Skip to main content

Temperature and particulate matter as environmental factors associated with seasonality of influenza incidence – an approach using Earth observation-based modeling in a health insurance cohort study from Baden-Württemberg (Germany)

Abstract

Background

Influenza seasonality has been frequently studied, but its mechanisms are not clear. Urban in-situ studies have linked influenza to meteorological or pollutant stressors. Few studies have investigated rural and less polluted areas in temperate climate zones.

Objectives

We examined influences of medium-term residential exposure to fine particulate matter (PM2.5), NO2, SO2, air temperature and precipitation on influenza incidence.

Methods

To obtain complete spatial coverage of Baden-Württemberg, we modeled environmental exposure from data of the Copernicus Atmosphere Monitoring Service and of the Copernicus Climate Change Service. We computed spatiotemporal aggregates to reflect quarterly mean values at post-code level. Moreover, we prepared health insurance data to yield influenza incidence between January 2010 and December 2018. We used generalized additive models, with Gaussian Markov random field smoothers for spatial input, whilst using or not using quarter as temporal input.

Results

In the 3.85 million cohort, 513,404 influenza cases occurred over the 9-year period, with 53.6% occurring in quarter 1 (January to March), and 10.2%, 9.4% and 26.8% in quarters 2, 3 and 4, respectively. Statistical modeling yielded highly significant effects of air temperature, precipitation, PM2.5 and NO2. Computation of stressor-specific gains revealed up to 3499 infections per 100,000 AOK clients per year that are attributable to lowering ambient mean air temperature from 18.71 °C to 2.01 °C. Stressor specific gains were also substantial for fine particulate matter, yielding up to 502 attributable infections per 100,000 clients per year for an increase from 7.49 μg/m3 to 15.98 μg/m3.

Conclusions

Whilst strong statistical association of temperature with other stressors makes it difficult to distinguish between direct and mediated temperature effects, results confirm genuine effects by fine particulate matter on influenza infections for both rural and urban areas in a temperate climate. Future studies should attempt to further establish the mediating mechanisms to inform public health policies.

Peer Review reports

Background

Influenza, which is caused by the influenza-A or the influenza-B virus, is transmitted through the respiratory organs via physical contact with an infectious person or with contaminated objects, or via airborne droplets or droplet nuclei, with the latter having a radius of < 2.5 µm. Up to 1 billion infections occur per year world-wide, with up to 5 million severe cases and 50,000 deaths attributable to influenza [1]. However, substantial seasonal variation exists, with infection rates culminating in winter, at least in temperate climates in the southern and northern hemisphere. Thus, influenza bears many similarities with SARS-CoV2, which has been causing the Covid-19 pandemic since December 2019.

The astounding seasonal variation appears to be related to several factors that can be grouped into four categories: pathogen abundance; environmental factors; host behavior; and host susceptibility [2]. The environmental factors, temperature, humidity, and vapor pressure have been implicated [3]. Notably, seasonality increases with latitude [4]. Thus, incidence rates depict a single peak in the temperate climate zones, with peaks in January to March in the northern hemisphere and in July to September in the southern hemisphere. Tropical climates may depict one, two (example: Hong Kong [5]), or no peak at all. The observation suggests that seasonal variation is not due to one factor alone, but rather to a combination of factors.

In addition to the meteorological factors, air pollutants such as ozone (O3) and sulfur dioxide (SO2) may modulate influenza transmission [6]. Associations of O3 and coarse particulate matter (PM10) with influenza-hospitalization, but not with influenza-mortality have been reported for Hong Kong [7] The finding has been confirmed by a pediatric cohort study in Brisbane (Australia), which also reports significant interaction between pollutant and temperature effects [8]. For Nanjing, a more temperate climate, associations were also found for nitrogen dioxide (NO2). A recent study from Wuhan attributes an additional role to SO2 [9]. Notably, that study also demonstrates the strongest linkage with a one- or two-day lag time only, suggesting a rather acute mediation of the effect. Another recent study of mortality in Milan during the winter 2016/2017 suggests that air pollution, low temperature and influenza infection jointly mediate excess mortality [10]. Moreover, re-assessment of historical data suggests that smog powerfully contributed to the death toll of London’s 1952 influenza epidemic [11]. Another study posits particulate matter (PM) arising from coal combustions severely aggravated 1918’s world-wide influenza pandemic [12]. Overall several studies have indeed examined the relationship between temperature, humidity and influenza [3, 13, 14], but only very few that propose effects of fine particulate matter (PM2.5) [15, 16]. However, all these studies are carried out under moderate to high air pollution conditions [7, 17] or for selected cities [15, 17, 18].

While environmental factors appear to strongly interact with each other in their impact on influenza incidence and severity, the fact that air pollutants themselves exhibit seasonality in temperate climates complicates the interpretation [19, 20]. Whether the association reflects non-causal co-variation or true causality cannot be discerned. A detailed understanding of the causative inter-relationships is particularly lacking for temperate climates, where the existing literature is based on in-situ ground measurements from larger cities. Furthermore, there is a general lack of studies including data that range from urban and highly-polluted to rural and less-polluted areas, thereby taking advantage of spatial variation in environmental factors in addition to their temporal variation. Spatial variation may be less error-prone, as it minimizes the effects of non-causative covariation on the observed outcome. In-situ ground measurements would be cumbersome to perform over wider geographical territories, but Earth observation in combination with twenty-first century computing power does allow for modeling of environmental exposure over wide-spread areas.

