Pollen and viruses contribute to spatio-temporal variation in asthma-related emergency department visits

Abstract

Background:

Asthma exacerbations are an important cause of emergency department visits but much remains unknown about the role of environmental triggers including viruses and allergenic pollen. A better understanding of spatio-temporal variation in exposure and risk posed by viruses and pollen types could help prioritize public health interventions.

Objective:

Here we quantify the effects of regionally important Cupressaceae pollen, tree pollen, other pollen types, rhinovirus, seasonal coronavirus, respiratory syncytial virus, and influenza on asthma-related emergency department visits for people living near eight pollen monitoring stations in Texas.

Methods:

We used age stratified Poisson regression analyses to quantify the effects of allergenic pollen and viruses on asthma-related emergency department visits.

Results:

Young children (< 5 years of age) had high asthma-related emergency department rates (24.1 visits/1,000,000 person-days), which were mainly attributed to viruses (51.2%). School-aged children also had high rates (20.7 visits/1,000,000 person-days), which were attributed to viruses (57.0%), Cupressaceae pollen (0.7%), and tree pollen (2.8%). Adults had lower rates (8.1 visits/1,000,000 person-days) which were attributed to viruses (25.4%), Cupressaceae pollen (0.8%), and tree pollen (2.3%). This risk was spread unevenly across space and time; for example, during peak Cuppressaceae season, this pollen accounted for 8.2% of adult emergency department visits near Austin where these plants are abundant, but 0.4% in cities like Houston where they are not; results for other age groups were similar.

Conclusions:

Although viruses are a major contributor to asthma-related emergency department visits, airborne pollen can explain a meaningful portion of visits during peak pollen season and this risk varies over both time and space because of differences in plant composition.

1. Introduction¹

In the United States there are 19.2 million adults and 5.5 million children with asthma (Centers for Disease Control and Prevention, 2018), and 2.6% of all pediatric and 0.9% of all adult emergency department visits are asthma-related (Qin et al., 2020). Of the approximately 1.2 million annual asthma-related emergency department visits, approximately 20% result in hospitalization (Lin et al., 2020). In 2010, each emergency department visit from a Medicaid/CHIP patient cost an average of $433; in Texas such visits annually cost $24 million (Pearson et al., 2014). Understanding the effect of environmental triggers on asthma-related emergency department visits (AREDV) could help to weigh the relative importance of potential asthma-related public health interventions for pollen (e.g., exposure avoidance) and viruses (e.g., infection prevention measures) with other asthma-related interventions such as e-health asthma management systems (Merchant et al., 2016; Miller et al., 2017), home interventions to reduce allergens (Matsui et al., 2017; Rabito et al., 2017), and educational campaigns (Pinnock et al., 2017). As the absolute and relative importance of these environmental triggers vary over space and time, so might the impact of potential interventions.

Many studies have found associations between airborne pollen concentrations and AREDV (Annesi-Maesano et al., 2023; Batra et al., 2022; De Roos et al., 2020; Erbas et al., 2018; Lee et al., 2019; Osborne et al., 2017; Sun et al., 2016). However, very few studies have investigated associations between pollen and AREDV in multiple places (but see Dales et al., 2008). This makes it difficult to understand how attributable risk in AREDV varies over space; it is challenging to directly compare results across studies because of differences in study methodology (Shrestha et al., 2021). Thus, even though differences in pollen concentrations among cities are recorded by pollen monitoring networks and described in the literature (Lo et al., 2019), there is not a good understanding of how the public health consequences of pollen vary among cities. Analyzing the effects of pollen on AREDV simultaneously in multiple areas could also help to untangle the public health effects of spatial differences in plant composition, such as those caused by regionally important plant taxa.

Viruses are an important trigger of asthma exacerbations and their contribution to AREDV has long been recognized (Eggo et al., 2016; Jackson & Johnston, 2010; Johnston et al., 2005). The prevalence of upper respiratory viral infections is highly seasonal, which could confound time series analyses of other seasonal triggers such as pollen and there is also the possibility of interactions between pollen and viruses (Gilles et al., 2020) although considerable uncertainty about these interactions remains (Martikainen et al., 2023). Moreover, viral prevalence and the associated AREDV can vary spatially (Eggo et al., 2016; Trinh et al., 2018). Thus, epidemiological investigations of pollen should ideally include viral prevalence to prevent potential confounding; however, few studies include both viral prevalence and airborne pollen (but see Erbas et al., 2015; Murray et al., 2006).

