key: cord-0811651-5hakexo7
authors: Isaia, Giancarlo; Diémoz, Henri; Maluta, Francesco; Fountoulakis, Ilias; Ceccon, Daniela; di Sarra, Alcide; Facta, Stefania; Fedele, Francesca; Lorenzetto, Giuseppe; Siani, Anna Maria; Isaia, Gianluca
title: Does solar ultraviolet radiation play a role in COVID-19 infection and deaths? An environmental ecological study in Italy
date: 2020-11-20
journal: Sci Total Environ
DOI: 10.1016/j.scitotenv.2020.143757
sha: 540164e94bb1a1bd5e2c0fcbc7fc398893c83284
doc_id: 811651
cord_uid: 5hakexo7

A significantly stronger impact in mortality and morbidity by COVID-19 has been observed in the northern Italian regions compared to the southern ones. The reasons of this geographical pattern might involve several concurrent factors. The main objective of this work is to investigate whether any correlations exist between the spatial distribution of COVID-19 cases and deaths in the different Italian regions and the amount of solar ultraviolet (UV) radiation at the Earth's surface. To this purpose, in this environmental ecological study a mixed-effect exponential regression was built to explain the incidence of COVID-19 based on the environmental conditions, and demographic and pathophysiologic factors. Observations and estimates of the cumulative solar UV exposure have been included to quantify the amount of radiation available e.g., for pre-vitamin D3 synthesis or SARS-CoV-2 inactivation by the sunlight. The analysis shows a significant correlation (p-value <5 × 10−2) between the response variables (death percentage, incidence of infections and positive tests) and biologically effective solar UV radiation, residents in nursing homes per inhabitant (NHR), air temperature, death percentage due to the most frequent comorbidities. Among all factors, the amount of solar UV radiation is the variable contributing the most to the observed correlation, explaining up to 83.2% of the variance of the COVID-19 affected cases per population. While the statistical outcomes of the study do not directly entail a specific cause-effect relationship, our results are consistent with the hypothesis that solar UV radiation impacted on the development of the infection and on its complications, e.g. through the effect of vitamin D on the immune system or virus inactivation by sunlight. The analytical framework used in this study, based on commonly available data, can be easily replicated in other countries and geographical domains to identify possible correlations between exposure to solar UV radiation and the spread of the pandemic.

importance of other environmental (Sect. 3.4.1), comorbidity and demographic factors (Sect. 3.4.2 and 3.4.3) . The results of our analysis, reported in Sect. 4, and further discussed in Sect. 5, may provide a significant step towards supporting the role of solar UV radiation in the COVID disease transmission and clinical outcomes. Furthermore, the statistical setup employed in this study can be applied to other series and geographical domains in an easy and effortless way, being it based on publicly available data, in order to provide, and to motivate deeper and more focused investigations.

This section presents some of the main studies about the biological effects of UV radiation with reference to the present pandemic (Sect. 2.1), and some examples of ecological studies related to COVID-19 and including different environmental variables (Sect. 2.2).

While an exhaustive review is out of the scope of this article and can be found elsewhere (e.g., Dobricic et al., 2020) , here we offer a quick overview of the current scientific knowledge with the aim of framing the problem and motivating the present study.

Solar UV radiation may affect the spread of the SARS-CoV-2 virus and the outcomes of COVID-19 in patients through multiple pathways. First of all, high levels of solar radiation, especially in the UV-B region (formally defined in the wavelength range 280-315 nm, but mainly detectable at wavelengths larger than 290 nm at the Earth's surface, due to absorption by the atmosphere), might reduce the dissemination of the virus, notably SARS-CoV-2, by acting as a disinfectant for contaminated nonporous materials, as found by Sagripanti and Lytle (2020) . Additionally, Ratnesar-Shumate et al. (2020) detected inactivation of SARS-CoV-2 on surfaces in culture media when exposed to simulated sunlight representative of the summer solstice at 40°N latitude at sea level on a clear day, and mainly attributed this process to UV-B radiation. Herman et al. (2020) used the data from both previous studies to determine the inactivation times from 290-315 nm UV-B solar radiation for SARS coronaviruses CoV and CoV-2 using global satellite data. The exposure duration needed to reduce the virus particles by 90% depend on the assumed inactivation J o u r n a l P r e -p r o o f Journal Pre-proof effective doses and the type of virus, ranging from ca. 90 minutes for SARS CoV viruses for midlatitude sites between March and September, to even less for SARS CoV-2.

Second, the exposure to UV radiation favours the synthesis of vitamin D in the body. In fact, exposure to UV radiation determines the photo-conversion of the pro-vitamin D3 (7dehydrocholesterol) in the skin to pre-vitamin D3 (Norval et al., 2010) . Subsequently, it meets two distinct hydroxylations, the first in the liver, where it is converted to 25(OH)D, and the second in the kidney, with synthesis of its active form (1-25OH 2 D 3 or calcitriol).