The German Aerospace Center as a research center (DLR) and the AOK Baden-Württemberg as the 5th largest statutory health insurance formed a unique partnership in 2020 to understand the complex system of health, environment and social structures. Man-made climate change is changing decisive factors that affect the health of the population in a variety of ways. For the AOK Baden-Württemberg, it is of crucial importance to understand the effects of climate change on health in detail to make significant progress in clarifying the dose–response relationship. Particular attention is paid to the protection of vulnerable subpopulations.

We, therefore, applied a novel approach combining state-of-the-art Earth observation-based numerical modeling with health insurance data to assess effects of residential exposure to pollutants and meteorological factors on the seasonality of influenza infections in, both, rural and urban environments of a temperate climate zone. We hypothesized that concentrations of PM, NO2, SO2 and ozone would be positively associated with influenza incidence, and that temperature would be negatively associated.

Methods

Study population

The study was designed as an observational cohort study, making use of all residents of Baden-Württemberg insured with Allgemeine Ortskrankenkasse (AOK) in Baden-Württemberg. In Germany, each federal state has its own AOK, which is the public corporation commissioned to provide health insurance to people. Until 1996, health insurance corporations had been linked to professions in Germany, and AOK therefore has been and continues to be the main health insurer for the working class.

Baden-Württemberg is Germany’s third most populated state, with 11 million inhabitants of which 4.5 million are currently health-insured with AOK Baden-Württemberg. The state’s capital is Stuttgart, with approximately 5 million residents in its metropolitan region. Baden-Württemberg’s total area amounts approximately 36,000 km2 comprising highly rural and urban environments.

We included data from all AOK clients residing in Baden-Württemberg collected for the period between January 2010 and December 2018.

The study is in accordance with the declaration of Helsinki and was approved by the Ethical Committee of the Medical Council (Landesärztekammer Nordrhein) within the VARIAQ study (lfd Nr 2,020,092). Data management plans were approved by data protection officers on both sides (DLR and AOK) before data exchange commenced.

Identification of cases

We gathered and merged at person-level health data from 5 AOK databases, as separate data bases exist for in-patient hospital cases, for out-patient hospital cases, for sick-leave, and two separate data-bases that cover data from other out-patient cases. Duplication of individual records was prevented through AOK-internal identifiers. The data sources primarily serve to manage remuneration of health care providers. As remuneration is occurring on a quarterly basis, the resulting data base contains quarterly information on diagnoses.

We classified all cases with ICD-10 codes J09 (influenza due to certain identified influenza viruses), J10 (influenza due to other identified influenza virus) or J11 (influenza due to unidentified virus with other respiratory manifestations) in a given quarter as prevalent cases. They were furthermore classified as new cases when the ICD-10 codes J09, J10 or J11 were present in a given quarter but not in the preceding quarter. From these data, we aggregated the number of new cases and the number of existing cases per quarter, per 5-digit postcode, and per gender. Sex and age information was abandoned to preclude exposure of personally identifiable information, and to be in keeping with data protection regulations. This approach was necessary in particular for remote areas with low population densities.

Environmental data

As stated above, our goal was to model environmental exposure for the entire region of Baden- Württemberg for all quarters between 2010 and 2018. Therefore, we retrieved air pollution data, i.e. surface level concentrations of PM10, PM2.5, NO2, SO2 and O3, from the Copernicus Atmosphere Monitoring Service (CAMS) European air quality reanalyses, which is a dataset resulting from an ensemble of seven (nine, after the upgrade in October 2019) chemical transport models composed of daily forecast and analysis data on the main pollutants’ concentration. The dataset has a native temporal resolution of one hour and a horizontal resolution of 0.1° × 0.1° [21, 22]. Forecast and analysis data are annually validated and adjusted by assimilation of data from the Copernicus in-situ component like measurements from the European Environmental Agency’s (EEA) station network. The re-analysis is publicly available on the Copernicus Atmosphere Data Store (ADS, https://ads.atmosphere.copernicus.eu).

We retrieved meteorological data from the Climate Data Store (CDS) of ECMWF, a free and open access platform gathering quality assured climate data derived from Earth observation, global and regional climate re-analyses of past observations, seasonal forecasts and climate projections. Moreover, we obtained data of downward ultraviolet (UV) radiation at the surface with hourly temporal resolution and with horizontal resolution of 0.25° X 0.25° from the ERA5 reanalysis dataset. This parameter is defined as the amount of UV radiation with a wavelength of 0.2–0.44 µm reaching the surface [23]. The ERA5 dataset is produced within the Copernicus Climate Change Service (C3S) that provides records of global atmosphere, land surface and ocean from 1959 onwards [24]. We retrieved air temperature data, dewpoint temperature at 2 m above the surface, from the ERA-5-Land dataset, which is also a reanalysis dataset available on the ECMWF CDS with enhanced resolution compared to ERA5. The dataset is available for the time period from 1950 until 2–3 months before the present time. It consists of a replay of the land component of ERA5 climate reanalysis with an applied sea mask and with the assimilation of observational data [25]. This dataset offers a native temporal resolution of one hour and a horizontal resolution of 0.1°X0.1°. A summary of the data features is provided in Table 1.

Table 1 Summary of information on sources used to gather the environmental data used for this study

From the 2 m temperature data we calculated daily aggregates of minimum, mean, and maximum. Similarly, we also converted dewpoint temperature data to °C and derived vapor pressure using an empirical formula [26], see Eq. 1. Afterwards, we produced daily minimum, mean, and maximum surface vapor pressure aggregates.

$$e=6.112*\mathrm{exp}( \frac{17.67*{T}_{d}}{{T}_{d}-243.5})$$
(1)

where: e = Vapor pressure in hPa, Td: dew point temperature in °C

We aggregated UV radiation data on a daily basis, calculating the sum of the radiation reaching the surface. We also converted the unit from J/m2 to W/ m2 dividing by the integration time in seconds.