Thus, there remains a need for a better understanding of how both pollen and viruses contribute to spatio-temporal variation in AREDV. In this study, we quantify the spatio-temporal risk of AREDV attributable to several viruses and pollen types including pollen from a regionally-important species of Cupressaceae. To do so, we analyze time series of these variables over five years for populations living near eight pollen monitoring stations in Texas using age stratified distributed lag non-linear models.

2. Materials and methods

2.1. Emergency department visits

Emergency department visits in Texas from October 2015 through December 2020 were obtained from the Texas Department of State Health Services via the Texas Health Care Information Collection. This study was approved the Texas Department of State Health Services Institutional Review Board and determined to be exempt by the University of Texas at Austin Institutional Review Board. We identified inpatient and outpatient visits where asthma was the primary diagnosis through standardized diagnosis codes used in the International Classification of Diseases (ICD 9: 493, ICD 10: J45). This data included the census block group of each patient’s home address; we retained all visits where the census block group centroid was within 25 km of an included pollen monitoring station. 25 km is similar to the maximum distance included in several previous epidemiological analyses of allergenic pollen (Darrow et al., 2012; Osborne et al., 2017; Sun et al., 2016) and sensitivity analyses were conducted for cut-off distances of 10 km and 50 km. For visits where the census block group was unavailable we instead used the census tract centroid (3.9%) or zip code centroid (4.8%). Catchment population was calculated based on the 2017 American Community Survey (US Census Bureau, 2016) estimates accessed via the R package ‘tidycensus’ (Walker & Herman, 2022). Population based incidence rates expressed per 1,000,000 person-days were calculated by dividing the number of AREDV per day by the population of census blocks within that area and multiplying by 1,000,000. After exploring several possible age categories that were informed by known variability in allergic sensitization and asthma-related emergency department visits,(Qin et al., 2020; Quest Diagnostics Health Trends, 2011) we conducted separate analyses for young children (ages 0 – 4), school-aged children (5 – 17), and adults (≥18); these strata match those used in similar studies (Darrow et al., 2012).

2.2. Airborne pollen

In the United States, airborne pollen concentrations are measured by the National Allergy Bureau (NAB) at approximately 80 pollen monitoring stations. We obtained pollen data from 2015 to 2020 from eight of these NAB stations in Texas, including: College Station, Dallas, Flower Mound, Georgetown (adjacent to Austin), Houston, San Antonio A, San Antonio B, and Waco (see map in SI 1). These NAB stations use Burkard samplers that intake air at a rate of 10 L/min. NAB monitoring guidelines stipulate that pollen measurements are to be collected in integrated 24-hour samples at heights of 5 – 15 m and that samplers are placed away from overhanging vegetation in areas where the air flow is not obstructed by buildings or other structural features. Samples are prepared, stained, and generally observed at a magnification of 400×. Trained and NAB certified personnel manually identify all pollen along at least one latitudinal traverse, which typically contains pollen from approximately 0.5m³ of sampled air. Here, we group pollen taxa in three types: Cupressaceae, trees (not including Cupressaceae), and other pollen types (including herbaceous plants such as ragweed and Poaceae). We separate out Cupressaceae pollen in our analysis because it is especially important allergenically (as described below), peaks when there are few other types of pollen in the air, varies substantially in concentrations across the study region, and because it represents 32% of all airborne pollen collected; composition of pollen at each station is provided in SI 2.

