key: cord-0664216-fkybmonn authors: Perone, Gaetano title: The determinants of COVID-19 case fatality rate (CFR) in the Italian regions and provinces: an analysis of environmental, demographic, and healthcare factors date: 2020-08-01 journal: nan DOI: nan sha: 5ace6702e87c60f4b5f813126c898e44f2869f3a doc_id: 664216 cord_uid: fkybmonn The Italian government has been one of the most responsive to COVID-19 emergency, through the adoption of quick and increasingly stringent measures to contain the outbreak. Despite this, Italy has suffered a huge human and social cost, especially in Lombardy. The aim of this paper is dual: i) first, to investigate the reasons of the case fatality rate (CFR) differences across Italian 20 regions and 107 provinces, using a multivariate OLS regression approach; and ii) second, to build a taxonomy of provinces with similar mortality risk of COVID-19, by using the Ward hierarchical agglomerative clustering method. I considered health system metrics, environmental pollution, climatic conditions, demographic variables, and three ad hoc indexes that represent the health system saturation. The results showed that overall health care efficiency, physician density, and average temperature helped to reduce the CFR. By the contrary, population aged 70 and above, car and firm density, level of air pollutants (NO2, O3, PM10, and PM2.5), relative average humidity, COVID-19 prevalence, and all three indexes of health system saturation were positively associated with the CFR. Population density, social vertical integration, and altitude were not statistically significant. In particular, the risk of dying increases with age, as 90 years old and above had a three-fold greater risk than the 80 to 89 years old and four-fold greater risk than 70 to 79 years old. Moreover, the cluster analysis showed that the highest mortality risk was concentrated in the north of the country, while the lowest risk was associated with southern provinces. Finally, since prevalence and health system saturation indexes played the most important role in explaining the CFR variability, a significant part of the latter may have been caused by the massive stress of the Italian health system. The novel coronavirus disease (COVID-19) is a severe acute respiratory syndrome detected for the first time in December 2019 in Wuhan, Hubei province, China. On March 11, 2020, the World Health Organization (WHO, 2020) declared that COVID-19 could be characterized as a pandemic. As of July 26, 2020, according to Worldometer (2020) , the virus has spread across 213 countries and territories, affecting over 16,4 million people and causing more than 650 thousand deaths. Italy was one of the countries hit the worst by the pandemic, with almost 250 thousand confirmed cases and over 35 thousand deaths at the time of writing. Despite the widely recognized Ogen (2020) used spatial analysis to examine the relationship between long-term exposure to NO2 and COVID-19 mortality in 66 administrative regions in France, Germany, Italy, and Spain. He showed that 78% of the total COVID-19 deaths were located in north Italy and central Spain, i.e. the regions with the highest level of NO2. Bianconi et al. (2020) , using multiple linear regression models, showed that mean annual exposure to PM10 and PM2.5 was significantly and positively associated with COVID-19 death rate in Italian 20 regions. Yao et al. (2020) , by using spatial analysis and multivariate linear regression for China as of April 12, 2020, found that every 10 μg/m 3 increase in PM10 and PM2.5 concentrations is associated with a 0.24% and 0.26% increase in the COVID-19 mortality rate, respectively. Brandt et al. (2020) also stressed that air pollution levels are strongly associated with densely populated urban areas in the USA. Hamidi et al. (2020) investigated both direct and indirect impacts of population density on death rates in 913 USA metropolitan countries by using structural equation modeling. They found that larger metropolitan areas and counties with higher population density were significantly associated with higher mortality rates. In early March 2020, Bayer and Kuhn (2020) advanced the hypothesis that Italian higher vertical social integration may arise the COVID-19 fatality rate. They used a sample of 24 countries with at least 200 COVID-19 confirmed cases and found a positive correlation between the share of the population aged 30-49 living with parents and COVID-19 death rate. However, this relation was strongly criticized by Belloc et al. (2020) , which showed how the sign of the correlation turned negative when considering the variation within Italian