Potential role of particulate matter in the spreading of COVID-19 in Northern Italy: first observational study based on initial epidemic diffusion
Public health
Original research
Leonardo Setti1, Fabrizio Passarini1, Gianluigi De Gennaro2, Pierluigi Barbieri3, Sabina Licen3, Maria Grazia Perrone4, Andrea Piazzalunga5, Massimo Borelli3, Jolanda Palmisani2, Alessia Di Gilio2, Emanuele Rizzo6, Annamaria Colao7, Prisco Piscitelli8, Alessandro Miani9
A number of studies have shown that the airborne transmission route could spread some viruses over a distance of 2 meters from an infected person. An epidemic model based only on respiratory droplets and close contact could not fully explain the regional differences in the spread of COVID-19 in Italy. On March 16th 2020, we presented a position paper proposing a research hypothesis concerning the association between higher mortality rates due to COVID-19 observed in Northern Italy and average concentrations of PM10 exceeding a daily limit of 50 µg/m3.
Methods To monitor the spreading of COVID-19 in Italy from February 24th to March 13th (the date of the Italian lockdown), official daily data for PM10 levels were collected from all Italian provinces between February 9th and February 29th, taking into account the maximum lag period (14 days) between the infection and diagnosis. In addition to the number of exceedances of the daily limit value of PM10, we also considered population data and daily travelling information for each province.
Results Exceedance of the daily limit value of PM10 appears to be a significant predictor of infection in univariate analyses (p<0.001). Less polluted provinces had a median of 0.03 infections over 1000 residents, while the most polluted provinces showed a median of 0.26 cases. Thirty-nine out of 41 Northern Italian provinces resulted in the category with the highest PM10 levels, while 62 out of 66 Southern provinces presented low PM10 concentrations (p<0.001). In Milan, the average growth rate before the lockdown was significantly higher than in Rome (0.34 vs 0.27 per day, with a doubling time of 2.0 days vs 2.6, respectively), thus suggesting a basic reproductive number R0>6.0, comparable with the highest values estimated for China.
Conclusion A significant association has been found between the geographical distribution of daily PM10 exceedances and the initial spreading of COVID-19 in the 110 Italian provinces.
http://creativecommons.org/licenses/by-nc/4.0/
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Introduction
COVID-19 (due to the new SARS-CoV-2 virus), is known to spread via respiratory droplets and close contact.1 However, this unique transmission model does not seem to explain properly the different initial spreading of the virus observed in Italy from February 24th 2020 to March 13th 2020. The huge virulence of COVID-19 in the Po Valley is not comparable to the milder contagiousness observed in the Central-Southern regions. Demographic factors related to the ageing of the population and the possibility of infection without clinical symptoms for a quite long time—together with the high rate of asymptomatic people that characterises COVID-19 (estimated at 50%–75% of infections)—may only partially explain the fast spreading of the virus in Lombardy and Northern Italy.2 3 Cai et al3 reported different incubation periods in patients infected in Wuhan,4 but an epidemic model based only on respiratory droplets and close contact could not fully explain the regional differences in the spread of COVID-19 in Italy, which was fast and dramatic only in Lombardy and Po Valley. At the same time, a number of studies have shown that the airborne transmission route could spread viruses over a distance of 2 meters from an infected person.5–19 Paules et al4 highlighted that—besides close distance contacts—airborne transmission of SARS-CoV can also occur.5 It has been reported that for some pathogens airborne transport can reach long distances.6–8 Reche et al8 described the aerosolisation of soil-dust and organic aggregates in sea spray that facilitates the long-range transport of bacteria, and likely of viruses, free in the atmosphere.
In particular, virus deposition rates were positively correlated with organic aerosol <0.7 µm, implying that viruses could have longer persistence time in the atmosphere and, consequently, can be dispersed further.8 Moreover Qin et al analysed the microbiome of the airborne particulate matter (PM2.5 and PM10) in Beijing over a period of 6 months in 2012 and 2013, putting in evidence variability of the composition that depended on the months analysed.9 Temporal distribution of the relative abundance of the microbiome on PM showed the highest presence of viruses in January and February, just in coincidence with most severe pollution due to PM. Chen et al demonstrated the relationship between short-term exposure to PM2.5 concentrations and incidence of measles in 21 cities in China.10 Their meta-analyses showed that the nationwide measles incidence was significantly associated with an increase of 10 µg/m3 in PM2.5 levels.
