key: cord-318856-f0m3wuyj
authors: Hoogeveen, Martijn J.; van Gorp, Eric C.M.; Hoogeveen, Ellen K.
title: Can pollen explain the seasonality of flu-like illnesses in the Netherlands?
date: 2020-10-22
journal: Sci Total Environ
DOI: 10.1016/j.scitotenv.2020.143182
sha: 
doc_id: 318856
cord_uid: f0m3wuyj

Current models for flu-like epidemics insufficiently explain multi-cycle seasonality. Meteorological factors alone, including the associated behavior, do not predict seasonality, given substantial climate differences between countries that are subject to flu-like epidemics or COVID-19. Pollen is documented to be allergenic, it plays a role in immuno-activation and defense against respiratory viruses, and seems to create a bio-aerosol that lowers the reproduction number of flu-like viruses. Therefore, we hypothesize that pollen may explain the seasonality of flu-like epidemics, including COVID-19, in combination with meteorological variables. We have tested the Pollen-Flu Seasonality Theory for 2016-2020 flu-like seasons, including COVID-19, in the Netherlands, with its 17.4 million inhabitants. We combined changes in flu-like incidence per 100K/Dutch residents (code: ILI) with pollen concentrations and meteorological data. Finally, a predictive model was tested using pollen and meteorological threshold values, inversely correlated to flu-like incidence. We found a highly significant inverse correlation of r(224)= -0.41 (p < 0.001) between pollen and changes in flu-like incidence, corrected for the incubation period. The correlation was stronger after taking into account the incubation time. We found that our predictive model has the highest inverse correlation with changes in flu-like incidence of r(222) = -0.48 (p < 0.001) when average thresholds of 610 total pollen grains/m3, 120 allergenic pollen grains/m3, and a solar radiation of 510 J/cm2 are passed. The passing of at least the pollen thresholds, preludes the beginning and end of flu-like seasons. Solar radiation is a co-inhibitor of flu-like incidence, while temperature makes no difference. However, higher relative humidity increases with flu-like incidence. We conclude that pollen is a predictor of the inverse seasonality of flu-like epidemics, including COVID-19, and that solar radiation is a co-inhibitor, in the Netherlands.

makes pollen less airborne, and cools the bio-aerosol down. Very high humidity levels (RH 98%) are even detrimental to pollen (Guarnieri, 2006) . An RH 98% effect on pollen could thus provide an alternative explanation as to why flu-like incidence in tropical countries is higher during the rainy season, and reduced during the rest of the year.

We hypothesize that pollen bio-aerosol has an inverse effect on flu-like incidence, including COVID-19 (see Figure 1 ), whereby pollen is known to be triggered and influenced by meteorological variables, which can then jointly explain the seasonality of flu-like incidence. This indirect explanation of the pollen effect is based on the fact that pollen bio-aerosol and UV light exposure lead to immuno-activation, and sometimes allergic symptoms, which seem to protect against flu-like viruses, or at least severe outcomes from them. The indirect pollen effect is explained by the spread of pollen bio-aerosol under sunny and dry conditions. Further, it is unknown how viral bio-aerosol and pollen bio-aerosol interact with each other in the air, and whether anti-viral phytochemicals in pollen could then play a role in an alternative explanation.

To further understand the impact of pollen as an environmental factor influencing the life cycle of flulike epidemics, the objective of this study is to determine the correlations of pollen and meteorological variables with (changes in) flu-like incidence and develop and test a discrete predictive model that combines pollen and meteorological co-inhibitors. Our main hypothesis, therefore, is that pollen is the missing link, jointly explaining with certain meteorological variables, flu-like seasonality, and that a compound threshold based factorcombining detected flu-inhibitorsis a good unified predictor of such seasonality. Regarding COVID-19, we have limited ourselves to observing whether or not COVID-19 at the tail-end of the 2019/2020 flu-like season is able to break with the flu-like seasonality pattern.