Cupressaceae includes Juniperus ashei (J. Buchholz), commonly known as mountain cedar or Ashe’s juniper, which is one of North America’s most infamous pollen producing plants (Mendoza & Quinn, 2021) and whose pollen causes severe allergic responses that are locally known as “cedar fever.” This species is abundant in central Texas near Austin, San Antonio, and Waco (Ellenwood et al., 2015); a map of J. ashei basal area is included in SI 1. In San Antonio, 40% of atopic patients were sensitized to Cupressaceae pollen (Calabria et al., 2007). This species is so clinically important in part because it is an exceptionally prolific producer of allergenic pollen; male trees can produce up to 1.3 million pollen cones annually, each of which contains approximately 400,000 pollen grains (Bunderson et al., 2012). Cupressaceae pollen is also especially prone to rupturing (Galveias et al., 2021), which increases exposure to allergens contained within J. ashei pollen, such as Jun a 1, a large and heavily glycosylated protein that activates allergic responses through both IgE dependent and independent pathways (Mendoza & Quinn, 2021).

Some pollen data were missing, including after October 26, 2019 for College Station and after March 13^th 2018 for Dallas; these were excluded from the analysis. In addition, all pollen time series were missing some data ranging from 1.3% (Flower Mound) to 59.0% of observations (College Station), often because some stations did not take measurements on weekends (Georgetown, Houston, and Waco). Epidemiological analyses of pollen often use linear interpolation to fill in missing data (Sun et al., 2016) although other interpolation methods are available (Picornell et al., 2021), and after exploring several modeling methods, we used linear interpolation to fill in gaps of either one or two days (and in doing so, assume that the data are missing at random). To assess the accuracy of the linear interpolation, we removed a subset of random observations (10%) and modeled them using linear interpolation, and then compared the predictions of interpolated datapoints with actual values. The R² between the interpolated and actual values for the different pollen categories were: 0.57 (Cupressaceae), 0.53 (other trees), and 0.65 (other pollen); further details on both missing data and the accuracy of the linear imputations are provided in SI 3. The accuracy of our linear interpolation is similar to those reported elsewhere (Picornell et al., 2021).

2.3. Viruses

The National Respiratory and Enteric Virus Surveillance System (NREVSS) collects data on several respiratory viruses by week at the city scale. While NREVSS is a passive and voluntary system, does not provide patient demographic information (e.g., age), and the number of contributing labs can vary seasonally, it can still be very useful for quantifying the seasonality of respiratory infections (Killerby et al., 2018; Midgley et al., 2017). We were provided access to the number of tests and the proportion of positive tests for rhinovirus, seasonal coronavirus, respiratory syncytial virus (RSV), and influenza virus by the Texas Department of State Health Services. Test approaches included antigen, PCR, and viral isolation methods. During our study period, the average annual number of tests administered for each virus across our study areas were: 8,396 (rhinovirus), 7,453 (seasonal coronavirus), 14,388 (RSV), and 24,865 (influenza virus). The number of tests varied among areas as well; for example, for influenza virus tests, there were an average of 826 tests per year reported in College Station compared to 33,653 in the Dallas metropolitan area. We explored several ways to include this data in the model (including the number of positive tests, the proportion of positive tests, and the number of total tests administered, and moving averages of each variable from 1 – 4 weeks), and ultimately selected a three-week moving average of the proportion of positive tests. The moving average of viral data allowed us to reduce stochasticity, which was especially important in areas with lower populations and therefore fewer viral tests (e.g., College Station). Seasonal coronavirus data were not available for Houston, so we used the state average instead. Influenza data were also not available for Austin (near the Georgetown pollen monitoring station) from June 4, 2016 – September 24, 2016; influenza rates in other cities were near zero during this period, so we assumed that there were no positive tests of influenza in Austin in that period. Viral prevalence in Texas can vary over finer time scales than the 3 week average used here (Eggo et al., 2016) and age information is not included so surveillance data may not match the ages included in the present analysis. To better capture these viral dynamics and the direct effects of whether students were in school, we also included an additional term for holidays, along with up to a two-week lag, based on models of viral dynamics in Texas (Eggo et al., 2016). While other viruses (e.g., metapneumovirus and bocavirus) have been documented to trigger asthma exacerbations (Coverstone et al., 2019; Papadopoulos et al., 2011), their prevalence is often lower (Monto et al., 2014) and a lack of data precluded their inclusion here.