For all the layers of environmental variables, we increased the horizontal resolution via a grid-granularization of the grid while using the nearest neighbor interpolation method. The new grid resolution is 0.067° X 0.067°.

To obtain geographical data, we derived shapefiles for all 5-digit postcodes in Baden-Württemberg from the ESRI Deutschland databank [27]. Geographical and demographical information in the dataset were extracted from ©OpenStreetMap contributors and from data of the 2011 census of Germany’s federal authority of statistics (https://www.zensus2011.de/DE/Home/home_node.html). The number of inhabitants in the shapefile was calculated for each polygon based on population density per 1km2 according to the DESTATIS’s Zensus2011 dataset. We used the shapefiles to mask environmental layers and to perform a spatial aggregation for each polygon. Thus, we obtained daily aggregates of all the environmental variables considered for each 5-digit postcode area. Finally, we performed another aggregation step to yield quarterly (i.e. three-month-averaged) data, calculating the mean values within each postcode-quarter aggregate window.

Data scrutiny

At the stage of data exchange, records with unknown age were excluded, leading to a loss of 0.56% of observations. Such spurious records in insurance data bases typically emerge from accidents and emergency for which the identity of the patient is not fully known. Next, a check for personally identifiable information was performed, which meant that all information with less than 3 persons (including cases and non-cases) per aggregate window (postcode, quarter) had to be discarded. However, the loss amounted to 0.0032% of individual-level data only.

Statistical methods

Before statistical model fitting, we explored pairwise collinearities among the environmental variables. According to the correlation matrix given in Table 2, we observed substantial collinearity in particular for the variables O3, PM10, UV radiation and vapor pressure (right and lower quadrants of the Table 2). Accordingly, to reduce issues arising from collinearity, we restricted statistical modeling to PM2.5, NO2, Temperature and Precipitation. Figure 1 presents joint distribution plots for environmental variables presenting collinearity.

Table 2 Matrix with Pearson’s correlation coefficients for environmental variables
Fig. 1
figure 1

Relationship of UV radiation, NO2 and PM2.5 with temperature. Joint distribution plots for UV, vapor pressure, NO2 and PM2.5 versus temperature. The distributions shown in each plot refer to mean daily aggregates for each ZIP-code area in Baden-Württemberg between 2010 and 2018. The dataset used to produce the plots are those reported in Table 1

With these four selected environmental independent variables, we fitted generalized additive models [28]. Generalized additive models allow for a flexible specification of outcome dependence on the variables as the predictor comprises a sum of (possibly nonlinear) smooth functions. The number of new influenza cases per post code represents count data (= non-negative integers) and hence the modeling approach was based on a negative binomial model [29] using the number of new cases as outcome variable. Regarding the independent variables, we used smooth P-spline functions for environmental variables, a Gaussian Markov random field smoother over postcodes and an offset for the number of AOK clients per postcode. Of note, entering postcode as an independent variable adjusts for site-specific traits such as population density, urbanity and other factors. In addition, we included two variables (‘quarter’ and ‘season’) to model temporal effects. Firstly, we used quarter as a means of accommodating seasonal variation that is not necessarily independent of environmental stressors under study. However, as much as stressors vary with ‘quarter’, this approach may not fully capture the true stressor effects. Therefore, we built a full model that does include quarter (called full model henceforth), and another model that does not include quarter (henceforth called restricted model). Accordingly, the restricted model attempts to explain seasonal variation in influenza incidence by seasonal variation in environmental stressors only. As a second temporal variable, we collapsed quarters 3 and 4 with the quarters 1 and 2 of the subsequent year as ‘season’, with the intent to model antigen shift and antigen drifts of the influenza virus. The 2011/12 seasonal influenza epidemics has started outstandingly late and was very mild [30]. Therefore, we ran models both with all seasons including season 2011/12, as well as with the recent seasons (2012/13 and later) excluding that season.

We performed statistical analyses and modeling with R in its version 4.0.4 (www.r-project.org). To assess periodicity, we subjected time series data to spectral analysis, using a maximum entropy method (MEM) in Python (https://github.com/martini-alessandro/Maximum-Entropy-Spectrum). The MEM is very well suited for short time series [31], however, an appropriate order of the underlying autoregressive process must be chosen in advance [32]. In our case, we decided for an order of 10 for a time series of 42 data points. This is in accordance with S Wust and M Bittner [33].

The function ‘bam’ from the r-package ‘mgcv’ was used to fit generalized additive models. Using the additive model fits, we estimated the number of influenza cases per 100,000 AOK clients per year, using the ‘predict’ function for all environmental stressors, with setting the other stressors to their medians, setting quarter = Q1, season to S14/15, and post code = D-68159. From these estimates, we computed the number of cases at the 5% and 95-percentiles of each respective stressor, keeping the values of the other independent variables constant, and we obtained the stressor-specific gain as the difference between predictions for these percentiles.

Given that the environmental stressor data are the result of a joint activity by many researchers, and that ICD-coded insurance data are highly standardized and controlled, there seems little potential for bias. To further reduce any bias, all data processing and statistical scripts, as well as interim results have been checked by at least two experienced programmers from the authors’ team.

