; Spain, 607; Italy, 567; USA, 399; Germany, 52; and Norway, 46.

Data from the Project for European Mortality Monitoring (2020 -Weeks 1 to 23) shows that some countries were more vulnerable to respiratory human-to-human transmitted epidemics that others, depending on population density and heating technologies. Belgium, Spain, and the UK were hardest hit (2 to 3 times influenza's z-score peak); whereas, several countries from southern, central, and northern Europe had peaks often lower than that of recent influenza. In spite of significant peaks representing a moderate excess mortality, France had a peculiar plunge in mortality after lockdown was lifted. Except for Portugal, Greece, Germany and Austria, who are prone to epidemics and successfully "skipped their turn," countries that had high peaks for influenza epidemics in 2017 and 2018 had high Covid-19 peaks, while those who had mild influenza epidemic peaks in 2017 and 2018 had even milder Covid-19 peaks.

The high-level model . 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 September 9, 2020. . In Europe, during the winter of 2020, Spain, Belgium, and the UK had a very different story than that of Portugal, Greece, Denmark or Germany with lower rates of infection. This can be particularly interesting, because of the high clustering nature of Covid and low dispersion factor. Once super-spreaders and clusters come under control, the propagation rate returns below 1 and the epidemic disappears.

For environmental/weather factors, we examined 5 key variables: UV, Ozone, Humidity, Sunshine, and Temperature. These were selected on the basis that the corresponding data was available across cities and countries. However, as the data sets for UV and Ozone are time-series and for Humidity, Sunshine and Temperature are monthly averages, we did the initial analysis separately.

The epidemic took on a very different course between the northern and southern hemispheres. It picked up in late Fall/Autumn until winter in Wuhan (Northern hemisphere) and Brazil (South hemisphere). It then picked up in Europe and United States in late winter to subside as spring arrived.

We have found that for a variety of European and Southern hemisphere cities, there is strong association between daily UV levels (as published by Tropospheric Emission Monitoring Internet Service (http://www.temis.nl/uvradiation/UVarchive/stations_uv.html) and deaths, as well as new cases. The association is particularly strong when lethality and new cases are lagged by 10-15 days (where the correlation coefficients range between |0.2| and |0.6|) as shown in the charts below. The amounts of sun needed may vary depending upon individual skin nature and pigmentation.

For sunshine, the 4-month average of daily sunshine is correlated with death per capita, cases per capita, death count, and case count at a borderline significance level of p = 0.1. There was no correlation for temperature. Interestingly enough, the epidemic increased in the Middle East, India and Pakistan in late spring despite high UV levels. This apparent contradiction may be explained by the population of these countries increasing their use of air conditioning, avoiding sun, and spending more time in closed places, thus creating better propagation conditions and further explaining the need for sun and fresh air to reduce spread or even reaching a point of disappearance.

The South to North gradient of the epidemic curve dropped in Europe as spring arrived, combined with lower peaks in countries where their parks were open and people could access sun, further indicating a beneficial effect of sun or UV's or light or vitamin D . 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint (Mitchell, 2020) . Other researchers have had similar findings (Alipio, 2020; Daneshkhah et al., 2020; Ebadi & Montano-Loza, 2020; Ilie, Stefanescu, & Smith, 2020) . It has often been suggested that vitamin D may play a protective role in many respiratory and inflammatory diseases. Such suggestion may also be true for SARSCOV-2. These observations indicate a possible beneficial effect of sun exposure on hosts beyond the direct effect UVs may have outdoors on the virus.

For Ozone, the correlation was weak, and it was discarded

We found an inverse correlation between humidity levels and absolute death counts. A similar trend is also confirmed in the study from Beihang University in Beijing, China that R0 is lowered by 0.0225 and 0.0158, respectively for absolute temp in C and relative % in humidity (Wang et al., 2020) . This is highly correlated with studies with influenza, too (Wang et al., 2020) . This humidity correlation may contribute to explain why coastal cities often faired better. London, Madrid, Brussels, Milan, and Paris were the hardest hit in Europe in that order, while none are coastal. The hardest hit coastal cities were Barcelona, Stockholm, and Amsterdam, which had either no lockdown or a very mild lockdown. Humidity did not appear to be a factor

In Figure 2 , we present the Cross-Correlation of UV index and Ozone for several cities:

. 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 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.