Finally, it binds to specific receptors (VDRs), acting on different tissues with a similar hormonal mechanism.

Among other effects, such as the regulation of calcium and phosphate metabolism and the possible prevention of several diseases (Maretzke et al 2020) , vitamin D plays a well-established role in the innate and adaptative immune defense of viral and bacterial infections (Calder et al, 2020) . In this perspective, a recent review (Grant et al., 2020) has highlighted that vitamin D is able to keep the integrity of the tight junctions and the pulmonary barrier, by its antiviral properties and its possible role in mitigating pneumonia and hyper inflammation (Borella et al., 2014) . These properties are thought to originate from the vitamin D ability to modulate gene expression by activating the VDRs, ubiquitously distributed in many target cells, including immune cells, and by promoting the expression of antimicrobic peptides such as cathelicidins and beta-defensins, which also have antiviral and immunomodulatory activity (Tripathi et al., 2015) . Indeed, VDRs have been found on both resident immune cells and respiratory epithelial cells (Lange et al., 2009 ) and a metanalysis, considering interventional studies in more than 11,000 patients, reported that vitamin D supplements reduce by two-thirds the incidence of acute respiratory infections (not induced by in patients with low levels of blood serum 25(OH)D (Martineau et al., 2019) .

Some observational outcomes endorse the idea that vitamin D (and hence an environmental component such as solar UV radiation) may also play a role in the prevention and development of the SARS-CoV-2 pandemic event (Isaia and Medico, 2020 (Rhodes et al., 2020) , this hypothesis is supported by circumstantial evidences that associate the outcomes of COVID-19 to vitamin D levels (Mitchell, 2020) : specifically, a correlation between vitamin D status and COVID-19 incidence and death rates has been reported in twenty European countries. Notably in Spain, Italy and Switzerland severely low Vitamin D levels in the aging and vulnerable populations were found (Ilie et al., 2020) . The terrestrial amounts of solar UV radiation mediated by vitamin D status could also explain differences in susceptibility of particular ethnic groups: this occurred, for example, in Connecticut (USA), where African-American population known at higher hypovitaminosis D risk (O'Connor et al., 2013) , have a higher rate of infection and death by COVID-19, in comparison to lighter-skinned people at a comparable latitude (Laurencin et al., 2020) . Moreover, a retrospective, observational analysis of deidentified tests was performed in over 190,000 patients from all US 50 states to determine if circulating 25-hydroxyvitamin D [25(OH)D] levels were associated with severe acute respiratory disease coronavirus 2 (SARS-CoV-2) positivity rates: SARS-CoV-2 positivity was strongly and inversely associated with circulating 25(OH)D levels, and this relationship persisted across latitudes, races/ethnicities, both sexes, and age range. (Kaufman et al., 2020) . Finally, a very interesting study demonstrated that administration of a high dose of 25-hydroxyvitamin D, a main metabolite of vitamin D endocrine system, significantly reduced the need for Intensive Care Unit treatment of patients requiring hospitalization due to proven COVID-19 (Castillo et al 2020) .

Since more than 90% of the vitamin D in the majority of individuals is effectively produced by solar UV radiation (mainly in the UV-B band) and only a modest supplement can be provided from food (though less than 20%), there are large seasonal differences in its production mainly caused by the variation of solar elevation and different skin exposure to UV radiation (Webb et al 2010) . This is particularly true for Italy, where UV-B irradiance reveals large differences during the year, with winter months characterized by an UV irradiance about tenfold lower than summer months (Calgani et al., 2016) ; the same study J o u r n a l P r e -p r o o f Journal Pre-proof reports, as a consequence, that the seasonal stratification of vitamin D concentrations revealed an evident trend with the minimum mean value recorded in April and a maximum mean value obtained in September. Therefore, blood sampling seasonality should be regarded as an important preanalytical factor in vitamin D assessment and the amount of total vitamin D synthesized during the summer should be high enough to maintain the levels > 50 nmol/L throughout the remaining part of the year (Bonelli et al., 2016) . Although Italy is located in southern Europe, a large part of the elderly is lacking vitamin D (Lips et al., 2019) . In particular, previous research proved that more than 70% of the elderly women in Italy have low levels of vitamin D (Isaia et al, 2003) . Unfortunately, no recent 25(OH)D analyses are available on the whole Italian territory, which compelled us to use the exposure to ambient solar UV radiation as an alternative proxy of the vitamin D status in this study.

Several studies carried out throughout the world have attempted to establish a relationship between one or more environmental factors and the spread (or the consequences) of the present pandemic. The most frequently studied environmental variables are temperature and humidity. Iqbal et al. (2020) for example, explored the temporal correlation between daily average temperature and daily new COVID-19 cases in Wuhan, China, in the period between January and March 2020, and found a positive association between both datasets.