To study the relationship between pollen and flu-like incidence in the Netherlands, we used the public datasets of Elkerliek Hospital (Elkerliek.nl) about the weekly allergenic, low-level allergenic and total pollen concentrations in the Netherlands in grains/m 3 , whereby for 42 types of pollen particles the J o u r n a l P r e -p r o o f Journal Pre-proof numbers are counted and averaged per day per 1m 3 of air. The common Burkard spore trap was used, through which a controlled amount of air was ingested. The applied classification and analysis method conforms to the EAACI (European Academy of Allergology and Clinical Immunology) and the EAN (European Allergy Network) standards. Allergenic pollen includes nine types of particles that are classified as moderate (Corylus, Alnus, Rumex, Plantago and Cedrus Libani) , strong (Betula and Artemisia), or very strong allergenic (Poaceae and Ambrosia). Additionally, we included low-level allergenic pollen concentrations in addition to the allergenic ones because we assume that they may also have effects. Low-level allergenic pollen includes the other 33 particles that are classified as nonallergenic to low-level allergenic (Cupressaceae, Ulmus, Populus, Fraxinus, Salix, Carpinus, Hippophae, Fagus, Quercus, Aesculus, Juglans, Acer, Platanus, Pinus, Ilex, Sambucus, Tilia, Ligustrum, Juncaceae, Cyperaceae, Ericaceae, Rosaceae, Asteraceae, Ranunculaceae, Apiaceae, Brassicaceae, Urtica, Chenopodiaceae, Fabaceae, Humulus, Filipendula, and Indet) . Total pollen concentration is the sum of the average allergenic and low-level allergenic pollen concentrations.

Advantages of using the total pollen metric are that there are hardly any 0 values (only 3 out of 266), and we did not need to limit ourselves to just parts of the seasonal cycle, which might introduce subseasonal bias into our research. We also assumed that long-distance pollen transport is accounted for, as foreign pollen will also be counted by a pollen measuring station that works all year round. Furthermore, we used the data from the Dutch State Institute for Public Health (RIVM.nl) gathered by Nivel (Nivel. nl) about weekly flu-like incidence (WHO code "ILI" -Influenza Like Illnesses) reports at primary medical care level, per 100,000 citizens in the Netherlands. Primary medical care is the using the number of influenza-like reports per primary care unit divided by the number of patients registered at that unit. This is then averaged for all primary care units and then extrapolated to the complete population. The datasets run from week 1 of 2016 up to week 18 of 2020 (n = 226 data points) to include the recent COVID-19 pandemic at the tail-end of the 2019/2020 flu-like season. To underpin the relative importance of COVID-19: SARS-CoV-2 has been detected in the Netherlands since week 9, 2020. According to the figures of Nivel.nl (2020, see Figure 2 ), from week 13 onward SARS-CoV-2 is the outcome of the (vast) majority of positive tests for patients at primary care level with flu-like complaints, and by week 18 100% of positive tests indicate SARS-CoV-2 (other tested viruses are five Influenza A and B subtypes, RSV, Rhinovirus and Enterovirus).

Furthermore, we also included meteorological datasets from the Royal Dutch Meteorological Institute (KNMI.nl), including average relative humidity/day, average temperature/day and global solar radiation in J/cm 2 per day as an indicator of UV radiation. These datasets were obtained from the KNMI's centrally located De Bilt weather station. Next, we calculated the weekly averages for the same periods that featured in the other datasets. De Bilt is traditionally chosen as it provides an approximation of modal meteorological parameters in the Netherlands, which is a small country.

Furthermore, all major population centers in the Netherlands, which account for around 70% of the total Dutch population, are within a radius of only 60 kilometers from De Bilt. We therefore assumed in this study that the measurements from De Bilt are sufficiently representative for the meteorological conditions typically experienced by the Dutch population.

To test allergenic versus low-level allergenic pollen assumptions, against hay fever and pre-COVID-prevalence of allergic rhinitis that is more or less similar to that in Western Europe, being around 23%, and frequently undiagnosed (Bauchau & Durham, 2004) . Furthermore, it can be noted that the prevalence of allergic diseases in general in the Netherlands is around 52% (Van de Ven et al, 2006) .

Datasets were complete, except that three weekly pollen concentration measurements were missing (1.3%). This was due to a malfunctioning monitoring station during week 26 of 2016, week 21 of 2017 and week 22 of 2019. These missing measurements appeared to be completely random. We imputed missing values to avoid bias and maintain power. We used a four-week surrounding average to estimate the three missing data points and thus avoid breaking lines in visuals. We checked that the missing data has no material impact on the results by comparing these averages with the data of previous years for similar periods, and by observing whether removal from statistical tests had any effect on outcomes and conclusions.