2.4. Statistical analysis

Distributed lag non-linear models (DLNM) provide a way to analyze the cumulative effects of exposure across multiple lag lengths (Gasparrini et al., 2010) with the R package ‘dlnm’ (Gasparrini, 2011). This approach is widely used in environmental epidemiological analyses (Bhaskaran et al., 2013), and has previously been applied to allergenic pollen (Ito et al., 2015; Lee et al., 2019; Sun et al., 2016). Here, the relationship between pollen concentrations and AREDV was modeled using lags of up to seven days, approximately the maximum length used in some other analyses of pollen (Guilbert et al., 2016; Lee et al., 2019; Sun et al., 2016) and which corresponds to the approximate timing of late-phase allergic responses and priming effects (Badorrek et al., 2011; Jacobs et al., 2012) including for J. ashei (Jacobs et al., 2014). Lags can also help account for differences in the timing of pollen release within a city (Katz et al., 2019). Separate analyses were conducted for each age stratum. We modeled the effect of pollen on AREDV as well as the shape of the lag with natural splines (indicated by ‘ns’ in the equation below). We explored using between 1 and 5 degrees of freedom for the response and the lag shape and for each age stratum selected the degrees of freedom using quasi-Akaike Information Criterion (QAIC) (Gasparrini et al., 2010). QAIC was also used for other model selection decisions (e.g., which type of viral index to include); we selected the model with the lowest QAIC. Day of the week (weekday) and fixed effects for each sampling station were also included in each model. We accounted for general seasonal trends and trends over time (e.g., reduced AREDV after COVID-19 cases beginning in March 2020) using a natural spline for the day since the start of study (df = 4 per year). We also extracted school calendar data for each metro area and included a term for holiday, including a two-week lag as described above. Minimum and maximum daily temperatures were also included in the model. Models for each age group were fit with a generalized linear model assuming a quasi-Poisson distribution where the expected count of asthma-related ED visits λ on day t near NAB station m, with surrounding population pop was modeled as a function of pollen categories Cupressaceae pollen, tree pollen, and other pollen at lags of 7 days, proportion of positive tests for rhinovirus, coronavirus, RSV, and influenza, and city fixed effect city. β represents the log relative rates for each virus, and δ represents the log relative rates for temperatures. The final model for young children was:

$$\log \left({\lambda }_{t,m}/{\mathit{pop}}_m\right)=\text{ns}({\mathit{Cupressaceae\; pollen}}_{t,m,},3\mathit{df})+\text{ns}\left({\mathit{tree\; pollen}}_{t,m,},3\mathit{df}\right)+\text{ns}({\mathit{other\; pollen}}_{t,m,},3\mathit{df})+{\text{β}}_{\mathit{rhinovirus}}\times {\mathit{rhinovirus}}_{,\mathit{week}\mathit{\left(}t\mathit{\right)},m}+{\text{β}}_{\mathit{Corona}}\times {\mathit{coronavirus}}_{,\mathit{week}\mathit{\left(}t\mathit{\right)},m}+{\text{β}}_{\mathit{RSV}}\times {\mathit{RSV}}_{,\mathit{week}\mathit{\left(}t\mathit{\right)},m}+{\text{β}}_{\mathit{influenza}}\times {\mathit{influenza}}_{,\mathit{week}\mathit{\left(}t\mathit{\right)},m}+{\mathit{city}}_m+{\mathit{weekday}}_t+{\delta }_1\times \mathit{Minimum\; temperature}+{\delta }_2\times \mathit{Maximum\; temperature}+\text{ns}\left(\text{day of study},4\mathit{df}/\mathit{year}\right)+\text{ns}\left({\mathit{school\; vacation}}_{t,m,},3\mathit{df}\right)$$