Results

The cohort of 3.85 million people resided within a total of 1194 postcodes that stretched out over at total of 35,712 km2. Population density was highly variable between postcodes (Fig. 2A), with a median of 249.5 inhabitants/km2, ranging from 2.4/km2 to 16,797/km2. Over the period of 9 years (36 quarters), a total of 513,404 influenza cases were observed, of which 53.6% occurred in quarter 1 (January to March), and 10.2%, 9.4% and 26.8% in quarters 2, 3 and 4, respectively. Overall, the influenza incidence amounted to 1482 cases per 100,000 AOK clients per year. A time series of influenza incidence is presented in Fig. 2B, and Fig. 2C depicts a spectrogram of these data. These plots reveal pronounced 1-year periodicity, in which incidences in quarter 1 consistently outnumber incidences in quarters 4, 2 and 3.

Fig. 2
figure 2

Geographical distribution of the cohort and influenza seasonality. A population density for each ZIP-code area in Baden- Württemberg. Population density data were derived from the DESTATIS’s Zensus2011 dataset. B boxplot reporting incidence per postcode for each quarter between of the years between 2010 and 2018. The y-axis has been cut at a value of 70, to allow better representation of boxes (C) Spectrogram of the data in (B). Note the pronounced peak at period = 1 year, and also the first harmonic at period = 2 years

Such quarterly variation with 1-year periodicity is also visible for the time series of Temperature, PM2.5 and NO2, but not for precipitation, as shown in the left panels of Fig. 3 and in Table3. The right panels of that figure depict spatial variation in the same stressors across the different postcodes in Baden-Württemberg.

Fig. 3
figure 3

Seasonal and geographical variation in environmental stressors. On the left side, box plots for stressors per postcode and quarter reflect temporal variation from 2010 to 2018. For total precipitation, the y-axis has been cut off at 200 mm/day, so that box sizes have reasonable spread. To reflect spatial variation, the diagrams on the right side display the averaged stressor levels over the observation period 2010 to 2018, plotted in geographical postcode coordinates. The dataset used to produce the plots are those reported in the Table 1

Table 3 Means and percentiles of environmental data

Statistical modeling yielded significant effects of temperature, precipitation, PM2.5, and NO2 for all four different models (all P < 10–16). When exploring the stressor-specific gains, effects were largest for temperature, ranging from 2639 (full model, all seasons) to 3499 (restricted model, recent seasons) cases per 100,000 AOK clients per year that can be attributed to temperature (Table 4). Plotting the estimated temperature effects revealed that the negative association between temperature and influenza incidence was most pronounced below 5 °C, and that the effect tapered off towards the higher temperatures (Fig. 4). The second largest effect was exerted by PM2.5, with a stressor-specific gain between 297 (restricted model, all seasons) and 502 (full model, recent seasons) cases per 100,000 AOK clients per year (Table 4). As can be seen from Fig. 4, the positive association was relatively steadily over the entire range of PM2.5 concentrations. Effects for precipitation and NO2 were in the order of 200 cases per 100,000 AOK clients per year only. Counterintuitively, however, NO2, was negatively related to influenza incidence (Fig. 4). It is also remarkable that the association for precipitation was only observed below 3mm/day.

Table 4 Attributable cases
Fig. 4
figure 4

Estimated cases in response to variation of environmental stressors. Estimations (and 95% their confidence intervals) were obtained with the R-function predict for the 5- to 95-percentile intervals of Temperature, Precipitation, PM2.5 and NO2 from the restricted model for 2013–2018 data. Here we display curves for the restricted model (excluding quarter as independent variable) with recent season set (from season 2012/13 onwards). Note that the other 3 statistical models yielded trend curves that are very similar to those displayed here. Also, including interaction terms e.g. for temperature and PM2.5 had no obvious bearing on the trend curves

Discussion

The important finding of our study combining Earth observation-based modeling with health insurance data is that seasonal variation in the incidence of influenza is attributable to temperature effects and to particulate matter. In this respect, our results are in line with previous in-situ urban studies [6, 7, 9, 34, 35] and extend the evidence to a mixed urban and rural cohort and to out-patients. However, associations with precipitation were at best moderate, and an incidence-promoting effect by NO2 cannot be supported on basis of our data. The associations found were robust across the four different statistical models used, which is particularly noteworthy with regards to the consistency between full and restricted models. The facts that the full models allowed quarter as an explanatory variable without direct involvement of the stressors, and that estimations did not deviate much when quarter was excluded as independent variable, suggest that the observed stressor effects reflect more than a mere temporal coincidence.

Even though our analyses found the strongest effect for environmental temperatures, the exact mechanisms by which temperature exerts the effects remain to be elucidated. Thus, as can be seen from Table 2, Pearson’s correlation coefficient between temperature and vapor pressure was 0.98, meaning that very similar results must be expected if temperature is replaced by vapor pressure. Indeed, previous work had suggested that vapor pressure rather than temperature per se is the effective agent that disrupts the influenza virus [14, 36]. It may also speak in favor of this idea that the observed relationship between influenza incidence and environmental temperature was strongly curvilinear (Fig. 4), which would be expected from Eq. 1 in this paper and Fig. 3 in J Shaman and M Kohn [14] if the temperature effect is mediated by vapor pressure. On the other hand, host susceptibility is known to be increased at low temperatures, e.g. via hampered interferon-β expression in respiratory cells and thereby blunted immune response [37]. And, as a third of many other alternative explanations, the temperature effect could also be partly mediated by UV radiation [38], with the latter being closely correlated to temperature (Table 2), and UV being well-known to inactivate viruses.