Regarding the incidence of flu-like symptoms, we calculated the weekly change compared to the previous week (ΔILI=ILI t -ILI t-1 ). This was to obtain an indication of the flu-like epidemic life cycle progression, whereby a decline is interpreted as Ro<1 and an increase as Ro>1 (Ro is the reproduction number of flu-like viruses). Furthermore, to cater, in one time-series metric, for changes in flu-like incidence as well as for an incubation period of up to two weeks, we calculated a three-week moving average (3WMA) of changes in flu-like incidence, of which two weeks are forward looking:

(ΔILI 3WMA = (ΔILI t + ΔILI t+1 + ΔILI t+2 )/3) Thus, ΔILI 3WMA has on average a one week lag. A general advantage of a moving average is that it reduces statistical noise. It should be noted that whenever we use the term incubation time, we also mean to include reporting delay (estimated to be around 4.5 days). We have not assumed delay effects for meteorological variables or pollen concentrations, so we have not calculated moving averages for other time series.

Compared with our previous study (Hoogeveen, 2020) , there is an overlap in datasets of less than 10%. The datasets are extended by the extension in time, the addition of meteorological datasets and non-allergenic pollen, and the introduction of newly calculated variables, such as total pollen J o u r n a l P r e -p r o o f Journal Pre-proof concentration, ΔILI 3WMA , the compound predictor and the log10 transformations on pollen, ILI and the hay fever index.

We formulated the following statistical null hypotheses for falsification. H1 0 : there are no inverse correlations for total pollen concentrations with flu-like incidence (corrected for incubation period).

H2 0 : there are no inverse correlations between pollen and changes in flu-like incidence (ΔILI or corrected for incubation time: ΔILI 3WMA ).

H3 0 : there is no predictive significance of a discrete model's compound value, based on thresholds for pollen and meteorological co-inhibitors, related to changes in flu-like incidence (ΔILI 3WMA ).

To understand the role of meteorological variables, to check whetherin our datasetsmeteorological variables show their well-established effects on pollen as assumed, and to select coinhibitors:

H4 0 : meteorological variablessolar radiation, temperature and relative humidityhave no effect on pollen and/or flu-like incidence change (ΔILI 3WMA ).

Low-level allergenic pollen is sometimes known to have a slight allergenic effect. To understand how to interpret adding none-to-low-level allergenic pollen to the total pollen metric, we wanted to verify their effects on the hay fever index: H5 0 : low-level allergenic pollen has no effect on hay fever and (changes in) flu-like incidence.

Note that with the exception of H5, all hypotheses are related to potential causality: the temporal sequentiality (temporality) of the respective independent variables, and flu-like incidence corrected for incubation period. Whenever we refer to temporality, we mean to indicate that the datasets behave as if there is causality, on the understanding that statistics alone cannot prove causality in uncontrolled settings.

Variables are presented with their means (M) and standard deviations (SD).

We calculated correlation coefficients to test the hypotheses and to assess the strength and direction of relationships. As a sensitivity analysis, we also calculated the bootstrapped correlation coefficients.

We used the full datasets, to avoid sub-seasonal bias, and by extending the number of years the distortions by incidental and uncontrolled events are supposed to be minimized. However, as a second sensitivity analysis, we removed from the datasets the autumn weeks between 42 and 50, which typically show low pollen concentrations of up to 20 grains/m 3 , which are applied to analyze the main outcome (H2 0 ). Further, as a third sensitivity analysis we calculate correlations per individual time lag included in ΔILI 3WMA in relation to H2 0 .

Next, linear regression (F-test) on identified inhibitors and interactions was used descriptively to determine whether the relationship can (statistically) be described as linear, and to determine the equation using estimates and intercept values, and produce probability, significance level, F-value, and the Multiple R squared correlation to understand the predictive power of the respective inhibitor.

Standard deviations and errors, and degrees of freedom (DF) were used as input for calculating the 95% probability interval. We have reported in the text the outcome of statistical tests in APA style, adapted to journal requirements. For relationships that appear non-linearlogarithmic or exponential we have used the log10 function to transform the data if that makes the relationship appear linear, before re-applying linear regression. We have also used the log10 transformed datasets for the calculation of correlation coefficients, to correct for skewness.

Finally, we created a simple, discrete model resulting in one compound value, using selected flu-like inhibitors. This was to determine the optimal average threshold values for these inhibitors, which have between total pollen and flu-like incidence, including the first cycle of the COVID-19 pandemic.