Remaining partial autocorrelation was accounted for by adding the one-day lagged deviance residuals to the model (Bhaskaran et al., 2013). Attributable risk from pollen and viruses was calculated using an established method within the DLNM model framework (Gasparrini & Leone, 2014). In this approach, the number of attributable cases on each day is defined as the fraction of attributable cases multiplied by the total number of observed cases. The fraction of cases is in turn calculated from the sum of the contributions from pollen and viruses from previous days within the lag window in the time series (i.e., the ‘backward perspective’) (Gasparrini & Leone, 2014). Empirical confidence intervals were calculated with Monte Carlo simulations. Attributable risk from each pollen and virus type was visualized by averaging the daily attributable risk at each site and week of the year across the study period. We also conducted additional sensitivity analyses to see how changing several variables affected AREDV, including: excluding either viruses or pollen from the model, changing the distance threshold from NAB stations to 10 km or 50 km, substituting average daily temperature for both minimum and maximum daily temperature. Potential interactions between viruses and pollen were also investigated, but interactions were minor and not retained in the final model. All analyses were conducted in R 4.0.3;(R Core Team, 2018) other packages used included sf (Pebesma, 2018), daymetr for downloading Daymet temperature data (Hufkens et al., 2018; Thornton et al., 2020), and several packages within the tidyverse (Wickham et al., 2019).

Results

3.1. TIME SERIES OF VARIABLES

From October 2015 – December 2020, there were a total of 649,754 emergency department visits in Texas for which asthma was the primary diagnosis. Of these, 177,988 were within 25 km of the eight pollen monitoring stations in Texas. In general, the population-based incidence rate was highest in San Antonio and tended to be lower in Dallas and Houston (Table 1). For school-aged children, the average daily number of AREDV per 1,000,000 residents living within 25 km of an NAB station were: 15.1 (Dallas), 15.3 (Georgetown), 15.0 (Flower Mound), 14.9 (Houston), 24.1 (San Antonio A), 35.0 (San Antonio B), and 26.4 (Waco). Averaging across study regions, young children had the highest rates (24.1), followed by school-age children (20.7) and adults (8.1). There were seasonal patterns in AREDV for young children (Fig. 1A) and school-age children (Fig. 1B) and somewhat weaker seasonal patterns for adults (Fig. 1C).

Table 1:

Description of study areas within 25 km of a pollen monitoring station including population, number and rate of asthma-related emergency department (ARED) visits as well as the total number of visits during the study period.

	Young children (0 – 4)			Children (5 – 17)			Adults (≥18)
Station	Total population	ARED visits (per 1,000,000 people/day)	Total ARED visits	Total population	ARED visits (per 1,000,000 people/day)	Total ARED visits	Total population	ARED visits (per 1,000,000 people/day)	Total ARED visits
College Station	13,293	23.4	598	30,829	19.5	1,153	170,535	6.9	2,251
Dallas	128,820	15.0	3,720	348,896	15.1	10,134	1,400,485	6.1	16,498
Flower Mound	68,189	16.2	2,117	197,997	15.0	5,693	769,937	7.2	10,704
Georgetown	48,593	19.8	1,847	132,531	15.3	3,886	518,190	6.9	6,892
Houston	227,334	17.6	7,693	536,443	14.9	15,311	2,165,265	7.7	31,848
San Antonio A	81,738	28.7	4,513	214,920	24.1	9,956	899,893	8.8	15,144
San Antonio B	50,265	39.3	3,790	127,825	35.0	8,605	448,155	10.6	9,138
Waco	16,519	32.4	1,027	40,295	26.4	2,045	172,096	10.44	3,425
Total	634,751		25,305	1,629,736		56,783	6,544,556		95,900

Figure 1:

Times series of dependent and independent variables: Daily asthma-related emergency department visits rolling weekly mean for individual stations (gray lines) and averaged across all stations (black lines) for A) Children aged 0 – 4, B) school aged children (5 – 17), and C) Adults (18 +); D) Airborne pollen concentrations for Cupressaceae pollen across the study period, E) airborne pollen concentrations for trees, F) airborne pollen concentrations for other pollen (weeds and grasses), G) Average proportion of tests that were positive for rhinovirus (green), influenza (blue), seasonal coronavirus (purple), and RSV (yellow) across the study regions.