The copyright holder for this preprint this version posted September 9, 2020. . https://doi.org/10. 1101 In Figure 3 , we present the correlation matrix of weather-related variables for all cities and countries aggregated together.

For social variables, we looked at two sets of data; demographic data (Density per km2, average daily ridership for cities with a metro, the GINI index of income distribution, the GDP per capita, and mobility data (changes in frequenting of residential, workplaces, parks, grocery/pharmacy, retail/recreation and transit stations).

The following correlations (and significance levels) were found.

. 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 September 9, 2020. This shows that increases in ridership and higher population densities are associated with both higher infection rates and deaths and that higher income inequality (as measured by Gini) is associated with a higher number of cases. In developing countries, the epidemic seems to hit wealthier populations harder; whereas, in Europe and United States, it seems to hit disadvantaged populations harder. . 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint

Using Google mobility data, we have found that there is a statistically significant correlation in most countries at +25 lag between the presence of transit stations and deaths from COVID-19. This pattern also applies to workplaces in countries with mild or no lockdown. Metro/Subway ridership is often an indication of office concentrations in modern buildings with shut windows, possibly recycled air or HVAC where clusters may form. Both variables may correlate.

Did metro mass transportation contribute to the high infection rate or was it office buildings or a combination (as there's high correlation between daily death count and cases count). Significance was set at 0.05, df = 25, so that the correlation coefficient should be > |0.22|. To a lesser extent a similar pattern applies to retail and recreation. This is altered in the sense that most recreation and retail had been closed in almost all countries and restricted . 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint in Sweden. All five hardest hit cities were thickly populated and had a dense subway system (London, New York, Madrid, Brussels, Milan, Paris).

. 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint

In Figure 5 , we present cross-correlation coefficients for Google Mobility Variables by Death Counts per Location . 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 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.

The copyright holder for this preprint this version posted September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint

Early in the epidemic, it appeared that co-morbidities played a role in the Covid-19 impact, including diabetes and obesity (needs reference), advanced age (where co-morbidity is much more common). To quantify the impact of lifestyle variables on Covid-19 mortality, we turned to "Our World In Data," which provides daily, collated, open source data for Covid-19 mortality (using the European Centre for Disease Prevention and Control, the World Health Organization and Johns Hopkins University). We used the latest cumulative natural log transformed total death per million and regressed it against the 2016 compilation of the adult prevalence rate of obesity (Country Comparison: Obesity -Adult Prevalence Rate, 2016).

In our first run, in which we did a worldwide regression, we found that the obesity adult prevalence rate (which is the proportion of obese people in the population per country) was a statistically significant predictor in determining the Total Deaths Per Million. For each unit of increase in obesity %, the average total death per million increased by 8.76% on a global average (with a highly significant p-value). However, as the epidemic was at different stages in different continents, we decided to do the regression by continent. This gave us the following results: 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 September 9, 2020. . mortality rates were low despite high obesity. However, those countries closed their borders very early. In the presence of obesity, other variables (diabetes prevalence, smoking, age 70 years or older, life expectancy, population density) were insignificant, further demonstrating the importance of obesity as a significant indicator in the epidemic course. Japan, the country with the oldest population in the world, with very high population density, with no lockdown, few restrictions and low testing outperformed by most other countries, has the lowest obesity rate in the world at 4.3% and had minimal deaths. Others have confirmed these observations (Lighter et al., 2020; Tan, He, & MacGregor, 2020) .

Most modern AC units are in fact HVAC. An AC unit proliferated viral transmission in a Guangzou restaurant through droplets and aerosols Lu et al., 2020) . Beyond that restaurant, the majority of infection clusters occurred in closed spaces, which were poorly ventilated or ventilated, cooled, and heated through HVAC. Health care facilities and hospitals typically have HVAC as do most slaughterhouses, hotels, social facilities, discotheques, buses, schools, companies of moderate size or larger, call centers, and gyms. Subways/metros rarely have HVAC but more likely artificial ventilation. Places of worship rarely have HVAC. Family reunions sometimes have HVAC.