The results for precipitation suggest a comparatively moderate positive association that is levelling off above 3 mm/day (Figure), which is somewhat unexpected. Of note, as much as precipitation is related to relative humidity and vapor pressure, this observation is opposed to the purported anti-viral effects for the latter two [36]. The expected association has indeed been found by a recent study from Gwangju in South Korea [34]. Of note, precipitation in Gwangju is accentuated in the summer months, but relatively independent of quarters in Baden-Württemberg (Fig. 3, Table). Moreover, summers are approximately 10 °C hotter in Gwangju than in Baden-Württemberg. It therefore seems possible that precipitation may elicit anti-viral effects in humid-hot climates, but promote virus infections e.g. via compromising host defense mechanism in cold-humid climates.

The second-largest association with influenza incidence in our data was observed for PM2.5. Of note, collinearity of PM2.5 with other stressors in Table 2 was substantially smaller than for temperature (with the exception of PM10). It therefore seems that the observed association points towards a genuine PM2.5 effect. Literature identifies two different, but not exclusive, types of explanations, namely by stabilization of airborne virus particles e.g. against UV light, and / or by compromising the host’s defense mechanisms. Thus, it is well established that PM2.5 elicits inflammatory responses [39], which are apt to chronically compromise airways and foster infections as typically seen in asthma and chronic obstructive pulmonary disease.

The pertinent question is how out-door PM2.5 levels, as used here, could affect people’s health when people spend most of their time indoors. Thus, an analysis of 2010 found that only 1.04 h per day are outdoor during weekdays, and only 1.64 h per day during weekends [40]. Indoor PM results from infiltration from outdoors, and from primary and secondary indoor sources [41, 42]. Primary sources include all types of combustion (e.g. for heating and cooking, but also including smoking), but also sources such as laser printing devices, handicrafts and organic aerosols from human and animal sources. Secondary indoor sources include particles infiltrated from outside that chemically react with particles from indoor sources. B Wang, Y Liu, Z Li and Z Li [43] have demonstrated that such indoor sources, in particular coal combustion have a strong bearing on influenza infections. In addition, it is well established that indoor PM significantly depends on outdoor PM [41]. However, virtually all buildings in Baden-Württemberg afford central heating, and heating affects indoor PM levels in two ways. Firstly, by directly fostering the convection-driven resuspension of PM, and second by reducing relative humidity and thereby hampering PM suspension [44]. Therefore, although speculative at this moment, there is a possibility that outdoor temperature impacts on influenza via heating-related seasonal variation in indoor PM levels.

Finally, the present study does not support the view of NO2 being a major driver for influenza, an idea that was put forward with regards to the current SARS-CoV2 pandemic [45, 46]. Instead, we observed a very moderate negative, rather than a positive, association between NO2 levels and incidence (Fig. 4), which may more likely be due to statistical collinearity with temperature or PM2.5 than the hallmark of NO2-related salutogenesis. Naturally, differences could exist with regards to virus-environment interactions between influenza and SARS-CoV2. The mechanism that have to date been proposed for mediating environmental stressor effects are very similar for these two viruses. Thus, observations made for influenza infections are certainly of interest also for a better understanding of the Covid-19 pandemic. Many studies world-wide have reported modulation by environmental stressors, albeit with mixed results [47,48,49,50]. Thus, even though environmental factors certainly did not fully determine infection rates in the early days of the pandemic [51], it does seem reasonable to consider environmental factors for disease control. This applies foremost to factors that can be modified with justifiable effort.

An important limitation of the present study is the limited spatial resolution (postcode area), and also the lack of information regarding the sojourn of individuals. Whilst it has become technically possible to monitor environmental exposure on a personal basis [52], the cost involved and the need of consenting participants would preclude sample sizes as in the present study. Therefore, given the lack of personal mobility data, the coarse graining of spatial information is not necessarily a disadvantage, in particular because spatial variation of stressor levels was fairly high in our data set (Fig. 3). The environmental data used for this research work have a native horizontal resolution of 0.1° × 0.1°, that, at the latitudes of interest corresponds to an area of approximately 55 km2. The main advantage of using these datasets consists in having a constant and reliable data availability for the spatial and temporal domain considered. Furthermore, the datasets used are all defined as reanalysis. This implies that a consistent quality check by comparison with observational data and their assimilation in the datasets has already been carried out. However, the cohort design of this study required a spatial aggregation of the environmental to the ZIP-code areas in Baden-Württemberg. Many of these areas are smaller with respect to the resolution of the dataset.

Another weakness of the present study is the unknown proportion of diagnoses that are based on clinical inspection only, with no confirmation by laboratory testing. However, obtaining confirmatory testing would have been impossible in the current study setting, and it would be extremely resource-demanding in any study sample of that magnitude with 34.7 million observed person-years and more than half a million diagnosed cases. Moreover, it has been demonstrated that clinical surveillance for influenza is synchronous with laboratory surveillance [5], suggesting that clinical diagnosis is a valid proxy of influenza infections.

Finally, the question of generalizability of the results arises. Whilst AOK used to be the designated health insurance for the working class until 1996, nowadays there is a free choice of health insurance in Germany. We regard the predominant inclusion of a working-class cohort as a strength of our approach, because it reduces influences from socio-economic status, and because populations of lower socio-economic status are typically more affected by environmental health issues. Therefore, a slight bias exists in the present cohort with relatively higher fractions of working class in the older as opposed to the younger age groups. However, the morbidity-related risk structure compensation, which is a monetary mechanism to enforce equity between German insurance companies, is very similar between AOK-BW and the average of the other German health insurance companies, which gives us confidence that the cohort underlying this study represents Germany’s population to a large degree.