Each pollen type exhibited strong seasonal patterns; Cupressaceae pollen was highest in the winter (approximately December – February; Fig 1D), tree pollen was highest in the spring (March – April; Fig. 1E), and other pollen types tended to be highest in the fall (September – October; Fig. 1F). Average weekly concentrations of both Cupressaceae and tree pollen types often exceeded 1,000 pollen grains/m³ during peak seasons, whereas the other pollen category was only occasionally >100 pollen grains/m³.

The proportion of positive tests for viruses also exhibited strong seasonality (Fig. 1G). Rhinovirus infections tended to peak in September but were also present throughout much of the year whereas influenza peaked in December – March, seasonal coronavirus peaked in December – January, and RSV peaked in December. There were also substantial differences in the seasonality of the proportion of positive tests for viruses among years (e.g., spikes in influenza at the end of 2017 and rhinovirus in the spring of 2017). For some viruses, there were meaningful differences between regions (e.g., rhinovirus and RSV), whereas for others (e.g., seasonal coronavirus) there were few differences between regions (SI 4).

3.2. Associations between ED visits, pollen, and viruses

For school-aged children, there were positive associations between AREDV and tree pollen and Cupressaceae pollen (Fig. 2A–D). For adults (Fig. 2E–F), there were positive associations between AREDV and each pollen type (Cupressaceae, tree pollen, and other pollen). The strongest effects were often at the highest pollen concentrations and sometimes at the longest lags. The associations between pollen concentrations and AREDV rates for young children ages 0 – 4 were small but statistically significant (SI 5).

Figure 2:

Relative risk (RR) of emergency department visits for school age children (ages 5 – 17) and adults (ages 18+) associated with Cupressaceae pollen across all lags (A, E) and at each lag and pollen concentration (B, F). RR for tree pollen across all lags (C, G) and at each lag and pollen concentration (D, H). The distribution of observed pollen concentrations are shown in rug plots at the top of figures A, C, E, G; darker areas indicate more observations. Figures are truncated at the 99^th percentile of pollen concentrations.

3.3. Attributable risk

Attributable risk (Table 2) for young children was highest for rhinovirus (40.8%), followed by RSV (4.2%), influenza (3.5%), and coronavirus (2.7%); tree pollen (1.1%) and Cupressaceae pollen (0.9%) were substantially lower. The sum of the attributable risk explained by these variables was 53.2%. For school aged children, the relative importance of attributable risk was: rhinovirus (42.1%), influenza (8.3%), RSV (4.2%), seasonal coronavirus (2.4%), tree pollen (2.8%), and Cupressaceae pollen (0.7%). The total percent of risk explained was 60.5%. For adults, the percent of attributable risk explained was far lower (30.8%), which was attributable to rhinovirus (18.0%), seasonal coronavirus (6.2%), RSV (1.2%), tree pollen (2.3%), other pollen (2.3%) and Cupressaceae pollen (0.8%). In total, over the study period, 4,113 visits were attributable to pollen.

Table 2:

Attributable risk for pollen (Cupressaceae and trees) and viruses over the study period for each age group (modeled separately). Estimates whose 95% confidence intervals do not cross zero are considered statistically significant.

	Attributable Risk for young children (0 – 4)		Attributable Risk for school-aged children (5 – 17)		Attributable Risk for adults (18+)
	Number of cases (mean ± 95 % CI)	Percent of cases (mean ± 95 % CI)	Number of cases (mean ± 95 % CI)	Percent of cases (mean ± 95 % CI)	Number of cases (mean ± 95 % CI)	Percent of cases (mean ± 95 % CI)
Cupressaceae	189 (76 – 298)	0.9 (0.4 – 1.5)	319 (135 – 495)	0.7 (0.3 – 1.1)	619 (394 – 838)	0.8 (0.5 – 1.1)
Trees	224 (59 – 384)	1.1 (0.3 – 1.9)	1,291 (1,036 – 1,535)	2.8 (2.3 – 3.4)	1,671 (1,372 – 1,961)	2.3 (1.9 – 2.7)
Rhinovirus	8,186 (6,914 – 9,353)	40.8 (34.5 – 46.7)	19,108 (17,123 – 20,957)	42.1 (37.8 – 46.2)	13,260 (10,238 – 16,194)	18.0 (13.9 – 21.9)
Coronavirus	549 (−503 – 1,554)	2.7 (−2.5 – 7.8)	1,105 (−534 – 2,708)	2.4 (−1.2 – 6.0)	4,552 (2,719 – 6,331)	6.2 (3.7 – 8.6)
RSV	839 (107 – 1,538)	4.2 (0.5 – 7.7)	1,909 (773 – 3,001)	4.2 (1.7 – 6.6)	873 (−570 – 2,246)	1.2 (−0.8 – 3.0)
Influenza	711 (−48 – 1,431)	3.5 (−0.2 – 7.1)	3,781 (2,684 – 4,827)	8.3 (5.9 – 10.6)	−695 (−2,279 – 844)	−0.9 (−3.1 – 1.1)
Total	10,698	53.2	27,513	60.5	22,705	30.8