regions. Finally, several studies have shown that the presence of at least one comorbidity, such as hypertension, diabetes, cardiovascular disease, and chronic lung disease, negatively affects outcomes of patients hospitalized with COVID-19 (Guan et al., 2020; Istat-ISS, 2020; Jordan et al., 2020; Wang et al., 2020; Wu and McGoogan, 2020) . The aim of this study is to estimate the main determinants of COVID-19 mortality rate at the peak of the virus outbreak in 20 Italian regions and 107 Italian provinces. Specifically, I chose the peak of mild and severe cases of COVID-19, which was approximately reached on April 3-4, 2020 ( Figure 2 ), because it can be considered as the moment of maximum health care system saturation. First, I used a multivariate cross-sectional OLS (ordinary least squares) approach to identify the main determinants at regional at province level, and then I applied the Ward's hierarchical agglomerative clustering method (Ward 1963) to build a "taxonomy" of provinces with similar mortality risk of COVID-19. The rationale behind the choice of cross-sectional regression methodology instead of a panel approach is as follows: i) comprehensive daily data on March 2020 are not currently available for environmental pollution and climatic variables; and ii) demographic and health system variables change very little or not at all in the short term. At regional level, I used the following 17 explanatory variables (Table 1) : an overall index (IPS) of the Italian health system performance in 2017-2018, the public health expenditure per capita in the period 2015-2017, the total specialist doctors and general practitioners per 1,000 inhabitants in the period 2016-2018, the total ordinary hospital beds per 1,000 inhabitants in the period 2016-2018, an index of car and firm density in 2015-2017, the electric power consumption (kWh per capita) in the period 2016-2018, 1 the proportion of population aged 70 and over in 2019, the proportion of population aged 80 and over in 2019, the proportion of population aged 90 and over in 2019, the vertical social integration proxied by the share of adults aged 18-34 living with their parents in 2019, the average relative humidity levels registered during March 2020, the average historical diurnal temperature range (DTR) in March, the average historical temperature in March, the average prevalence of COVID-19 on April 3 and 4, 2020, the ratio between the COVID-19 prevalence and the ordinary hospital beds, an two ad hoc indexes that indicate the saturation of ordinary hospital beds (OB) and critical care beds (CCB) at the peak of the outbreak, respectively. The dependent variable is represented by the average case fatality rate of COVID-19 on April 3 and 4, 2020. 2 At the province level, I used the following 18 explanatory variables (Table 2) : the average general practitioners per 1,000 inhabitants in the period 2015-2017, the historical average temperature in March, the historical average diurnal temperature range (DTR) in March, an ordinal index of the population structure (rural-intermediate-urban) in 2013, the population density (people per sq. km) in 2019, the proportion of population aged 70 and over in 2019, the proportion of population aged 70-79 in 2019, the proportion of population aged 80-89 in 2019, the proportion of population aged 90 and over in 2019, the average emissions (in µg/m 3 ) of particulate matter less than 10 micrometers in diameter (PM10) in 2017-2018, the number of days in which PM10 exceeded the legal limit of 50 µg/m 3 , 3 the average emissions (in µg/m 3 ) of particulate matter less than 2.5 micrometers in diameter (PM2.5) in the period 2017-2018, the average emission of nitrogen dioxide (NO2) expressed in µg/m 3 in 2017-2018, the number of days in which ozone (O3) exceeded the limit of 120 µg/m 3 , 4 the average altitude of the capital city of each province, the prevalence of COVID-19 on March 31, 2020, and the ratio between the COVID-19 prevalence on March 31 and the average number of ordinary hospital beds in in the period 2016-2018. The dependent variable is represented by the case fatality rate of COVID-19 on March 31, 2020. 5 Further details are provided in Tables 1 and 2. 