Other recent studies have also reported associations between PM and infectious diseases (eg, influenza, haemorrhagic fever with renal syndrome) as inhalation could bring PM deep into the lungs, and viruses attached to particles may directly invade the lower part of the respiratory tract, thus enhancing the induction of infections, as demonstrated by Sedlmaier et al. 11
Zhao et al showed that the majority of the patients positive for the highly pathogenic avian influenza H5N2 in Iowa (USA) in 2015 might have been infected by airborne viruses carried by fine PM from infected farms, both within the same state and from neighbouring states.12 Ma et al13 observed a positive correlation between the incidence of measles and PM10 in western China during the period 1986–2005; the condensation and stabilisation of the bioaerosol, generating aggregates with atmospheric particles from primary (ie, dust) and secondary particulates, have been indicated as mechanisms able to transport airborne bacteria and viruses to distant regions, even by the intercontinentally transported dust.13
Ferrari et al14 showed measles outbreaks occurring in dry seasons and disappearing at the onset of rainy seasons in Niger,14 while Brown et al15 found that the most severe measles epidemic in the USA occurred in Kansas in 1935 during the Dust Bowl period.15 Coming to recent specific studies, laboratory experiments by van Doremalen et al16 indicated that airborne and fomite transmission of SARS-Cov-2 is plausible, since the virus can remain viable and infectious in aerosol for several hours.16 On-field measurement carried out by Liu et al showed evidence of coronavirus RNA in indoor air samples from Wuhan hospitals and even in ambient air in close proximity, during the COVID-19 outbreak, highlighting the airborne route as a possible important pathway for contamination that should undergo further confirmations.17
Santarpia et al reported the presence of airborne SARS-COV-2 in indoor air samples at the Nebraska University Hospital,18 while—the opposite—some negative evidence of presence of virus in air reported by Ong et al19 comes from explicitly poor sampling schemes.19 Recently, we have published the first evidence in the world of the presence of COVID-19 on outdoor PM in samples collected between 23 February and 9 March in the province of Bergamo (Lombardy, Italy), which experienced the highest diffusion and mortality rates in Italy.20
A research carried out by the Harvard School of Public Health seems to confirm an association between increases in PM concentration and mortality rates due to COVID-19.21 On March 16th 2020, we have released an official position paper highlighting that there is enough evidence to consider the airborne route as a possible additional factor for interpreting the anomalous COVID-19 outbreaks notified in Northern Italy, known to be one of the European areas characterised by the highest PM concentrations.22 23
This article presents the data that led to the publication of the position paper and triggered high interest in the research community working on the hypothesis of a possibility of further transmission via airborne dust,24–26 taking into account the possibility that the potential survival of the virus could be influenced by climatic parameters such as humidity and temperature as well as by fine dust concentrations.27
Other papers support the possible merging of contaminated aerosol with fine particulate in the atmosphere.28–30
The concentration of fine particles has also been repeatedly recognised by other authors as an important cofactor causing higher mortality rates in heavily contaminated areas.31 32
This study is aimed at searching for a possible association between the initial spread of COVID-19 in Italy, registered from the end of February to the first weeks of March 2020 (February 24th to March 13th), and the frequency of high daily average concentrations of PM, recorded before the lockdown, taking into account the lag period of the infection (February 9th to February 29th). The research hypotheses that we addressed is the possibility that air pollution could produce a ‘boost effect’ of the COVID-19 epidemic, thus representing a kind of an exceptional ‘super-spread event’.
Methods
In the frame of an observational design of the study, we have analysed daily data relevant to ambient PM10 levels, urban conditions and COVID-19 incidence from all Italian provinces, in order to reliably determine the association between PM pollution levels and the initial spread of COVID-19 in Italy. Daily PM10 concentrations were collected by the official air quality monitoring stations of the Regional Environmental Protection Agencies, publicly available on their websites. The number of daily PM10 limit value exceedances (50 µg/m3) detected in the different provinces, divided by the total number of PM10 monitoring stations for each selected province was computed.
Population data related to each Italian province were collected from the National Institute for Statistics for all the 110 Italian provinces,33 paying specific attention to the absolute number of inhabitants and their density (number of inhabitants/km2 for each province) as well as to the number of commuters (people travelling from other provinces for job reasons) and its proportion with respect to the province population.
We have computed the number of COVID-19 infected people for each province and the infection rate based on the number of inhabitants from February 24th to March 13th (the date when the lockdown was decided), as reported by the official government website, updated with daily frequency.34
The number of PM exceedances were computed between February 9th and February 29th, as we had to take into account the maximum lag period of 14 days, which is the average time elapsed between the contagion and the first weeks of the Italian epidemic (February 24th to March 13th). Further covariates related to the different provinces have also been considered: the number of air quality monitoring stations available in each province, and the longitude and the latitude of the province city centre.