Our analysis of available AC demand showed a correlation between commercial demand as per JRAIA 2018 data and death from Covid-19 until June 25th at significance level of p = 0.1 using 15 European countries which are at the end of their epidemic wave and for which data was available. Figure 5 presents the correlation coefficients related to air conditioning:

. 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 September 9, 2020. . The absolute death count as of June 25 is correlated with 2018 commercial AC demand at 0.86; 2018 total AC demand at 0.71; and commercial AC per capita at 0.58. Death per capita was associated with 2018 commercial AC demand at 0.77; commercial AC per capita at 0.64; and 2018 Total AC demand 2018 at 0.61 With regard to influenza spread, improvement of ventilation in high density public spaces could be an important and relatively easy-to-implement strategy supplementing vaccination, corresponding to a vaccination coverage of 60-70% for an efficacy of 40%, 50-60% for 60% efficacy, and 40-50% for 80% efficacy. In the aerosol model, consistently improved ventilation beats vaccination even with full coverage if efficacies are low. It can be done fast and it works for aerosols, known droplets, and yet unknown viruses (Lazzari, Smieszek, & Salathé, 2019) . All five hardest hit cities had a high concentration of modern office buildings and hotels with shared and/or a culture of centralized HVAC in office buildings, malls, homes (London, New York, Madrid, Brussels, Milan) . For most buildings, the easiest way to deliver outside air directly across the building envelope is to open a window. Window ventilation not only bypasses ductwork but increases outside air fraction and increases total air change rate as well (79). Administrators and building operators should discuss a plan for increasing perimeter, and specifically window, ventilation when outdoor temperatures are adequate for this practice (Dietz et al., 2020) .

. 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 September 9, 2020. . Two studies hinted that blood type plays a role. We correlated multiple countries data mortality with blood type. We could not confirm these findings, with a slight statistical tendency the other way.

Asian countries easily wore masks and were spared. Denmark, Finland, and Norway skipped masks and were spared. African countries tried to wear masks and took them off and were spared. Open air music festivals and demonstrations on June 21 st with or without masks did not show an increase in cases or the appearance of clusters within 14 days.

Given a median delay between infection and symptoms of 5.1 days, depending upon testing policies, the effects of masks wearing should appear within 5 days and become definitive within 14 days. California observed an increase of positivity rates in spite of making masks mandatory in the entire state 25 days after they went to lockdown. California made masks mandatory immediately upon leaving home on June 18 th . (Data from https://covidtracking.com/data/download). Of course, we have no data upon people's compliance with mask wearing.

. 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 September 9, 2020. . https: //doi.org/10.1101 //doi.org/10. /2020 Texas also had an up curve in which masks did not appear to show any effect. Texas made mask wearing mandatory on July 2nd. In these 2 states introduction of masks failed to bend the curve.

We used simulation tools to reproduce the dynamics of the epidemic comparing different regions and countries to understand dynamics of the epidemic. We ran a first set of simulations matching expansion, death rates and available antibody tests leading to 2 scenarios (seasonal and herd immunity). We published in late March preliminary results on propagmath.org indicating that both scenarios would lead to the epidemic subsiding by June 2020 in western Europe with a progressive gradient from South to North.

We observed that severity and fatality differed significantly from region to region beyond differences in accounting methods or healthcare systems needed to refine calculations, epidemic propagation, and severity. We explored the literature and isolated and tested significant factors, proposing a set of factors describing what may be an "eco-system" favorable to epidemic propagation or contraction. The most significant factors were hours of daylight and level of ultraviolet radiation (taking into account that different populations need different levels of UVs), closed spaces with recycled air (public transport, closed large gathering, events, shared A/C buildings with recycled air, slaughterhouses, and the like), humidity, age, and smoking status. We also accounted for the fact that data has accumulated to show that the virus started earlier than initially thought and thus has a lower R0. We saw that much of the propagation came from "super-spreaders" and in closed clusters. Super-spreaders are often more exposed and tend to be exposed earlier in the epidemic (for example, a transport driver would be exposed to hundreds of travelers and could in turn contaminate hundreds more). These super-spreaders are more likely to be infected in the first half of the epidemic. Once they subside, the R0 drops drastically.

Our simulations on these factors led us to a herd immunity in major urban cities of around 20%-30% and in rural areas of around 7%-10% when weather conditions are most favorable to propagation and around 7%-14% in in major urban cities and 3%-6% in rural areas when weather conditions are least favorable to propagation. That would not stop clusters from happening in sunny, humid summers in slaughterhouses where temperatures are moderately cold, air is recycled, and people kept away from daylight. Epidemic spread conditions would not have occurred outside of a slaughterhouse. These simulations and herd immunity conclusions confirm other models showing that herd immunity may be much lower and faster than initially expected.