On the other hand, based on data collected in the same period in the 10 most affected provinces in China, Shahzad et al. (2020a) showed that the relationship between temperature and COVID-19 actually differs for different provinces, the correlation between the temporal series being mostly positive for some areas and negative for others. Sajadi et al. (2020) found that the initial spread of the virus occurred on a narrow latitude belt, roughly located between 30°N and 50°N, characterised by specific conditions of temperature (5-11 °C) and humidity (4-7 g/m³), consistently with the behaviour of a seasonal respiratory virus. In a recent study, Bherwani et al. (2020) included temperature and relative humidity in their epidemiological model for India, which also combined socio-J o u r n a l P r e -p r o o f Journal Pre-proof behavioural aspects such as social distancing and lockdown regulations. They found fair correlation with temperature, compatible with disruption of the lipid layer of coronavirus at higher temperatures, but undermined by the most influencing factor of social distancing and confounded by other co-varying environmental factors. No clear role of relative humidity was established.

Air quality is another factor potentially impacting on the spread and impacts of the pandemic (Sciomer et al., 2020) . Particulate matter (PM), for example, might act as a carrier of the virus, whose RNA was found in PM samples collected in northern Italy, the epicenter of the pandemic in Europe (Setti et al. 2020) , or contribute, together with other atmospheric pollutants, to comorbidity and worsening the prognosis (Fattorini and Regoli, 2020) , by triggering inflammatory response at molecular, cellular and organ levels (Mescoli et al., 2020) . Fareed et al. (2020) correlated the air quality index and average relative humidity with COVID-19 mortality in Wuhan between January and March 2020, showing a negative relationship between air quality and deaths and relative humidity and deaths. For the same period, Bianconi et al. (2020) investigated the spatial correlation between PM and demographic factors with the total cases and deaths for the different Italian regions and provinces using adjusted regression models, and found significant association between exposure to PM, COVID-19 incidence and death. An even wider dataset, encompassing PM2.5, PM10, nitrogen dioxide, carbon monoxide, benzene, sulfur dioxide and ozone was used by Cazzolla Gatti et al. (2020) to train a machine learning algorithm and predict total cases and deaths in the Italian regions and provinces up to June 2020. Air pollution was found to be the main predictor for the effects of SARS-CoV-2, an increase of pollution by 5-10% being responsible of an increase by 19-28% of the cases and by 4-14% of the deaths.

Using daily concentrations of carbon monoxide, nitrogen dioxide, sulfur dioxide, PM2.5, and 2020, in addition to positive correlation of the cases with relative humidity and negative correlation with temperature. Shahzad et al. (2020b) , correlated daily averages of temperature and PM2.5 with daily new cases of COVID-19 (up to July 2020) in the 4 most populated regions of Spain, showing that cases decrease at higher temperature and lower PM2.5. Generally speaking, although different research methods are used in the scientific literature and various confounding factors are not always taken into account, the major findings are consistent and highlight the contribution of pollutants such as PM2.5 and nitrogen dioxide in enhancing the COVID-19 spread and lethality (Copat et al., 2020) . Sehra et al. (2020) explored the association between COVID-19 incidence and maximum daily temperature, precipitation, and solar UV radiation in the USA and showed that a higher UV index was associated with a lower incidence rate. Using a more recent dataset, Tang et al. (2020) found significant spatial and temporal negative correlation in the USA between COVID-19 (and other coronaviruses) cases and UV radiation (weekly and monthly radiation weighted by the erythemal, vitamin D and DNA damage action spectra and integrated over a given time period). Finally, Fazzini et al. (2020) examined the temporal relationships between several climate parameters, including solar radiation, and J o u r n a l P r e -p r o o f new daily positive swabs for COVID-19 for the Lombardy region in the period from March to April 2020. The multiple linear model resulting from the variables that explain at least 3% of the total variability showed that the spread of the pandemic is favoured by increased sunshine (in contrast to most studies) and high relative humidity, while the contribution of temperature is negligible. To the best of our knowledge, no further studies relating COVID-19 cases and solar radiation, notably in the UV band, in Italy have been conducted so far.