All the provinces have been assigned to two geographical areas (Northern or Southern Italy). The data set is publicly available on our web page35 along with statistical analyses reproducible code in R language.36
To investigate how PM exceedances might relate to infection diffusion, we started performing an exploratory analysis on exceedances of PM10 considering the recursive partitioning tree approach, as implemented into the party package.37
Such implementation connects the exploratory techniques to the classical statistical test approach, presenting the advantage to exploit a motivated stopping criterion when pruning the tree (ie, the p value of a significance test on independence of any covariate and response).38 Within recursive partitioning analyses, the response variable was represented by the proportion of COVID-19 cases over province population; the log-transforms of responses of such proportions were reported in figures. Cut-offs identified by recursive partitioning tree analyses were subsequently used in binomial generalised linear models, both univariate and multivariable (ie, logistic regression).
The response of the binomial generalised linear models is expressed as a two-dimensional vector39 obtained by binding the number of COVID-19 cases and the rest of the province population. In the presence of overdispersion, quasi-binomial distributions were addressed. When suitable, association in the contingency table has also been expressed in terms of ORs, and Fisher’s exact test was used to assess statistical significance. Exploratory analyses of exceedance rates of PM2.5 were held by the recursive partitioning tree approach too. Correlation between PM2.5 and PM10 exceedance rates per province have been addressed by using a linear model.
Pearson’s coefficient was applied to evaluate correlation; diagnostics plots were used to assess model adequacy.
Similarly, we performed statistical inference analyses on data from Milan and Rome, in order to observe the potential association between PM levels and spread of COVID-19 in big cities located in different geographical areas with remarkable differences in exceedances of PM10, but presenting at the same time similar urbanisation, lifestyle, population, ageing index and number of commuters.
The Roman municipality is far more extensive, with 1287 km2 of surface area compared with just 182 km2 of Milan. Talking about population, Rome has 2.87 million inhabitants compared with 1.35 million of Milan, but it is much less densely populated: 2232 inhabitants/km2 versus 7439 (mainly due to the huge extension of the city of Rome). However, an additional one million inhabitants live in Roman neighbourhoods with an average density between 6720 and 9231 inhabitants/km2. The underground extensions of Milan and Rome are currently about 98 km and 54 km, respectively. Looking at the numbers of annual visitors, Rome has about 29.0 million tourists per year, compared with 12.1 million people visiting Milan.
The number of daily commuters (people moving to Rome due to working reasons or similar) is higher in Rome (2.04 million trips) compared with Milan (1.66 million trips).Patient and public involvement
Patient and public involvement
Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Results
The spatial distribution of ambient PM10 exceedances between Italian cities was geographically heterogeneous and it is presented in figure 1A.
The highest number of exceedances were generally located in Northern Italian regions suffering from a rapid diffusion of the COVID-19 epidemic, while zones with a lower contagion were located in the central and southern regions. The maps shown in figure 1 illustrate the mean values of exceedances of PM10 on the number of PM10 stations in all Italian provinces during the period February 9th to February 29th 2020 (figure 1A), compared with the total COVID-19 infection per province observed in the period March 3rd to March 13th (figure 1B–E). Overall, there were 17 660 of the 60.4 million inhabitants in Italy infected at the time of the study. The highest incidence rates of COVID-19 were recorded in cities located in Northern Italy, and particularly in the Lombardy region, including Milan. The lowest incidence of COVID-19 was observed in Southern Italy as well as in the Lazio region (which includes Rome).
SARS-Cov-2 has been subsequently recognised as a highly contagious virus transmitted by airborne direct contact, showing super-spread event characteristics that has pushed the Italian Government to adopt extraordinary measures (namely total lockdown) to contain the outbreak.40 In figure 2A, two main discontinuity trends are evident and can be attributed to the Italian lockdown. If continuing the observation beyond the dates of the lockdown (11 March to 13 March), it was possible—by analysing the trend of new daily COVID-19 infections—to observe a first reduction in the spreading rate of the contagion around March 22nd (reflecting the school closure ordered on March 5th) and a second one around March 28th (reflecting the lockdown ordered between March 11th and March 13th).
Figure 2
(A) Daily new COVID-19 infections in Italy from February 24th to April 4th 2020. (B) Trend of COVID-19 spread in Italy during the first 15 days of the epidemic.
As the lag period for SARS-COV-2 infection can be estimated in maximum 14 days, our study analysed the Italian outbreak before March 11th, when the incidence growth rate was showing a typical exponential trend of the spread (figure 2B). In the univariate analysis, the daily limit value of exceedances of PM10 appear to be a significant predictor (p<0.001) of infection with a cut-off value of 1.29 (figure 3). The cut-off divides the provinces into two classes, characterised by higher (n=43) and lower (n=67) PM10 concentrations, respectively: the less polluted provinces had a median 0.03 infection cases over 1000 inhabitants (first to third quartile 0.01–0.09; range 0.00–0.56), while the most polluted provinces had a median 0.26 infection cases over 1000 inhabitants (first to third quartile 0.14–0.51, range 0.00–4.92). The box plots in figure 3 are log-transformed to enhance figure readability.
CONTINUE TO READ MORE ON THE ORIGINAL SOURCE
download PDF