Our simulations also show that in case of a new epidemic, changing A/C systems to fresh air, reducing population on public transport, favoring some teleworking, shutting down . 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 September 9, 2020. . large, closed gathering places that offered prolonged contact, and favoring open air activities could reduce spread drastically without lockdown side effects, and even benefit the economy because of investments, innovation, and a healthier population. The population would be better protected and prepared for aerosol and droplet pathogens.

Given the correlation to UVs and the need for UV exposure combined with the literature on the benefits of vitamin D on respiratory diseases, inflammatory diseases, and even Covid-19, authorities could lead campaigns to ensure their populations have adequate sun exposure and vitamin D levels to protect in fall/winter from the common cold, influenza or a milder Covid.

Our simulations led us to a crude corrected mortality rate between .06% and 0.4%, consistent with many serological studies except for Holland with 0.6% and Spain with 0.9%. Over-mortality can be calculated after the epidemic ends and may reduce these figures as it will take into account co-morbidities which seem to play a major role in this epidemic.

Our simulations also lead us to 3 major scenarios Best case -Herd immunity or isolation is successful in the southern hemisphere before October and the virus disappears.

Memory immunity case-The virus returns and 8% -15% of major cities' population (3%-5% rural) has lost its acquired immunity but can rely on memory immunity to produce quickly protective cells, thus reducing R 0 faster and severity faster. That would lead us to a curve that is 30% lower in number of cases and 38% lower in severity if little action is taken and 50% lower in number of cases and 70% lower in severity if effective moderating action is taken.

Seasonal case-The virus returns and 8% -15% of major cities' population (3%-5% rural) has retained its acquired immunity thus reducing R 0 faster and severity even faster. That would lead us to a curve that is 60% lower in number of cases and 70% lower in severity if little action is taken and 70% lower in number of cases and 80% lower in severity if some effective moderating action is taken. If these projections are correct, in the following years we would have experienced the birth of a new common cold.

In all cases, we expect the next stages to be milder because of natural evolution, our ability to adapt, and maybe our ability to take reasonable constructive smart action.

Our models, simulations are online to explore on . 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 September 9, 2020. . https://doi.org/10. 1101 propagmath.org

Taking into account Dissemination factor, probable moderate association between susceptibility and superspreading, presence of a probably more resistant significant population, herd immunity may be as low as 10 to 20 % when natural conditions are maintaining the Rn of the virus at a slow propagation rate (1.2 to 1.5). In case of the return of clustering, density, aeration and weather conditions that raise the R value to 3 or 4, the required immunity would be 67 % and 75 % respectively, not accounting for the dissemination factor which reduces that requirement as per our estimates to 37% and 52%, respectively. Less dense regions that are not prone to respiratory epidemics and have not been hard hit, may continue to have mild or even milder episodes since they do not provide such conditions that raise the R value of the virus in spite of having less herd immunity.

Attention should be paid, to (1) regions that are prone to respiratory human to human epidemics such as influenza and have avoided it, as they may need careful anti-clustering measures (reducing density of people in transportation, reducing the use of HVAC, and limiting large events); (2) regions that in certain situations see an increase in population density with closed space gathering or long stays of large populations in places with suboptimal fresh air; (3) populations with insufficient vitamin D levels; (4) populations with poor diet and the presence of co-morbidities.

The benefit of such a situation, is that assuming immunity or memory immunity is retained after an aggressive round may be attained making the virus non-viable when conditions become less favorable, assuming immunity or memory immunity is protective. If the pathogen doesn't fade away or is reintroduced, acquired immunity from the initial round would still mitigate the epidemic as simulated below.

Learned lessons and simple anticipations may actually mitigate or facilitate disappearance even before the appearance of a vaccine. These actions may also help for influenza and any droplets or aerosol-based future epidemics Given that protected population genetics, epigenetics, cross immunity, cellular immunity, T cells, antibodies, could provide a herd immunity ranging between 44 % and 75 %, it is possible that hard hit regions may have milder epidemic episodes.