Conclusions

Our study is in line with and extends previous studies with in-situ ground measurements reporting strong associations between the environmental stressors temperature and particulate matter upon influenza incidence. In particular, our study suggests that such associations exist also for a region with temperate climate encompassing urban as well as rural areas. However, the present results put previous reports on associations with precipitation and NO2 into perspective. Whilst the true effects of temperature observed here are, at least partly, exerted via collinear effects such as ultraviolet radiation, ozone, and other routes. In contrast, effects observed for particulate matter were statistically less collinear with other stressors. We, therefore, suggest that reduction of particulate matter could offer leverage for public health policies in relation to influenza and other viral infections. Future studies should aim at gaining information on indoor levels of particulate matter, and at gaining more individualized information on exposure levels. In addition, it should be analyzed which costs can be directly linked to the environmental stressors. The DLR and AOK BW want to continue their partnership for further analyses.

Finally, this investigation also demonstrates that Earth observation-based modeling in combination with health insurance data constitutes a powerful concept for public health studies. In the medium and long term, there is the vision for a location-independent and individualized risk prediction that will decisively advance healthcare and prevention and draw the population of Baden-Württemberg to the effects of climate change adequately prepared and protected.

Availability of data and materials

The data that support the findings of this study are available on request from the corresponding author JR. The data are not publicly available due to the high standards of data protection for health data.

Abbreviations

ADS:

Athmosphere data store (of Copernicus)

AOK:

Allgemeine Ortskrankenkasse (public health insurance company)

CAMS:

Copernicus atmosphere monitoring service

CDS:

Climate data store

C3S:

Copernicus climate change service

DLR:

German Aerospace Center

EEA:

European environmental agency

MEM:

Maximum entropy method

NO2 :

Nitrogen dioxide

O3 :

Ozone

PM:

Particulate matter

PM10 :

Coarse particular matter

PM2.5 :

Fine particular matter

SO2 :

Sulfur dioxide

UV:

Ultraviolet

References

  1. Lambert LC, Fauci AS. Influenza vaccines for the future. N Engl J Med. 2010;363(21):2036–44.

    Article  CAS  Google Scholar 

  2. Dowell SF. Seasonal variation in host susceptibility and cycles of certain infectious diseases. Emerg Infect Dis. 2001;7(3):369–74.

    Article  CAS  Google Scholar 

  3. Lipsitch M, Viboud C. Influenza seasonality: lifting the fog. Proc Natl Acad Sci U S A. 2009;106(10):3645–6.

    Article  CAS  Google Scholar 

  4. Tamerius J, Nelson MI, Zhou SZ, Viboud C, Miller MA, Alonso WJ. Global influenza seasonality: reconciling patterns across temperate and tropical regions. Environ Health Perspect. 2011;119(4):439–45.

    Article  Google Scholar 

  5. Yang L, Wong CM, Lau EH, Chan KP, Ou CQ, Peiris JS. Synchrony of clinical and laboratory surveillance for influenza in Hong Kong. PLoS ONE. 2008;3(1):e1399.

    Article  Google Scholar 

  6. Martins LC, Latorre Mdo R, Cardoso MR, Goncalves FL, Saldiva PH, Braga AL. Air pollution and emergency room visits due to pneumonia and influenza in Sao Paulo, Brazil. Rev Saude Publica. 2002;36(1):88–94.

    Article  Google Scholar 

  7. Wong CM, Yang L, Thach TQ, Chau PY, Chan KP, Thomas GN, Lam TH, Wong TW, Hedley AJ, Peiris JS. Modification by influenza on health effects of air pollution in Hong Kong. Environ Health Perspect. 2009;117(2):248–53.

    Article  CAS  Google Scholar 

  8. Xu Z, Hu W, Williams G, Clements AC, Kan H, Tong S. Air pollution, temperature and pediatric influenza in Brisbane, Australia. Environ Int. 2013;59:384–8.

    Article  CAS  Google Scholar 

  9. Meng Y, Lu Y, Xiang H, Liu S. Short-term effects of ambient air pollution on the incidence of influenza in Wuhan, China: a time-series analysis. Environ Res. 2021;192:110327.

    Article  CAS  Google Scholar 

  10. Murtas R, Russo AG. Effects of pollution, low temperature and influenza syndrome on the excess mortality risk in winter 2016–2017. BMC Public Health. 2019;19(1):1445.

    Article  Google Scholar 

  11. Bell ML, Davis DL, Fletcher T. A retrospective assessment of mortality from the London smog episode of 1952: the role of influenza and pollution. Environ Health Perspect. 2004;112(1):6–8.

    Article  Google Scholar 

  12. Clay K, Lewis J, Severnini E. Pollution, Infectious Disease, and Mortality: Evidence from the 1918 Spanish Influenza Pandemic. J Econ Hist. 2018;78(4):1179–209.

    Article  Google Scholar 

  13. Shaman J, Pitzer VE, Viboud C, Grenfell BT, Lipsitch M. Absolute humidity and the seasonal onset of influenza in the continental United States. PLoS Biol. 2010;8(2):e1000316.

    Article  Google Scholar 

  14. Shaman J, Kohn M. Absolute humidity modulates influenza survival, transmission, and seasonality. Proc Natl Acad Sci U S A. 2009;106(9):3243–8.

    Article  CAS  Google Scholar 

  15. DeFelice TP. Relationship between temporal anomalies in PM2.5 concentrations and reported influenza/influenza-like illness activity. Heliyon. 2020;6(8):e04726.

    Article  CAS  Google Scholar 

  16. Su W, Wu X, Geng X, Zhao X, Liu Q, Liu T. The short-term effects of air pollutants on influenza-like illness in Jinan, China. BMC Public Health. 2019;19(1):1319.