6 The OLS equation estimated at the regional and provincial level was as follows: = 0 + 1 1 +. . . + [1] Where i represented the regions/provinces, 0 was the intercept, X1 to Xn were the independent variables for each region/province, and εi was the random error. Then, I conducted a cluster analysis on selected significant variables obtained from multivariate OLS on province data. The procedure encompassed the following 4 sequential steps. First, since variables are not on the same scale, I calculated the standardized data matrix by using the following formula: = − [2] Where x was the value of the variable in the original dataset, μ was the arithmetic mean of the original variable, and σ was the standard deviation of the latter. Second, I computed the Euclidean distance. Given two points X and Y in d dimensional space, the Euclidean distance between X and Y was equal to: Then, I applied Ward's hierarchical clustering method, which allowed to obtain clusters of provinces with features as similar as possible by minimizing the total within-cluster variance. Specifically, at each step, the pair of clusters that are characterized by minimum between-cluster distance were merged. Therefore, Ward's method merging cost formula between two clusters, p and q, was given by: [4] 2 According to Baud et al. (2020) and Scheiner et al. (2020) , there is a delay between infection and death of about 14 days. However, on one hand, the prevalence on March 19-21 and April 3-4 had an almost perfectly positive correlation of 0.96 (elaboration on data from http://www.salute.gov.it), and on the other, the CFR increased exponentially until the peak, and then it has continued to grow much more slowly (https://ourworldindata.org/grapher/coronavirus-cfr). 3 The legal limit is laid down in Directive 2008/50/EC (https://eur-lex.europa.eu/legalcontent/en/ALL/?uri=CELEX%3A32008L0050). 4 The average emissions of air pollutants for each province have been calculated by using the annual data coming from a maximum of 266 urban and suburban monitoring stations, spread all over the country (Istat, 2019) . 5 Istat released only the overall death for March 31, April 30, and May 31, 2020. 6 Some descriptive statistics are also reported in the Supplementary Material (Tables S1 and S2). From this, it was obtained: Where mj was the center of cluster j, and nj was the overall number of points included in cluster j. In Table 3 (a, b), I presented the OLS estimations at the regional level. All the 12 estimated OLS models were statistically significant; in fact, the Fisher-Snedecor distribution assumed values far higher than the tabulated critical values at the 1% level of significance. In particular, the r-square showed that models were able to explain from 0.44% to 0.88% of CFR variability. Furthermore, since Breusch-Pagan (1979) and Shapiro-Wilk (1965) tests allowed to accept the null hypothesis of homoscedasticity and normality of residuals, models seemed well specified. However, due to the small sample, I preferred to adopt a conservative approach, by applying the HC2 correction proposed by MacKinnon and White (1985) . 7 It is important to stress that OLS cross-sectional analysis is very sensitive to the presence of outliers, which can cause misspecification issues (Mur and Lauridsen 2007) , especially in such small samples (Wooldridge 2015, p. 334) . Therefore, I investigated the presence of highly influential points. The analysis revealed that none of the leverage points (h) are beyond the cutoff value (ℎ > 2 / ) proposed by Belsley et al. (1980) . Finally, I also identified the presence of multicollinearity. The variance inflation factors (VIF) ranged from 1.46 to 8.28 8 and were less than 10, i.e. the rule of thumb suggested by the relevant literature (Belsley 1982; Hair et al. 1995) . Therefore, I concluded that the independent variables just suffered from weak linear dependency and there were no serious multicollinearity issues. The results showed that the IPS index and physicians were significantly and negatively associated with COVID-19 CFR. 9 By the contrary, health expenditure, car and firm density, people aged 70 and above, humidity, DTR, prevalence, the ratio of prevalence/ordinary beds, and saturation indexes of ordinary and critical care beds were significantly and positively correlated with the CFR. Hospital beds, kWh per capita, and social vertical integration showed no statistical significance. Even if the sign of health expenditure is unexpected, the IPS index seemed to confirm the importance of health system effectiveness in all its dimensions more than only health expenditure. The regression coefficients also gave interesting information. The change in 1 unit of physician per 1,000 inhabitants and IPS index was correlated on average with a change of -2.87% and -0.53% in the CFR, respectively. 