Masks . 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 September 9, 2020. . Significant clusters have formed in gyms, hospitals, and slaughterhouses where masks were mandatory. The studies are not conclusive. A 2009 randomized, controlled trial in Japan did not demonstrate that masks prevented health care workers from contracted cold symptoms or colds. A 2010 systemic review found: "There is some evidence to support the wearing of masks or respirators during illness to protect others, and public health emphasis on mask wearing during illness may help to reduce influenza virus transmission. There are fewer data to support the use of masks or respirators to prevent becoming infected (Cowling, Zhou, Ip, Leung, & Aiello, 2010 ). An RCT cautioned against the use of cloth masks due to moisture retention, reuse, and poor filtration, resulting in an increased risk of infection. Further research is needed to inform the widespread use of cloth masks globally (MacIntyre et al., 2015) . Evidence as to efficacy is weak, and it becomes weaker when transposed into real life for large population and raises the question as to long term effects of prolonged mask wearing in populations.

For example, females wearing Hijab or Niqab had low plasma 25(OH)D levels, attributed to low sunlight or ultraviolet exposure (Mallah et al., 2011) . Masks reduce the surface area exposed to sunlight. The long-term use of traditional niqab can adversely affect the functional vital capacity and the fractional expiratory volume at one second (FEV1) among Saudi adult females (Alghadir, Aly, & Zafar, 2012) . Respiratory infections and asthma were significantly more common among veils users (p < 0.00001 and p < 0.0003, respectively, probably secondary to infection (Ahmad et al., 2001) . Discomfort levels increase with duration of time wearing masks and nasal resistance increases upon removal of an N-95 mask or a surgical facemask, potentially due to nasal physiological changes (Zhu, Lee, Wang, & Lee, 2014) . Masks wearing, is a political, sociological decision. The data does not support that it's being a meaningful factor in real life conditions where it may actually give an illusion of protection.

Density. Density is another driving factor and calls for attention, as the density of regions changes over time (especially from tourism), thus exposing naive populations who have had limited exposure to the virus. This may be limited when it happens in the summer, with some possible protection from sun and many open-air activities. These 2 driving factors may mitigate each other. However, the situation is different if people spend increasing time in closed places, aggravated by recycled air or HVAC where clusters may form. Authorities and people should consider alternatives to confined mass transport which may favor this epidemic and future ones beyond masks' relative protection. Authorities and companies may want to consider improving air circulation with renovated fresh air to protect employees and even gain productivity. SMIEZEK, et al. show how improving air exchange rates in buildings may correspond to a vaccination coverage of 60-70% for an efficacy of 40%, 50-60% for 60% efficacy, and 40-50% for 80% efficacy. In the aerosol model, consistently improved ventilation beats vaccination even with full coverage if efficacies are low (Lazzari et al., 2019) .

. 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 September 9, 2020. . Data also indicates a very high rôle and proportion of deaths that are due to comorbidities mostly related to obesity and vitamin-d defficiency. This is further confirmed by the existence of many countries like Denmark observing an undermortality in spite of having some COVID deaths.

Prolonged anxiety, related to this episode is likely to increase obesity and vitamin-d defficiency making population more vulnerable. This calls for special attention in populations and governments to increase awareness as to the importance of lifestyle, healthy lifestyle, exercise and diet. Data also indicates that increase of close dense places and increase in A/C may play a rôle in COVID epidemic and others.

Given the proportion of comorbidities, treatments and vaccines may be welcome but will only contribute to reducing a portion of mortality in fragile populations exposed.

Studying Sweden where the virus has had much of its natural course because of the limited restrictions imposed by the government and by the fact that it is far north and thus benefits from little sun protection, when we try to assess Covid Lethality, the actual number of infected cases can only be extrapolated making all IFR (Infection Fatality Rate) values approximations since the actual number of infected is unknown. If we account for specific antibodies detected through serology tests and add memory immunity T cells, cross immunity, cellular immunity take us to values between 14% and 30%, which puts us in a . 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint range between 1,400,000 and 3,000,000 cases. This results in an IFR or lethality between 0.17% and 0.34%.

Having a closer look we notice a 3,725 count in over-mortality instead of 5,420 Covid declared deaths. Some of this could be due to deaths with comorbidities having been attributed to Covid. IFR may actually be lower and would drop to a range between 0.12 % to 0.34 %.