Furthermore, we can reject H5 0 in favor of our assumption that it makes sense to also include low-level allergenic pollen concentrations in our study. Low-level allergenic pollen is inversely correlated to flulike incidence (r(221) = -0.37, p < .00001), especially when corrected for the 2 weeks incubation time (r(219) = -0.53, p < .00001).

The fact that the correlations become stronger when taking into account incubation time, implies temporality. Furthermore, we can also observe from Figure 3 that flu-like incidence starts to decline after the first pollen bursts. Moreover, flu-like incidence starts to increase sharply after pollen concentrations become very low or close to zero. This is a qualitative indication of temporality.

Furthermore, we can notice that the first COVID-19 cycle behaved according to pollen-flu seasonality, at least does not break with it.

When testing the impact on ΔILI, the weekly changes in medical flu-like incidence, the extended dataset till 2020, including COVID-19, shows a strong and highly significant inverse correlation with total pollen (r(226) = -0.26, p = 0.000063). Therefore, we can falsify the null-hypothesis (H2 0 ) that there is no inverse correlation between the weekly pollen concentrations and weekly changes in flulike incidence (ΔILI), including the period covering the first cycle of the COVID-19 pandemic. This inverse correlation therefore provides further support for the alternative hypothesis that the presence of an elevated level of pollen has an inhibiting effect on flu-like incidence, and starts to immediately influence the direction and course of the epidemic life cycle. Also, during the COVID-19 dominated period of the last 9 weeks, it appears that flu-like incidence behaves according to the expected pollenflu seasonality. This strengthens the idea that COVID-19 might itself be seasonal, like all other flu-like pandemics since the end of the 19 th century. Also when studying other data from RIVM.nl about COVID-19 hospitalizations, we cannot conclude that COVID-19 breaks through the seasonal barrier.

For example, new COVID-19 hospitalizations decreased from a peak of 611 on March 27 to just 33 on May 3, the last day of week 18.

Using the three-week moving average (ΔILI 3WMA ) of changes in flu-like incidence, the correlation coefficients become stronger and are again highly significant for total pollen concentration (r(223) = -J o u r n a l P r e -p r o o f Journal Pre-proof 0.41, p < 0.00001). The bootstrapped correlation coefficient calculation gives a comparable outcome (r(223) = -0.38, p < 0.0001). As a second sensitivity analysis, we used the reduced dataset (minus the weeks of low pollen activity) and again found similar correlations (r(191) = -0.44, p < 0.0001; bootstrapped r(188) = -0.44, p < 0.0001, CI 95% -0.46 to -0.25)). Finally, as a third sensitivity analysis, we analyze each time lag included in the ΔILI 3WMA calculation separately. Per individual time lag there are as well highly significant inverse correlations: as given before r(226) = -0.26, p = 0.000063 in case of no time lag (ΔILI t ); r(225) = -0.22, p = 0.000713 in case of a time lag of one week (ΔILI t+1 ); r(225) = -0.23, p = 0.000552 in case of a time lag of two weeks (ΔILI t+2 ); and the bootstrapped correlations for these are similar. We can thus also reject the null-hypothesis (H2 0 ) that there is no inverse relationship between pollen and changes in flu-like incidence including incubation time (ΔILI 3WMA or ΔILI t+1 or ΔILI t+2 ). These correlations (see also Figure 4 ) are a further indication of temporality and does not contradict the idea that COVID-19 is subject to pollen induced fluseasonality. The fact that the correlation with ΔILI 3WMA is stronger than those for each of the included time lags might be an indication of the noise reduction effect of this moving average, and makes thus the compound effect of the three covered time lags more visible.

Linear regression analysis shows that pollen has a highly significant inhibitory effect on flu-like incidence change (ΔILI 3WMA ) of F(1, 222) = 37.1, p < 0.001 (see Table 2 , line 1), as a further basis for using total pollen concentration as a predictor. A Log10 transformation of pollen to compensate for visual non-linearity leads to a similar outcome: F(1, 219) = 43.87, p < 0.001 (see Table 2 , line 4). At least visually, it is a good fit (see Figure 4 ).

Of the meteorological variables, only solar radiation has a highly significant inverse correlation with changes in flu-like incidence (ΔILI 3WMA ): (r(224) = -0.25, p = 0.000156).