Our multidisciplinary study involves data of solar UV ambient exposure (i.e., energy of the solar radiation reaching the Earth's surface over a specified time period per unit area) to quantify the amount of radiation necessary to activate pre-vitamin D3 synthesis. These are closure of schools of every order on the national territory (4 th March 2020, marker 2), of retail business and restaurants (9 th -11 th March 2020, marker 3), and of all other "unnecessary" activities (21 th March 2020, marker 4). The "phase 2" started on 4 th May 2020 (marker 5), with the progressive easing of the regulations, culminating in returning to the free circulation across the regional boundaries on 3 rd June 2020 (marker 6). The period chosen in this study to investigate the evolution of COVID-19 cases and outcomes (response variables, Sect. 3.3) spans the interval between 25 th February and 31 st May 2020, i.e. it includes the whole "phase 1" and the beginning of "phase 2", and encompasses J o u r n a l P r e -p r o o f 97% of the total cases in the "first wave" of the pandemic in Italy and 96% of the total deaths. In other words, these data describe the impact of the pandemic on a rather uniform population that was not prepared for an event of this size and strength. We refrained from considering the summer period, since the number of the cases is greatly reduced (thus increasing the statistical noise), and other variables might have altered the evolution of the pandemic. Among them, the mean age of the affected decreased (younger people were affected by coronavirus during the summer period), better treatments in hospitals and better preparation of the people (frailty populations have been more protected by the social isolation and distancing and by the use of masks). Notably, the mortality observed from summer to autumn is probably altered by these provisions. Fig. 1 also shows that, although the day-to-day variability of solar UV radiation can be large due to the impact of clouds, the seasonal course (envelope) of the UV exposure is rather smooth. Moreover, since a long exposure interval (several months) is required in order for cholecalciferol to accumulate in the fat tissues of the human body (Sect. 2.1), the expected effects from vitamin D will be modest and difficult to interpret on the temporal scale of the current duration of the pandemic. Indeed, based on previous research, e.g. in Great Britain (Hyppönen and Power, 2007) and the United States (Kroll et al., 2015) , the average 25(OH)D levels in the population are projected to increase by no more than 20-30% in the period February-May considered in the present study. Incidentally, this also justifies the positive correlation (contrasting to most studies) between new daily positive swabs and UV exposure found by Fazzini et al. (2020) , in the too short period between March and April 2020. For this reason, no temporal correlation between the response variables and the environmental, demographic and pathophysiologic factors was attempted in this study, which is based instead on spatial correlations.

Different univariate models were built as linear regressions of the logarithm base 10 (Sajadi et al., 2020) of the response variables (listed in Section 3.3 and also reported in Table 1 ), as a function of each of several factors aggregated on a regional basis (independent J o u r n a l P r e -p r o o f Journal Pre-proof variables, Section 3.4 and Table 2a -b-c). For each independent variable, the parameters of the regression were fitted with the least-squares method and the statistics of the regression analysis were calculated. Multivariate analyses were also performed in order to determine how large is the increase of the explained variance when using multiple predictors, and the logarithm base 10 of every response variable was fitted by means of a multiple linear regression to several independent variables simultaneously. In this case, the independent variables were selected as the subset of variables that resulted as statistically significant in the univariate analysis (significance threshold value set to 0.05). The analysis was performed in the MATLAB R2019b environment.

The study was conducted considering the available clinical outcomes of all 20 Italian regions. We employed these official data, aggregated for each region, in order to investigate the COVID-19 outcome variations throughout the national territory. In particular, we selected the following clinical data: (i) the percentage of positive pharyngeal nose swabs, for viral RNA detection, relative to the population of the corresponding region, (ii) the percentage testing positive for COVID-19 relative to the number of swab tests performed regionally, (iii) the COVID-19 death percentage relative to the population of the corresponding region. The data (reported in Table 1) 

Several parameters were considered in the regression as independent variables, described in the next paragraphs in detail.

Italy is a Mediterranean country which extends in a wide latitudinal zone, ranging from about 47 to 36 °N. Its topography is very complex, mainly because of the presence of the J o u r n a l P r e -p r o o f Apennines extending from the north to the south, and the Alps at the north, the latter surrounding the wide Po basin. The large altitude gradients due to the presence of such high mountains and the latitudinal extent of more than 10° affect the meteorological conditions which vary significantly throughout the country, even within short horizontal distances of a few kilometers, resulting in a correspondingly large variability in the levels and the spectral distribution of the solar UV irradiance that reaches the surface (Meloni et al., 2000) . Topography also impacts air quality, by limiting the exchange of air masses and creating permanent atmospheric pollution hotspots, such as the Po basin (actually one of the strongest in Europe, also due to the very populated urban settlements), or, conversely, by favouring winds contributing to the dispersion of pollutants, e.g. close to the coastline.

The large variability of the environmental data among the different regions is clearly visible in Table 2a . These factors, described below in more detail, were aggregated for each Italian region as spatial averages, after weighting their original 2-D fields over the respective distribution of the population density (European Commission, Global Human Settlement, GHS-POP product release R2019A).

 Solar UV irradiance at the Earth's surface.

In the absence of recent clinical data assessing the distribution of 25(OH)D in Italy, we considered the ambient UV exposure spectrally weighted with the pre-vitamin D3 action curve, as defined by the International Commission on Illumination (CIE) (Bouillon et al., 2006) , as a proxy to quantify the capability of ambient UV radiation to initiate the synthesis of pre-vitamin D3 in the human skin the resulting quantity will be shortened hereafter to "vitamin D UV exposure". Although the solar spectral irradiance is greater in the UV-A region (formally defined between 315 and 400 nm), the CIE action spectrum for pre-vitamin D3 peaks at much shorter wavelengths. This makes the 300-320 nm spectral range (mostly to the expanded uncertainties of most of the used instruments and within the uncertainty of satellite products). Overall, the dependence of UV radiation on the latitude of the Italian stations, also found in previous research (Siani et al., 2013; Fountoulakis et al., 2020b) , is well captured by TEMIS (Fig. 3) .