Ledberg (2020) explored the epidemic, only looking at the aggressive window from the end of March to early May reaching a count of 4857 representing a 10.5 % over-mortality, lower than the 1940/1941 influenza outbreak in terms of mortality/capita. Looking at a wider window gives a better perspective as to actual over-mortality. Covid's effect on a naive population has led to a maximum increased mortality of 8 % compared to influenza. in previous years assuming no deaths were caused by influenza, the effect of Covid on the population in Sweden where little restrictions were applied was an 8% increase in death compared to previous years, assuming all excessive death was caused by Covid and none by Influenza.

The 2020 Mortality from all causes was 0.48 %. Within that the over-mortality related to Covid was 0.04%. This is no comfort for those who lost loved ones, but as immunity builds, future eventual outbreaks are likely to be milder and more can be done to prevent mortality. Over a semester of Covid in Sweden, where the population was recommended to be cautious and large events were cancelled, but also where influenza seemed almost absent, the situation in terms of overmortality over a semester led to a situation 8 % to 10 % more severe than that of previous years in which there were influenza outbreaks.

Sweden is a country that does not benefit from sun, and where, like much of Europe, there is significant obesity. On the other hand, Stockholm, the main city, is less dense than some European capitals.

In Denmark, the epidemic effect on mortality is invisible. Over the first 2020 half an undermortality can be observed despite a short lockdown of places. We observed a difference between under-mortality and Covid death suggesting a possible Covid death attribution that may be attributed to comorbidities. Over-mortality due to Covid-19 was -0.01 % Belgium Belgium is one of the 3 hardest hit countries in the world despite a severe prolonged lockdown. It has declared 9776 Covid deaths. Over-mortality was 3993, confirming the . 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 September 9, 2020. . same excess in Covid counts with respect to over-mortality as observed for Sweden and Denmark. This difference could be caused by 2 possible explanations that could each could contribute to this massive 60 % overcount. Either the Covid situation spared up to 5783 lives or simply 5783 deaths occurred with patients who died of comorbidities and Covid, but actually died of comorbidities. Belgium has a population of around 11,486,000. Over-mortality in 2020 is around 7% on of the top 3 hardest hit countries with an overmortality of 0.03 %.

France Insee's data provides a similar pattern to that of Sweden with an over-mortality of 6.73 % corresponding to 17,691 people. This is lower than the 29,779 declared by France. If there we set aside over-mortality caused by influenza or lockdown and assume that all of that over-mortality is attributed to Covid-19, that figure is still lower by 40 % than the COVID-19 death total, indicating most likely a very large comorbidities factor attributed to COVID. Given that May and June 2020 showed under-mortality, the over-mortality displayed during the Covid episode is comparable to that of the 2016 -2017 influenza.

. 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint When we examine the recent severe Influenza epidemic in 2016-2017 over the months of December 2016 to February 2017 and compare mortality to that of February 2020 to April 2020, the Covid year shows an excess mortality of 1.42 % in one of the 10 hardest countries in the World as per death per million inhabitants. The under-mortalities that followed further confirms the role of comorbidities in Covid deaths counts which could turn out to be 30 % to 60 % of Covid attributed deaths. Having a longer look at history gives us a better insight into what happened in one of the hardest hit countries in the world, but also one that displayed sufficient data transparency because of its structures. In France, 2020 in terms of mortality was slightly milder, or comparable to 2016/2017, 1973, 1997, 2000 influenzas and milder than 1969,1956,1963,and 1962. Monthly Death (INSEE data)/ Population (Estimate of linear evolution from 42 Millions to 67 Million. This suggests a disproportion between the epidemic's impact and the populations' overall reactions which could lead to actions and regulations that are counterproductive on physical, physiological, and the psychological health of populations, making them fragile against Covid-19.

From the multiple factors impacting the epidemic that we analyzed, and, comparing multiple countries across the world at a macro level on the one side and verifying occurrence at micro level through studies review or cluster analysis, 2 factors stand out as to predicting severity: (1) UV/Vitamin D levels and (2) the health of the population and especially obesity. At the time of writing this paper, we found no country with an obesity level under 8% having a severe epidemic. We also found that countries in which the population benefited from sun exposure or vitamin D supplementation and spent time outside, fared well.