Thus, of the meteorological variables, when it comes to solar radiation and relative humidity the nullhypothesis (H4 0 ) can also be rejected, as they seem to effect the flu-like epidemic lifecycle. Of these two, only solar radiation is a flu-like inhibitor in line with its positive effect on pollen concentration, its association with immune-activation and the effect that UV has on viruses.

A univariate linear regression also shows the highly significant negative correlation for solar radiation on flu-like incidence change (ΔILI 3WMA ) (F(1, 222) = 14.43, p < 0.001 (see Table 2 , line 2). As the correlation is weak (Multiple R-squared = 0.06), we have interpreted solar radiation as a co-inhibitor in relation to pollen; as a stand-alone independent variable its effect is too weak to explain flu-like seasonality.

Taking into account all these findings, we developed a discrete, compound model in which we included the changes in flu-like incidence (ΔILI 3WMA ), a threshold value for solar radiation (k r ), and both pollen threshold values for allergenic (k ap ) and total pollen (k p ). We found that the compound model has the highest inverse correlation (r(222) = -0.48, p < 0.001) for the following threshold values: k r : 510 J/cm 2 , k ap : 120 allergenic pollen grains/m 3 , and k p : 610 total pollen grains/m 3 . The bootstrapped correlation coefficient calculation gives a comparable outcome (r(222) = -0.47, p < 0.0001). In line with the previous outcomes, the inclusion of relative humidity, low-level allergenic pollen or temperature did not improve the correlation strength of this model. Furthermore, given that they showed no significant interaction effects with pollen, it was not necessary to take such interactions into consideration in the model.

In each of the observed years, the now (re)defined pollen thresholds are passed in week 10 (± 5 weeks), depending on meteorological conditions controlling the pollen calendar and coinciding with reaching flu-like peaks, and again in week 33 (± 2 weeks), marking the start of the new flu-like season.

There is a highly significant inverse relationship between our compound threshold-based predictor value with flu-like incidence change (ΔILI 3WMA ) of F(1, 222) = 65.59, p < 0.001 and a Multiple R-J o u r n a l P r e -p r o o f squared correlation of 0.2281 (see Table 2 , line 3). This confirms the usefulness of a discrete, pollen and solar radiation threshold-based model as a predictor of switches in flu-like seasonality, whereby the effect of pollen is stronger than that of solar radiation. As a consequence, we can reject the nullhypothesis (H3 0 ) that this compound pollen/solar radiation value has no predictive significance for flulike seasonality.

First of all we will discuss the possible implications of the results for our theoretic model and alternative explanations. Next, we will discuss our methods.

We found highly significant inverse relationships between pollen and solar radiation and (changes in) flu-like incidence: a higher pollen concentration or an increase in solar radiation in the Netherlands is related to a decline in flu-like incidence. This inverse correlation with pollen becomes stronger when the 2019/2020 period is included, which has been increasingly dominated by COVID-19 during the last 9 weeks. Given that more time will be needed to draw conclusions about whether the spread of COVID-19 is seasonal or not, from the data in this study it can only be observed that COVID-19 is not breaking with the flu-like seasonality pattern. Alternatively, social distancing may have contributed to flattening both the flu-like epidemic and COVID-19 pandemic curves at the tail-end of the 2019/2020 flu-like season. The Dutch government imposed hygiene measures from March 9, 2020 onward and a mild form of a lockdown, that included social distancing, from March 11. Such behavioral policies

The highly significant inverse correlation between hay fever and flu-like incidence confirms that allergic rhinitis makes it more difficult for flu-like viruses to propagate. Solar radiation, the only meteorological variable that has a co-inhibitive effect on changes in flu-like incidence, has a stimulating effect on aerosol pollen formation and is responsible for melatonininduced immuno-activation. Relative humidity reduces pollen aerosol formation, and correlates positively with flu-like incidence. We did not specifically look at precipitation, but it might make sense to explicitly consider this independent variable, given that it reduces pollen dissemination.

In our study we showed that temperature, aside from the fact that it influences pollen, has no predictive value for changes in flu-like incidence. Therefore, its inverse correlation with flu-like incidence might be interpreted in a number of ways: a) as spurious: the common causal factor is solar radiation, or b) as a stressor that has immediate effects on the functioning of the immune system of already infected persons. When discussing the influence of meteorological variables, we assume that the associated behavioral aspects are covered. These are sometimes summarized as seasonal behavior, but this independent variable might have a cultural dimension that needs to be better understood.