 Air temperature and relative humidity.

To account for possible influences of air temperature and relative humidity on the pandemic and its effects (Sect. 2.2), the time interval considered for these two variables is the same as the clinical data collection period, i.e. February-May 2020. For a better spatial coverage, J o u r n a l P r e -p r o o f the air temperature and relative humidity at 2 m above the surface were obtained for the whole Italian territory from the analysis of a numerical prediction model, COSMO 2I. Owing to its high resolution (about 2.2 km grid step) and its accurate physical scheme (nonhydrostatic, fully compressible), COSMO 2I allows a good representativeness of the meteorological fields even on complex terrains (Diémoz et al., 2019) . A detailed description of COSMO can be found elsewhere (Baldauf et al., 2011) .

Concentration measurements of airborne particulate matter with a diameter of 10 µm or less (named as PM 10 ) were downloaded from the website of the European Environment Agency (https://www.eea.europa.eu/data-and-maps/dashboards/air-quality-statistics-expertviewer). These consist in yearly regional averages from more than 500 stations in Italy, which were further averaged on the 2015-2019 period to account for the possible long-term effects of chronic exposure to PM 10 on comorbidities (e.g., Fattorini and Regoli, 2020) . The geographical pattern of PM 10 mostly arises from the environmental, social and economic characteristics of the different Italian regions, the rate of industrialization, and the population density. Since these are permanent or slowly changing features, the resulting geographical distribution is expected not to change even during the shorter period of the pandemic in Italy and could thus account for other possible short-term effects on human health and virus transmission.

For the second group of variables three chronic pathologies ( 

J o u r n a l P r e -p r o o f Finally, the third group includes the mean age, NHR per 100 inhabitants and percent mortality rate (demographic factors). These data (Table 2c) were provided by the National Institute for Statistic (ISTAT).

The analysis of the regression on the percentage testing positive for COVID-19 relative to the population of the corresponding region (hereafter named as affected/pop) and on the COVID-19 death percentage relative to the population of the corresponding region (deaths/pop) yielded very similar results (Tables 3a-3b ). For both response variables, the univariate analyses of the estimated coefficients showed significant correlations (p-value<5×10 -2 ) only with the following factors: the ambient vitamin D UV exposure, air temperature, NHR, the death percentage due to circulatory system diseases, to cerebrovascular diseases and to diabetes mellitus. An inverse relationship was found for all these factors, the only exception being NHR, which showed a direct relationship, i.e.

COVID-19 deaths percentage increased with an increase in NHR. The observed relationships will be discussed in the next section.

The same tables show that mean age variations for each Italian region did not significantly correlate with affected/pop (p=1.1×10 -1 ) and deaths/pop (p=1.6×10 -1 ). In particular, the mean age among the Italian regions did not show significant differences, thus preventing this factor to be statistically significant when correlated to COVID-19 infections and deaths.

In other words, the geographical variation of the mean age in Italy did not drive the geographical distribution of COVID-19 cases in a significant way. We also tested different metrics to describe the age distribution in each region, such as percentage of people older or younger than a specified threshold, and assess their correlation with the response variables, however the overall conclusion does not change, i.e. the population is about the same age throughout Italy.

The effect size of a specific factor is quantified as the percent change of the response variable resulting from a 1% increase of the mean of the factor itself, over the whole Italian territory. The vitamin D UV exposure had the larger effect size, both for affected/pop, -8.6%

J o u r n a l P r e -p r o o f (95% CI: -10.4% to -6.8% ) and death/pop, -10.3% (95% CI :-13.0% to -7.4%). The vitamin D UV exposure had also the largest percentage of variation explained, 83.3% and 74.2%

for affected/pop and death/pop, respectively, meaning that the univariate regression based on this factor showed the best goodness of fit ( Fig. 4a-b) . Interestingly enough, the analysis of the regression coefficient proved that, among the regional demographic factors, only the NHR had a significant effect on the response variables, with p-value <5×10 -4 for both response variables.

To assess whether the inclusion of all significant factors led to a large increase of the explained variance, both for the affected/pop and death/pop response variables, multivariate analyses were performed on the subsets of factors presented above, considering only those with p-value<5×10 -2 . It is important to highlight that the multivariate regression is not intended as a predictive model, since the goal of this regression is just to compare the maximum % of variation explained using all the relevant independent variables to the one from the univariate models. The combined effect of the vitamin D UV exposure, air temperature, NHR, death percentage due to circulatory system diseases, to cerebrovascular diseases and to diabetes mellitus for each region on the response variables was analyzed. In terms of % of variation explained, it is interesting to notice that the combined effect of multiple factors did not substantially improve the goodness of fit of the regressions. In fact, the formulated multivariate regressions had a total explained variation of 78.5% for death/pop and of 85.5% for affected/pop, i.e. only around 5% more than the values obtained with the univariate regressions with the vitamin D UV exposure.