    Article  Google Scholar 

  17. Huang L, Zhou L, Chen J, Chen K, Liu Y, Chen X, Tang F. Acute effects of air pollution on influenza-like illness in Nanjing, China: a population-based study. Chemosphere. 2016;147:180–7.

    Article  CAS  Google Scholar 

  18. Zhang Z, Chai P, Wang J, Ye Z, Shen P, Lu H, Jin M, Gu M, Li D, Lin H, et al. Association of particulate matter air pollution and hospital visits for respiratory diseases: a time-series study from China. Environ Sci Pollut Res Int. 2019;26(12):12280–7.

    Article  CAS  Google Scholar 

  19. Inness A, Baier F, Benedetti A, Bouarar I, Chabrillat S, Clark H, Clerbaux C, Coheur P, Engelen RJ, Errera Q, et al. The MACC reanalysis: an 8 yr data set of atmospheric composition. Atmos Chem Phys. 2013;13(8):4073–109.

    Article  CAS  Google Scholar 

  20. Theodosi C, Tsagkaraki M, Zarmpas P, Grivas G, Liakakou E, Paraskevopoulou D, Lianou M, Gerasopoulos E, Mihalopoulos N. Multi-year chemical composition of the fine-aerosol fraction in Athens, Greece, with emphasis on the contribution of residential heating in wintertime. Atmos Chem Phys. 2018;18(19):14371–91.

    Article  CAS  Google Scholar 

  21. Akritidis D, Katragkou E, Georgoulias AK, Zanis P, Kartsios S, Flemming J, Inness A, Douros J, Eskes H. A complex aerosol transport event over Europe during the 2017 Storm Ophelia in CAMS forecast systems: analysis and evaluation. Atmos Chem Phys. 2020;20(21):13557–78.

    Article  CAS  Google Scholar 

  22. Marécal V, Peuch VH, Andersson C, Andersson S, Arteta J, Beekmann M, Benedictow A, Bergström R, Bessagnet B, Cansado A, et al. A regional air quality forecasting system over Europe: the MACC-II daily ensemble production. Geosci Model Dev. 2015;8(9):2777–813.

    Article  Google Scholar 

  23. ERA5: data documentation [https://confluence.ecmwf.int/display/CKB/ERA5%3A+data+documentation]

  24. Hersbach H, Bell B, Berrisford P, Hirahara S, Horanyi A, Munoz-Sabater J, Nicolas J, Peubey C, Radu R, Schepers D, et al. The ERA5 global reanalysis. Q J Roy Meteor Soc. 2020;146(730):1999–2049.

    Article  Google Scholar 

  25. ERA5-Land hourly data from 1981 to present [https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=overview]

  26. Bolton D. The Computation of Equivalent Potential Temperature. Mon Weather Rev. 1980;108(7):1046–53.

    Article  Google Scholar 

  27. Esri Deutschland. Postleitzahlengebiete - OSM. In. Edited by Esri Deutschland Content. https://opendata-esri-de.opendata.arcgis.com/datasets/5b203df4357844c8a6715d7d411a8341_0.

  28. Wood SN. Generalized additive models : an introduction with R. 2nd ed. Boca Raton: CRC Press/Taylor & Francis Group; 2017.

  29. Hardin JW, Hilbe JM. Generalized linear models and extensions. 2nd ed. College Station, Tex.: Stata Press; 2007.

    Google Scholar 

  30. ECDC: Risk Assessment. Risk Assessment. Seasonal influenza 2011–2012 in Europe (EU/EEA countries). Stockholm: European Centre for Disease Prevention and Control; 2012. p. 1–18.

    Google Scholar 

  31. Chen W, Stegen G. Experiments with maximum entropy power spectra of sinusoids. J Geophys Res. 1974;79(20):3019–22.

    Article  Google Scholar 

  32. Ulrych TJ, Bishop TN. Maximum entropy spectral analysis and autoregressive decomposition. Rev Geophys. 1975;13(1):183–200.

    Article  Google Scholar 

  33. Wust S, Bittner M. Non-linear resonant wave-wave interaction (triad): Case studies based on rocket data and first application to satellite data. J Atmos Sol-Terr Phy. 2006;68(9):959–76.

    Article  Google Scholar 

  34. Joung YH, Jang TS, Kim JK. Association among sentinel surveillance, meteorological factors, and infectious disease in Gwangju, Korea. Environ Sci Pollut Res Int. 2022;29(12):17561–9.

    Article  Google Scholar 

  35. Lindner-Cendrowska K, Brode P. Impact of biometeorological conditions and air pollution on influenza-like illnesses incidence in Warsaw. Int J Biometeorol. 2021;65(6):929–44.

    Article  Google Scholar 

  36. Marr LC, Tang JW, Van Mullekom J, Lakdawala SS. Mechanistic insights into the effect of humidity on airborne influenza virus survival, transmission and incidence. J R Soc Interface. 2019;16(150):20180298.

    Article  CAS  Google Scholar 

  37. Nishioka K, Daidoji T, Nakaya T. Downregulation of calcium-regulated heat stable protein 1 expression by low-temperature stimulation causes reduction of interferon-beta expression and sensitivity to influenza viral infection. Virus Res. 2022;309:198659.

    Article  CAS  Google Scholar 

  38. Schuit M, Gardner S, Wood S, Bower K, Williams G, Freeburger D, Dabisch P. The Influence of Simulated Sunlight on the Inactivation of Influenza Virus in Aerosols. J Infect Dis. 2020;221(3):372–8.