10 Consistent with Du et al. (2020) , Ioannidis et al. (2020), Sannigrahi et al. (2020) , the risk of dying increased progressively with the increase in the elderly population. In fact, the regression coefficients of the population aged 90 and above were 4 times larger than that for the population aged 80 and above, and 8 times larger than that for the population aged 70 and above. Moreover, each 1 unit increase in humidity and DTR was positively associated on average with a change of 0.28% 11 and 2.22% in the CFR. The results for humidity are consistent with a recent study by Bianconi et al. (2020) on the Italian case, while DTR outcomes are in line with Ma et al. (2020) . Prevalence and health system saturation indexes were highly significant in all the models. In particular, saturation of ordinary beds had the largest coefficient, followed by COVID-19 7 As suggested by Erving and Long (2000, p. 220) , in presence of homoskedasticity, HC2 option has good small size properties. 8 I did not consider VIF for average temperature, which was constantly greater than 10. It is due to this that average temperature has been excluded in most part of the models. 9 In addition, in four OLS models, the average electric power consumption is negatively associated with the CFR. However, its coefficients are statistically significant only at the 10% level. 10 From now on, I only considered the average of the significant coefficients. 11 Model 10 also seemed to suggest that the relationship between humidity and CFR is non-monotonic. In other words, the CFR decreased for low-medium levels of humidity and increased for high values of humidity. However, since the lack of comprehensive data on humidity, further investigations are necessary. prevalence, the ratio prevalence/beds, and the saturation of intensive care beds. They played the most important role in explaining CFR variability; in fact, the average r-square of models (5-12) with prevalence and saturation indexes was 0.81, i.e. almost twice the models (1-4) without them (r = 0.46). In particular, the change in 0.1 unit of CCB and OB saturation was correlated with a change of 0.47% and 4.54% in the CFR. Therefore, a significant part of the CFR may be caused by the massive stress of the Italian health system. Finally, the correlation matrix in Figure S1 indicated that the average temperature and humidity were significantly and negatively correlated to the COVID-19 prevalence, with a Pearson's r of -0.8 and -0.46, respectively. By the contrary, the kWh per capita and car and firm density showed a good positive correlation with the latter, with a Pearson's r of 0.66 and 0.52, respectively. In Table 4 (a, b), I presented the OLS estimations at the provincial level. All 11 estimated OLS models were statistically significant; in fact, the Fisher-Snedecor distribution assumed values far higher than the tabulated critical values at 1% level of significance. In particular, r-square showed that models were able to explain from 0.27% to 0.45% of CFR variability. Since the Breusch-Pagan (1979) test revealed heteroscedasticity issues, I applied the HC2 correction. As stated by Ghasemi and Zahediasl (2012) , the violation of normality assumption should not be a major problem in a sample with enough observations (n > 40), such as in this case. Furthermore, the variance inflation factors (VIF) ranged from 1.05 to 2.56 and are much less than 10; thus, there were no multicollinearity issues. The OLS models showed that general practitioners, average temperature, and DTR were significantly and negatively associated with the CFR. 12 By the contrary, all the considered air pollutants (PM10, PM2.5, NO2, and O3), people aged 70 and above, prevalence, and saturation index for ordinary beds were significantly and positively correlated with the CFR. Urbanization, population density, and altitude showed no statistical significance. The change in 1 degree Celsius of temperature and DTR was correlated with an average change of -0.53% and -2.44% in the CFR, respectively. If the results for the temperature are consistent with other studies (Bianconi et al. 2020; Rahman et al. 2020; Wu et al. 2020b) , the sign of DTR is unexpected and in contrast with previous findings. Therefore, the relative impact is ambiguous and unclear. Among air pollutants, PM2.5 and PM10 had the largest impact on COVID-19 CFR, followed by NO2 and O3. The change in 10 μg/m 3 of PM2.5 and PM10 was associated with a change of 3.13% and 2.54% in the CFR, respectively. The change in 10 μg/m 3 of NO2 was "only" correlated with a change of 1.27% in the CFR. While, each day in which O3 exceeded the limit of 120 µg/m 3 , the CFR have increased of 0.11%. The inclusion of O3 is very significant, since it allowed r-square to increase to 0.42. The positive and significant relationship between air pollutants and CFR is consistent with the relevant literature (Bianconi et al., 2020; Ogen, 2020; Pansini