Considering all the available factors in the multivariate analyses, i.e. including those with pvalue > 5×10 -2 from the univariate analysis, thus likely overfitting the response variables, obviously gave the maximum explained variations achievable, equal to 91.1% and to 95.8%

for death/pop and affected/pop, respectively. These values were roughly 15% higher than the variations explained by the vitamin D UV exposure in the univariate regressions.

The analysis of the regression coefficients on the percentage testing positive for COVID-19

relative to the number of swabs performed in the corresponding region (affected/swabs %)

showed an overall worse index of goodness of fit (fifth column, % of variation explained, in Table 3c ), compared to the other response variables (fifth column, % of variation explained,

in Tables 3a and 3b ).

The univariate analysis of the estimated coefficients only showed significant correlations (p-value<5×10 -2 ) between the response variable and the vitamin D UV exposure, NHR, and the death percentage due to circulatory system diseases and to diabetes mellitus. The relationships between the response variable and the factors were the same as the other response variables presented above. The effect sizes of these factors on affected/swabs were roughly half the values of the effect sizes of the same factors on death/pop and on affected/pop. For this response variable as well, the vitamin D UV exposure had the largest percentage of variation explained, even though its value was as low as 41.4%. This analysis revealed that air temperature, death percentage due to vascular diseases and to diabetes mellitus did not have a significant correlation with affected/swabs, as opposed to affected/pop and death/pop. The multivariate regression had an explained variation of 43.8%, that was almost equal to the value obtained from the univariate regression with the vitamin D UV exposure as the only independent variable.

As shown in the previous section, the most effective variables explaining the geographical distribution of infections and mortality from COVID-19 in Italy were vitamin D UV exposure , temperature, NHR, and some chronic diseases (cardio and cerebrovascular diseases, and diabetes mellitus). These are discussed hereafter.

First of all, the inverse relation between COVID-19 cases and vitamin D UV exposure found in this study is in line with most previous studies. For example, it was shown through regression modeling that, as of May 2020, countries in the Northern hemisphere were manifesting relatively high COVID-19 mortality, with an estimated 4.4% increase in mortality per degree of latitude, north of 28°N (p=3.1×10 -2 ), after adjustment for age of population (Rhodes et al., 2020 (Radujkovic et al., 2020) promiscuous environments, such as nursing homes, are exposed to a higher risk than private homes.

The inverse relation between the number of deaths and common comorbidities was unexpected and deserves separate discussion. Considering the "past" comorbidity deaths (the latest data available referring to the year 2017) as a proxy for cardiovascular risk, the statistical analysis revealed that both COVID-19 death and incidence are lower in those regions where cardiovascular risk is higher. Therefore, our regionally-averaged data reveal that comorbidities alone are not able to justify the geographical distribution of COVID-19

incidence and death. In support of this important finding, the data recently published by the National Institute for Statistics (ISTAT) and by the Italian National Institute of Health (ISS) (https://www.istat.it/it/files//2020/07/ReportISS_Istat_Cause-di-morte-Covid.pdf) demonstrate that COVID-19 is directly responsible (from January to May 2020) for 89% of the deaths of people testing positive, while only a small fraction (11%) of the deaths is caused by cardiovascular diseases (4.6%), tumors (2.4%), respiratory illness (1%), type 2 diabetes (0.6%), dementia (0.6%), gastrointestinal diseases (0.5%) and other causes (1.3%). However, it should be also considered that a severe comorbidity rather than single illnesses alone could better explain a cause-effect relation with COVID-19, although current data do not allow to confirm or deny this statement. Future epidemiological investigations are necessary to address this question. Comorbidities must of course be taken into account as important factors when different age classes are considered, also given that they are frequently associated with hypovitaminosis D.

Finally, a short comment should be made about the lower variation explained by the regression describing affected/swabs. Although, in principle, this ratio should represent the best proxy of the infections and reduce the bias due to the different number of swabs performed in each region ("the more you search for, the more you will find"), the tests were performed following different regional strategies, in different ways and on different groups (e.g. old people, nursing home residents, doctors or nurses, relatives of infected etc.).

Furthermore, one person could be tested several times during the course of the disease.

These issues introduce large uncertainties in the infected/swabs ratio.