    Article  Google Scholar 

  39. de Haar C, Hassing I, Bol M, Bleumink R, Pieters R. Ultrafine but not fine particulate matter causes airway inflammation and allergic airway sensitization to co-administered antigen in mice. Clin Exp Allergy. 2006;36(11):1469–79.

    Article  Google Scholar 

  40. Diffey BL. An overview analysis of the time people spend outdoors. Br J Dermatol. 2011;164(4):848–54.

    Article  CAS  Google Scholar 

  41. Zhang L, Ou C, Magana-Arachchi D, Vithanage M, Vanka KS, Palanisami T, Masakorala K, Wijesekara H, Yan Y, Bolan N, et al. Indoor particulate matter in Urban households: sources, pathways, characteristics, health effects, and exposure mitigation. Int J Environ Res Public Health. 2021;18(21):11055.

    Article  CAS  Google Scholar 

  42. Han Y, Qi M, Chen Y, Shen H, Liu J, Huang Y, Chen H, Liu W, Wang X, Liu J, et al. Influences of ambient air PM(2).(5) concentration and meteorological condition on the indoor PM(2).(5) concentrations in a residential apartment in Beijing using a new approach. Environ Pollut. 2015;205:307–14.

    Article  CAS  Google Scholar 

  43. Wang B, Liu Y, Li Z, Li Z. Association of indoor air pollution from coal combustion with influenza-like illness in housewives. Environ Pollut. 2016;216:646–52.

    Article  CAS  Google Scholar 

  44. Zheng S, Zhang J, Mou J, Du W, Yu Y, Wang L. The influence of relative humidity and ground material on indoor walking-induced particle resuspension. J Environ Sci Health A Tox Hazard Subst Environ Eng. 2019;54(10):1044–53.

    Article  CAS  Google Scholar 

  45. Ogen Y. Assessing nitrogen dioxide (NO2) levels as a contributing factor to coronavirus (COVID-19) fatality. Sci Total Environ. 2020;726:138605.

    Article  CAS  Google Scholar 

  46. Filippini T, Rothman KJ, Goffi A, Ferrari F, Maffeis G, Orsini N, Vinceti M. Satellite-detected tropospheric nitrogen dioxide and spread of SARS-CoV-2 infection in Northern Italy. Sci Total Environ. 2020;739:140278.

    Article  CAS  Google Scholar 

  47. Basray R, Malik A, Waqar W, Chaudhry A, Wasif Malik M, Ali Khan M, Ansari JA, Ikram A. Impact of environmental factors on COVID-19 cases and mortalities in major cities of Pakistan. J Biosaf Biosecur. 2021;3(1):10–6.

    Article  CAS  Google Scholar 

  48. Ma Y, Zhao Y, Liu J, He X, Wang B, Fu S, Yan J, Niu J, Zhou J, Luo B. Effects of temperature variation and humidity on the death of COVID-19 in Wuhan, China. Sci Total Environ. 2020;724:138226.

    Article  CAS  Google Scholar 

  49. Ma Y, Pei S, Shaman J, Dubrow R, Chen K. Role of air temperature and humidity in the transmission of SARS-CoV-2 in the United States. medRxiv [Preprint]. 2020:2020.11.13.20231472. https://doi.org/10.1101/2020.11.13.20231472. Update in: Nat Commun. 2021;12(1):3602.

  50. Wu X, Nethery RC, Sabath BM, Braun D, Dominici F. Exposure to air pollution and COVID-19 mortality in the United States: a nationwide cross-sectional study. medRxiv. 2020:2020.2004.2005.20054502.

  51. Poirier C, Luo W, Majumder MS, Liu D, Mandl KD, Mooring TA, Santillana M. The role of environmental factors on transmission rates of the COVID-19 outbreak: an initial assessment in two spatial scales. Sci Rep. 2020;10(1):17002.

    Article  CAS  Google Scholar 

  52. Liang L, Gong P, Cong N, Li Z, Zhao Y, Chen Y. Assessment of personal exposure to particulate air pollution: the first result of City Health Outlook (CHO) project. BMC Public Health. 2019;19(1):711.

    Article  Google Scholar 

Download references

Acknowledgements

The authors are grateful to Klaus Hamacher for enabling this collaborative work.

Funding

Open Access funding enabled and organized by Projekt DEAL. This study was performed with internal funding of DLR and AOK Baden-Württemberg.

Author information

Authors and Affiliations

Authors

Contributions

JR, SDe, JJ, TA and MB conceived the research question and approach. LG and TE prepared environmental data, SDa prepared health data, and JR and LG merged the joint data base. SW performed spectral analyses, and MS and JR performed statistical analyses. LG performed the environmental data analysis and prepared Figures 1, 2A & B, and 3, S.W. prepared Figure 2C, and JR prepared Figure 4. JR, LG, MS and TE wrote the mansucript. All authors reviewed the manuscript. The author(s) read and approved the final manuscript.

Corresponding author

Correspondence to Jörn Rittweger.

Ethics declarations

Ethics approval and consent to participate

The study was approved by the Ethical Committee of the Medical Council (Landesärztekammer Nordrhein) within the VARIAQ study (lfd Nr 2020092).

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Rittweger, J., Gilardi, L., Baltruweit, M. et al. Temperature and particulate matter as environmental factors associated with seasonality of influenza incidence – an approach using Earth observation-based modeling in a health insurance cohort study from Baden-Württemberg (Germany). Environ Health 21, 131 (2022). https://doi.org/10.1186/s12940-022-00927-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12940-022-00927-y

Keywords

  • Air quality
  • Epidemiology
  • Disease burden
  • Infectious disease