and Fornacca, 2020; Wu et al., 2020a; Yao et al., 2020) . As found at the regional level, the CFR increased progressively with population age. Specifically, 90 years old and above had a three-fold greater risk than the 80-to-89 years old and four-fold greater risk than 70-to-79 years old. This result is consistent with Onder et al. (2020) and assumes great importance, since the susceptibility to the COVID-19 is constant across all age groups (Istat-ISS, 2020) . The prevalence and saturation indexes were highly and significantly correlated with the CFR in all the considered models (9-12). Every 0.1 unit increase in COVID-19 prevalence and saturation of hospital ordinary beds was associated with an average change of 1.66% and 0.44% in the CFR, respectively. Moreover, the inclusion of prevalence and saturation in the models allowed an increase in the average rsquare from 0.31 to 0.43. This is consistent with the previous findings. The correlation matrix ( Figure S2 ) gave other interesting information, especially regarding air pollutants. Air pollutants were highly and positively correlated with each other, with a Pearson's r that ranged from 0.48 to 0.84. Moreover, consistent with Brandt (2020), population density was moderately and positively correlated with air pollutants. Finally, as found in other studies (Bianconi et al. 2020; Pansini and Fornacca 2020; Setti et al. 2020; Zhu et al. 2020) , COVID-19 prevalence and air pollutant showed a good positive correlation, with a Pearson's r ranged from 0.36 to 0.65. In Figure 3 , I presented the dendrogram obtained using Ward's method. The optimal number of clusters was identified by using two different methods: i) the EM (Expectation-Maximization) algorithm for Gaussian finite mixture model, proposed by Fraley et al. (2012) in the package 'Mclust' (R environment); and iii) the package 'NbClust' (R environment) that proposes 30 different indexes. 13 In the first case, according to BIC (Bayesian Information Criterion) score, the best model was VVI (varying volume, varying shape, and equal orientation) with 3 clusters. The majority of indices (7 out of 30) included in 'NbClust', also proposed 3 clusters. Therefore, by cutting the dendrogram at an approximately height of 13, I obtained three clusters with an increasing risk of mortality (Table 5) . Each cluster was identified with a grey dotted rectangle ( Figure 3 ) and represented graphically through a map ( Figure 4 ). The cluster with the highest mortality risk (risk = 3) was composed by 25 provinces of northern Italy, 1 province of central Italy (Pesaro and Urbino), and 2 provinces of southern Italy (Frosinone and Avellino). The cluster with the medium mortality risk (risk = 2) was composed by 22 provinces of northern Italy, 18 provinces of central Italy, and 6 provinces of southern Italy (Chieti, L'Aquila, Oristano, Pescara, South Sardinia, and Teramo). Finally, the cluster with the lowest mortality risk (risk = 1) was composed by 2 provinces of central Italy (Latina and Rome), and 31 provinces of southern Italy (Table 6) . Therefore, the highest mortality risk was concentrated in the north of the country, while the lowest risk was associated with southern provinces. Specifically, cluster 3 had an 8.62% higher CFR than cluster 1, and 5.3% higher CFR than cluster 2. Cluster 3 had a temperature of 2.58 degrees lower and a number of general practitioners per 1,000 inhabitants of 0.24 lower than that for cluster 1. Moreover, cluster 3 had a PM10 of 11.23 μg/m 3 higher and a proportion of population aged 70 and abobe of 1.49% higher than that for cluster 1. Finally, the hospital bed saturation in cluster 3 was more than double than that for cluster 2, and 8 times greater than that for cluster 1. The results are consistent with Grasselli et al. (2020) , according to which Lombardy's intensive care units already had an 85% to 90% occupancy ahead of the outbreak. 14 To the best of my knowledge, this is one of the first studies to investigate the relationship between a wide set of heterogeneous factors and COVID-19 mortality. The OLS analysis showed that environmental, demographic, and healthcare factors play an important role in explaining the CFR variability. In particular, population aging, air pollutants (NO2, O3, PM10, and PM2.5), relative humidity, COVID-19 prevalence, and critical care and ordinary beds saturation are positively correlated with the CFR. By the contrary, overall health care efficiency (IPS), physician density, and average temperature are negatively associated with CFR. Specifically, the inclusion of the COVID-19 prevalence and saturation indexes of ordinary and critical care beds explains up to 86% of the CFR variability. Therefore, a significant part of the CFR variability may be caused by the massive stress of the Italian health system. The results are robust across several model specifications. Moreover, cluster analysis showed that the highest mortality risk was concentrated in northern Italy, while the lowest risk was associated with southern provinces. However, this study also has some limitations that can be summarized as follows: i) first, a significant part of the patients died in hospital presented at least one comorbidity ahead of 13 The outcomes of both methods are reported in the Supplementary Material ( Figure S3 and Table S3 ). 