More generally, the results obtained by our study were affected by a number of uncertainties. First of all, it is well known that, in general, all observational retrospective ecological studies may suffer from ecological fallacy. Moreover, possible geographical differences in the initial number of infected people, and different regional impacts of the lockdown measures could not be excluded and might have contributed to the observed correlation. Also, the input data, e.g. the number of infections, might not be entirely accurate since, as already mentioned, the tests may not be uniformly distributed within the Italian territory. Likewise, the number of deaths officially reported might have been underestimated due to organizational reasons, although this bias could affect the whole territory in a similar way. Even the independent variables, such as the vitamin D UV exposure, are affected by uncertainties. For example, the ambient UV exposure differs from the UV exposure received by anatomical sites (i.e. personal exposure) which depends upon different postures under several environmental conditions and the duration of exposure (Schmalwieser, 2020) . Furthermore, the exact shape of the effective pre-vitamin D3 action spectrum and the conditions under which it should be used are still under debate (Norval et al., 2010) . Finally, the previtamin D3 levels are also determined, for the same personal exposure, by the phototype and other personal characteristics, which might not be as uniform in Italy as in other countries (Yeum et al., 2016) . However, despite the existing uncertainties, and in the absence of recent countrywide results from screening campaigns of the levels of 25(OH)D in Italy, the spatial variability of the cumulative solar UV exposure obtained from TEMIS is expected to be a sufficiently representative proxy of the spatial variability of the vitamin D adequacy in the population, as already proven by previous studies (Kelly et al., 2016 , O'Sullivan et al., 2017 , Prodam et al, 2016 . Large scale analyses of the levels of 25(OH)D in Italians are however necessary in order to further J o u r n a l P r e -p r o o f clarify and definitely prove this assumption, and to discriminate the effect by vitamin D from the enhanced SARS-COV-2 inactivation by UV radiation.

In this study, we tested the correlation between the presently COVID-19 available clinical data for every Italian region and several environmental, comorbidity and demographic factors, obtained from public sources. Although the analysis can be further improved using more refined statistical techniques, such as non-linear models or machine learning algorithms, the main outcomes of the study are not expected to change significantly, also given the relatively low number of response variables (equal to the number of the Italian regions, for each regression).

Whilst the results do not directly imply a specific cause-effect relationship, they support the hypothesis that exposure to solar UV radiationtogether with other factors such as social, demographic, and environmental conditionsaffects SARS-CoV-2 transmission and its outcomes. The statistical setup and the available data did not allow us to discriminate the protective effect of vitamin D, synthesized after exposure to the sun, and the effect of Therefore, public health campaigns should be promoted to increase consumption of vitamin D rich food or properly controlled pharmaceutical supplementation, and to encourage more adequate sunlight exposure, especially in countries exposed at the risk of vitamin D insufficiency or deficiency (Lips et al. 2019; Isaia et al., 2003) .

Since the statistical framework employed in the present study relies on commonly available data, it can be easily applied to other contexts and geographical domains. Follow-up research could include, for example, an extension of the present analysis to the whole European continent.

Funding: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors J o u r n a l P r e -p r o o f J o u r n a l P r e -p r o o f J o u r n a l P r e -p r o o f

Investigation of effective climatology parameters on COVID-19 outbreak in Iran Science of

Operational Convective-Scale Numerical Weather Prediction with the COSMO Model: Description and Sensitivities

Exploring dependence of COVID-19 on environmental factors and spread prediction in India

Particulate matter pollution and the COVID-19 outbreak: results from Italian regions and provinces

Vitamin D supplementation and musculoskeletal health

Blood sampling seasonality as an important preanalytical factor for assessment of vitamin D Status

Vitamin D: a new anti-infective agent?

Action spectrum for the production of previtamin D3 in human skin

Active monitoring of persons exposed to patients with confirmed COVID-19 -United States

Optimal nutritional status for a well-functining immune system an important factor to protect against viral infections

Serum 25(OH)D seasonality in urologic patients from central Italy

Effect of Calcifediol Treatment and best Available Therapy versus best Available Therapy on Intensive Care Unit Admission and Mortality Among Patients Hospitalized for COVID-19: A Pilot Randomized

Machine learning reveals that prolonged exposure to air pollution is associated with SARS-CoV-2 mortality and infectivity in Italy

The role of air pollution (PM and NO 2 ) in COVID-19 spread and lethality: A systematic review

25-Hydroxyvitamin D concentrations are lower in patients with positive PCR for SARS CoV-2 Nutrients

Transport of Po valley aerosol pollution to the northwestern Alps -Part 1: Phenomenology

First national intercomparison of solar ultraviolet radiometers in Italy

Do environmental factors such as weather conditions and air pollution influence COVID-19 outbreaks

Investigating the Effects of Meteorological Parameters on COVID-19: Case Study of New Jersey, United States Environ Res 30

Co-variance nexus between COVID-19 mortality, humidity, and air quality index in Wuhan, China: New insights from partial and multiple wavelet coherence

Role of the chronic air pollution levels in the Covid-19 outbreak risk in Italy Environmental Pollution

Preliminary Analysis of Relationships between COVID19 and Climate, Morphology, and Urbanization in the Lombardy Region

Monitoring of solar spectral ultraviolet irradiance in Aosta Italy

Solar UV Irradiance in a changing climate: trends in Europe and the significance of spectral monitoring in Italy

The possible roles of solar ultraviolet-B radiation and vitamin D in reducing case-fatality rates from the 1918-1919 influenza pandemic in the United States Dermato-Endocrinology

Evidence that Vitamin D supplementation could reduce risk of location-specific temperature and UV index Journal of Public Health: From Theory to Practice