A few factors stand out as to propagation: (1) Commercial HVAC, (2) Density and poorly aerated gatherings, (3) Relative humidity; (4) Level of UVs, (5) Timely policies of limiting gatherings in closed clustered places until aeration is improved, and (6) reducing daily Metro ridership.

Multiple other factors explained little of the variance including population lockdowns, masks, blood types, and ozone. However, mask policies do not tell us the degree to which people complied with mask policies. Masks may have a mitigating role in situations of high density with HVAC or travel on poorly ventilated public transportation. Dispersing urban dwellers into the countryside and away from high occupancy buildings with HVAC could . 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 September 9, 2020. . reduce transmission. Contact tracing was not analyzed as very few countries would be able to apply it for long enough. Over-mortality observed is within ranges of severe past flu epidemics 2016/2017 or 1999/2000 and lower than older severe flu epidemics of the forties or seventies.

We calculated that COVID-19 mortality death counts in European countries is overestimated when taking into account over-mortality, further confirming the role of comorbidities. This is confirmed independently of lockdown policy. A few countries observed an under-mortality despite some deaths counts attributed to COVID.

Treatments or vaccines will protect the fraction that is not suffering from severe comorbidities.

Prevention measures such as reviewing aeration systems, enhancing diet and exercise, ensuring adequate levels of vitamin-D may turn out to be efficient in providing protection against COVID, possibly influenza and other epidemics, not to mention increasing the efficiency and wellbeing of populations.

Prevention measures and attention must be paid to anxiety resulting from this episode resulting in increased in obesity, addictions, vitamin-D deficiency, depression, suicide hunger and ultimately making populations more exposed to pathogens like Covid-19

Weather Related Variables . 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint Worldometers Numbers . 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 September 9, 2020. . https://doi.org/10.1101/2020.09.07.20190066 doi: medRxiv preprint

The Effect of Wearing the Veil by Saudi Ladies on the Occurrence of Respiratory Diseases. The Journal of asthma : official journal of the Association for the Care of

Effect of face veil on ventilator function among Saudi adult females

Vitamin D Supplementation Could Possibly Improve Clinical Outcomes

Incubation period of 2019 novel coronavirus (2019-nCoV) infections among travellers from Wuhan, China

Epidemiology and transmission of COVID-19 in 391 cases and 1286 of their close contacts in Shenzhen, China: a retrospective cohort study

Face masks to prevent transmission of influenza virus: a systematic review

The Possible Role of Vitamin D in Suppressing Cytokine Storm and Associated Mortality in COVID-19 Patients. medRxiv

2019 novel coronavirus (COVID-19) pandemic: Built environment considerations to reduce transmission

Perspective: improving vitamin D status in the management of COVID-19

Estimating the overdispersion in COVID-19 transmission using outbreak sizes outside China

Individual variation in susceptibility or exposure to SARS-CoV-2 lowers the herd immunity threshold. medRxiv

Targets of T Cell Responses to SARS-CoV-2 Coronavirus in Humans with COVID-19 Disease and Unexposed Individuals

The role of vitamin D in the prevention of coronavirus disease 2019 infection and mortality

The Incubation Period of Coronavirus Disease 2019 (COVID-19) From Publicly Reported Confirmed Cases: Estimation and Application

Assessing the Dynamics and Control of Dropletand Aerosol-Transmitted Influenza Using an Indoor Positioning System

Why herd immunity to COVID-19 is reached much earlier than thought

Evidence for probable aerosol transmission of SARS-CoV-2 in a poorly ventilated restaurant. medRxiv

Obesity in Patients Younger Than 60 Years Is a Risk Factor for COVID-19 Hospital Admission

Superspreading and the effect of individual variation on disease emergence

COVID-19 Outbreak Associated with Air Conditioning in Restaurant

A cluster randomised trial of cloth masks compared with medical masks in healthcare workers

Plasma concentrations of 25-hydroxyvitamin D among Jordanians: Effect of biological and habitual factors on vitamin D status

Vitamin-D and COVID-19: do deficient risk a poorer outcome?

Closed environments facilitate secondary transmission of coronavirus disease 2019 (COVID-19) medRxiv

Indoor transmission of SARS-Cov-2

Obesity and covid-19: the role of the food industry

High Temperature and High Humidity Reduce the Transmission of COVID-19

Effects of long-duration wearing of N95 respirator and surgical facemask: a pilot study