14 In fact, Lombardy was the hardest hit region on April 3-4, 2020, with 40% of the Italian confirmed cases. COVID-19 infection (ISS, 2020) ; ii) then, as pointed out in other studies (Spalt et al., 2016; Wang et al., 2019) , the utilization of air pollution implies unavoidable measurement errors since most people usually stay indoors; and iii) finally, climatic variables, such as average temperature and DTR, refer just to the average historical values. Finally, the study seemed to stress the importance of implementing quick and rational lockdown measures, of making patients comfortable, of implementing an action plan to discourage car use and decrease firm's pollution, and of buying ad hoc health care facilities, medical equipment, and devices to adequately tackle similar and unforeseeable emergencies. 16 The parameters used are the following: patient satisfaction, active patient mobility, passive patient mobility, legal fees for disputes, operating result, life expectancy, equality in health treatment, and economic hardship. In particular, each parameter is standardized, with mean = 100 and standard deviation = 10, and the final synthetic index is obtained by calculating the simple average of them. 17 Data are available at URL: http://dati.istat.it/. 18 The number of cars refers to those recorded in the Pubblico registro automobilistico (Public vehicle register). The number of largest firms (> 250 employees) refers to those that operate in the following sectors: (1) mining and minerals from quarries and mines; (2) manufacturing activities; (3) supply of electricity, gas, vapors, and air conditioning; and (4) supply sewerage, waste management and remediation activities. The index is compiled according to the following analytical method: i) first, I standardized the data according to surface area (cars and firms for 100 sq. km.); ii) than, the respective outputs are switched to fixed-base indexes (with mean = 100); iii) finally, I computed the simple arithmetic mean of the latter. Ages 90+ The proportion of population aged 90 and over for each region, in 2019. The share of unmarried young adults aged 18-34 living with at least one parent for each region, in 2019. I.Stat (database) The average relative humidity levels registered during March 2020, for each region. 19 The historical diurnal temperature range in March, for each region. www.il meteo.it 20 Mipaaf (2019a) Temperature The historical average temperature in March, for each region. Mipaaf (2019a) Prevalence The average ratio between the people who have been tested positive for COVID-19 and the overall population of each region on April 3 and 4, 2020. The ratio between the average COVID-19 prevalence on April 3 and 4, 2020, and the average number of ordinary hospital beds in 2016-2018, for each region. I.Stat (database), Ministry of Health (2020) Ministry of Health (2020) CCB saturation The ratio between the average people who have been recovered from COVID-19 in intensive care on 3 and 4 April 2020, and the average number of critical care beds (CCB) in the period 2016-2018, for each region. Health (2020) OB saturation The ratio between the average people who have been recovered from COVID-19 with mild symptoms on 3 and 4 April 2020, and the average number of ordinary hospital beds in the period 2016-2018 for each region. Ministry of Health (2020) An ordinal index that ranks population of each province by urban-rural structure: predominantly rural (1), intermediate (2) MacKinnon and White (1985) . Significance level: p-value < 0.01***; p-value < 0.05**: p-value < 0.1*. MacKinnon and White (1985) . Significance level: p-value < 0.01***; p-value < 0.05**: p-value < 0.1*. MacKinnon and White (1985) . Significance level: p-value < 0.01***; p-value < 0.05**: p-value < 0.1*. Standard errors (in brackets) are based on HC2 method developed by Real estimates of mortality following COVID-19 infection. The Lancet infectious diseases Intergenerational ties and case fatality rates: A cross-country analysis. Institute of Labor Economic (IZA) Cross-country correlation analysis for research on COVID-19 Assessing the presence of harmful collinearity and other forms of weak data through a test for signal-to-noise Regression diagnostics: Identifying influential data and sources of collinearity Particulate matter pollution and the COVID-19 outbreak: results from Italian regions and provinces These Are the World's Healthiest Nations Air pollution, racial disparities, and COVID-19 mortality Air pollution reduction and mortality benefit during the COVID-19 outbreak in China. The Lancet Planetary Health Air Pollution and COVID-19: The Role of Particulate Matter in the Spread and Increase of COVID-19's Morbidity and Mortality Can atmospheric pollution be considered a co-factor in extremely high level of SARS-CoV-2 lethality in Northern Italy? Environmental Pollution Indice di misurazione e valutazione dei sistemi regionali italiani La performance sanitaria. Indice di misurazione e valutazione dei sistemi regionali italiani Predictors of mortality for patients with COVID-19 pneumonia caused by SARS-CoV-2: a prospective cohort study NUTS3 regions mclust version 4 for R: normal mixture modeling for model-based clustering, classification, and density estimation Association of long-term exposure to community noise and traffic-related air pollution with coronary heart disease mortality Normality tests for statistical analysis: a guide for non-statisticians Critical care utilization for the COVID-19 outbreak in Lombardy, Italy: Early experience and forecast during an emergency response Comorbidity and its impact on 1590 patients with Covid-19 in China: A Nationwide Analysis Multivariate Data Analysis, 3 th Does Density Aggravate the COVID-19 Pandemic? Population-level COVID-19 mortality risk for non-elderly individuals overall and for non-elderly individuals without underlying diseases in pandemic epicenters Characteristics of COVID-19 patients dying in Italy Report based on available data on Tavole di dati. Ambiente Urbano Impatto dell'epidemia COVID-19 sulla mortalità totale della popolazione residente primo trimestre 2020 Covid-19: risk factors for severe disease and death Asymptomatic and presymptomatic SARS-CoV-2 infections in residents of a long-term care skilled nursing facility Coronavirus Disease 2019 (COVID-19) in Italy Using heteroscedasticity consistent standard errors in the linear regression model Effects of temperature variation and humidity on the death of COVID-19 in Wuhan Some heteroskedasticity-consistent covariance matrix estimators with improved finite sample properties Osservatorio agroclimatico, statistiche regionali Osservatorio agroclimatico, statistiche provinciali Outliers and spatial dependence in cross-sectional regressions Assessing nitrogen dioxide (NO2) levels as a contributing factor to the coronavirus (COVID-19) fatality rate Case-fatality rate and characteristics of patients dying in relation to COVID-19 in Italy COVID-19 higher induced mortality in Chinese regions with lower air quality Temperature Extreme May Exaggerate the Mortality Risk of COVID-19 in the Low-and Middle-income Countries: A Global Analysis. Preprints The overall mortality caused by covid-19 in the european region is highly associated with demographic composition: A spatial regression-based approach Mathematical modeling of COVID-19 fatality trends: Death kinetics law versus infection-to-death delay rule SARS-Cov-2RNA Found on Particulate Matter of Bergamo in Northern Italy: First Evidence Time-location patterns of a diverse population of older adults: the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air) Dati storici Estimates of the severity of coronavirus disease 2019: a model-based analysis. The Lancet infectious diseases Clinical characteristics of 138 hospitalized patients with 2019 Novel Coronavirus-Infected pneumonia in Wuhan Association between long-term exposure to ambient air pollution and change in quantitatively assessed emphysema and lung function Hierarchical grouping to optimize an objective function Characteristics of and important lessons from the coronavirus disease 2019 (COVID-19) outbreak in China: summary of a report of 72314 cases from the Chinese Center for Disease Control and Prevention Exposure to air pollution and COVID-19 mortality in the United States Effects of temperature and humidity on the daily new cases and new deaths of COVID-19 in 166 countries Association of particulate matter pollution and case fatality rate of COVID-19 in 49 Chinese cities Association between short-term exposure to air pollution and COVID-19 infection: Evidence from China