Inactivation times from 290 to 315 nm UVB in sunlight for SARS coronaviruses CoV and CoV-2 using OMI satellite data for the sunlit Earth

Hypovitaminosis D in British adults at age 45 y: nationwide cohort study of dietary and lifestyle predictors

Low Temperature and Low UV Indexes Correlated with Peaks of Influenza Virus Activity in Northern Europe during

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

The nexus between COVID-19, temperature and exchange rate in Wuhan city: New findings from partial and multiple wavelet coherence

Prevalence of hypovitaminosis D in elderly women in Italy: clinical consequences and risk factors

Associations between hypovitaminosis D and COVID-19: a narrative review

SARS-CoV-2 circulating 25-hydroxyvitamin D levels PLOS ONE

The contributions of adjusted ambient ultraviolet B radiation at place of residence and other determinants to serum 25-hydroxyvitamin D concentrations

Temporal Relationship between Vitamin D Status and Parathyroid Hormone in the United States PLOS ONE

Vitamin D, the immune system and asthma

First detection of SARS-CoV-2 in untreated wastewaters in Italy

The COVID-19 Pandemic: a call to action to identify and address racial and ethnic disparities

Current vitamin D status in European and Middle East countries and strategies to prevent vitamin D deficiency: a position statement of the European Calcified Tissue Society

Spectral character of sunlight modulates photosynthesis of previtaminD3 and its photoisomers in human skin

Aging decreases the capacity of human skin to J o u r n a l P r e -p r o o f produce vitamin D3

Role of Vitamin D in preventing and treating selected extraskeletal diseases-An Umbrella Review

Vitamin D supplementation to prevent acute respiratory infections: individual participant data meta

Solar UV Dose Patterns in Italy

The secretive liaison of particulate matter and SARS-CoV-2. A hypothesis and theory investigation Front

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

Is the action spectrum for the UV-induced production of previtamin D3 in human skin correct?

Guidelines for the management of osteoporosis and fragility fractures

The uncertain significance of low vitamin D levels in african descent populations: a review of the bone and cardiometabolic literature

Ambient UVB dose and sun enjoyment are important predictors of Vitamin D Status in an Older Population

Influence of ultraviolet radiation on the association between 25-hydroxy vitamin D levels and cardiovascular risk factors in obesity

Vitamin D Deficiency and Outcome of COVID-19 Patients

Simulated Sunlight Rapidly Inactivates SARS-CoV-2 on Surfaces

Perspective: Vitamin D deficiency and COVID-19 severity -plausibly linked by latitude, ethnicity, impacts on cytokines, ACE2, and thrombosis

Estimated Inactivation of Coronaviruses by Solar Radiation With Special Reference to COVID-19

Temperature, humidity, and latitude analysis to estimate potential spread and seasonality of Coronavirus Disease

Possibilities to estimate the personal UV radiation exposure from ambient UV radiation measurements

SARS-CoV-2 spread in Northern Italy: what about the pollution role? Environ Monit Assess

Maximum Daily Temperature, Precipitation, Ultra-Violet Light and Rates of Transmission of SARS-Cov-2 in the United States Clin Infect Dis May 30;ciaa681

Searching for SARS-COV-2 on Particulate Matter: A Possible Early Indicator of COVID-19 Epidemic Recurrence

Asymmetric nexus between temperature and COVID-19 in the top ten affected provinces of China: A current application of quantile-on-quantile approach

Effects of climatological parameters on the outbreak spread of COVID-19 in highly affected regions of Spain

Fluctuations in environmental pollutants and air quality during the lockdown in the USA and China: two sides of COVID-19 pandemic

Biologically effective surface UV climatology at Rome and

Sunlight ultraviolet radiation dose is negatively correlated with thepercent positive of SARS-CoV-2 and four other common humancoronaviruses in the U

Antiviral Activity of the human Cathelicidin, LL-37, and derived peptides on seasonal and Pandemic influenza A viruses

TEMIS UV index and UV dose operational data products, version 2. Royal Netherlands Meteorological Institute (KNMI)

A UV-Index Climatology for Europe Based on Satellite Data

The role of sunlight exposure in determining the vitamin D status of the U.K. white adult population

Impact of geographic location on Vitamin D status and bone mineral density

TEMIS UV product validation using NILU-UV ground-based measurements in Thessaloniki

Writing, review & editing. Henri Diémoz: Data curation, Methodology, Formal analysis, Writing, review & editing. Ilias Fountoulakis Formal analysis, Validation. Francesco Maluta: Formal analysis

Data curation, Writing, review & editing. Anna Maria Siani: Data curation, Writing, review & editing. Daniela Ceccon: Data curation. Stefania Facta: Data curation

Data curation. Giuseppe Lorenzetto: Data curation. Gianluca Isaia: Supervision, Writing original draft

The authors would like to thank the Aosta Valley weather forecast office (Centro 

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.J o u r n a l P r e -p r o o f