We showed that a compound value, based on threshold values for pollen and solar radiation, results in a stronger correlation with the flu-like lifecycle than the individual inhibitors. This model could form an empirical basis for understanding flu-like seasonality, its Ro and reliably predicting the start and end of each flu-like cycle. Given that behavior, in the form of hygiene and social distancing, is also widely seen as an inhibitor, it might be worthwhile to also include this factor in our compound value. This will probably lead to an even stronger predictor for the evolution of the reproduction number Ro of flu-like epidemics, although this might be beyond explaining the seasonality effect itself. For as long as the level of herd immunity (Fine et al., 2011) for COVID-19 is still below required thresholds for ending pandemics (Plans-Rubio, 2012) , it might make sense to also include indications of herd immunity levels in the theoretic model.

Finally, despite air pollution not been seen as an inhibitor of flu-like incidence (Coccia, 2020), it still might interact with pollen. A more complete theoretic model, controlling for the (interactions with) air pollution, could give more insight in how to interpret the findings of this or similar studies.

In general, statistical research cannot prove causal relationships in uncontrolled environments, even if datasets seem to behave as if there is causality. Such statistics, however, can provide indications and identify reliable predictors, help filter out bad ideas, and be the inspiration for testable hypotheses that can be verified in laboratory and other fully controlled experiments. With a predictor we mean that a reliable temporal relationship between two variables is identified, without yet having validated causality, i.e., a bellwether factor.

Although the datasets seem to be sufficiently representative, there appears to be room for improvement. For example, including the data of more weather stations might help to improve the approximation of the weather conditions the Dutch population experiences on average, and help to distinguish patterns per province. Furthermore, it might be useful to include wind speeds, given that these constitute a vector for the dispersal of pollen in the Netherlands, which has a maritime and temperate climate. Additionally, the effects of climate change on pollen maturation (Frei & Gassner, 2008 ) might also be an important factor. Another example of improving the representativeness would be by including more pollen types in the particle counts than are currently covered by the current methodology of the European Allergy Network. Further, reclassification or recalibration of pollen types on a rational scale in terms of allergenicity, let's say 0-100%, would be very useful. For example, if pollen types are identified with a reliable score of 0%, these could be used to differentiate 

Impact of Pollen

Eleven faces of coronavirus disease 2019

Herd Immunity": A Rough Guide

Seasonality in risk of pandemic influenza emergence

Towards a data-driven characterization of behavioral changes induced by the seasonal flu

Ripe pollen carbohydrate changes in Trachycarpus fortunei: the effect of relative humidity

Climate change and its impact on birch pollen quantities and the start of the pollen season an example from Switzerland for the period 1969-2006

Pollen exposure weakens innate defense against respiratory viruses

Nivel zorgregistraties eerste lijn -Surveillance -wekelijks bulletin over symptomen en aandoeningen op basis van gegevens van huisartsen

Why Lungs Keep Time: Circadian Rhythms and Lung Immunity

The vaccination coverage required to establish herd immunity against influenza viruses

Dynamical prediction of flu seasonality driven by ambient temperature: influenza vs. common cold

Current Understanding of the Pathophysiology of Allergic Rhinitis

COVID-19 transmission in Mainland China is associated with temperature and humidity: A time-series analysis

Humidity and Latitude Analysis to Predict Potential Spread and Seasonality for COVID

Absolute Humidity and Pandemic Versus Epidemic Influenza

The Role of Temperature and Humidity on Seasonal Influenza in Tropical Areas: Guatemala, El Salvador and Panama

Global influenza seasonality: reconciling patterns across temperate and tropical regions

Potential utility of melatonin in deadly infectious diseases related to the overreaction of innate immune response and destructive inflammation: focus on COVID-19

& Asfiang P. Correlation between weather and Covid-19 pandemic in Jakarta

Atopic diseases and related risk factors among Dutch adolescents

Receptor Recognition by the Novel Coronavirus from Wuhan: an Analysis Based on Decade-Long Structural Studies of SARS Coronavirus

Association between ambient temperature and COVID-19 infection in 122 cities from China

No Association of COVID-19 transmission with temperature or UV radiation in Chinese cities

Clinical characteristics of 140 patients infected with SARS-CoV-2 in Wuhan