The attributable risk of pollen and viruses varied greatly between seasons for young children (Fig. 3), school aged children (Fig. 4), and adults (Fig. 5). The timing and extent of risk also varied between cities for pollen (Table 3); for example, in January, the height of the Cupressaceae pollen season, school-aged children near the San Antonio A pollen monitoring station had a higher percent of attributable risk due to Cupressaceae pollen than Houston (9.0% vs 0.3%), reflecting underlying differences in Cupressaceae pollen abundance. For trees in March, the month with the highest tree pollen concentrations, the station with the lowest portion of attributable risk for school-aged children was Waco (7.3%), whereas the area with the highest portion of attributable risk was Houston (18.6%); averaged among cities the attributable risk from tree pollen was 11.6%.

Fig 3:

Estimated average attributable risk to young children (0 – 4) from pollen and viruses at each site across the calendar year (colored bands). The red line indicates the rate of AREDV observed within 25 km of each pollen monitoring station averaged for each week of the year. A one week rolling mean was applied to daily attributable risk values to ease visualization of seasonal patterns.

Fig 4:

Estimated average attributable risk to school-aged children (5 – 17) from pollen and viruses at each site across the calendar year (colored bands). The red line indicates the rate of AREDV observed within 25 km of each pollen monitoring station averaged for each week of the year. A one week rolling mean was applied to daily attributable risk values to ease visualization of seasonal patterns.

Fig 5:

Estimated average attributable risk to adults from pollen and viruses at each site across the calendar year (colored bands). The red line indicates the rate of AREDV observed within 25 km of each pollen monitoring station averaged for each week of the year. A one week rolling mean was applied to daily attributable risk values to ease visualization of seasonal patterns.

Table 3:

Effects of pollen during peak month for Cupressaceae (January) and tree pollen (March). Mean and 95% empirical confidence interval for attributable risk for pollen and viruses for each age group.

		Percent of attributable risk
		young children (0 – 4)	school-aged children (5 – 17)	adults (18+)
Effect of Cupressaceae during January	Austin/Georgetown	7.9 (2.4 – 13.0)	7.2 (3.1 – 11.2)	8.2 (5.4 – 11)
	College Station	1.3 (0.0 – 2.5)	0.9 (−0.1 – 1.9)	1.4 (0.7 – 2.0)
	Dallas	0.8 (0.1 – 1.4)	0.4 (0 – 0.9)	0.7 (0.3 – 1.0)
	Flower Mound	1.9 (0.3 – 3.4)	1.1 (−0.1 – 2.2)	1.7 (0.9 – 2.5)
	Houston	0.4 (0.0 – 0.7)	0.3 (0.0 – 0.5)	0.4 (0.2 – 0.6)
	San Antonio A	12.5 (6.3 – 18.1)	9.0 (4.2 – 13.3)	9.2 (5.7 – 12.5)
	San Antonio B	3.4 (0.5 – 6.2)	2.2 (0.1 – 4.2)	3.4 (1.8 – 4.8)
	Waco	3.8 (0.2 – 6.9)	2.1 (0.0 – 4.1)	3.3 (1.9 – 4.7)
Effect of Cupressaceae across all months	Austin/Georgetown	1.3 (0.4 – 2.1)	1.3 (0.6 – 2.1)	1.6 (1.0 – 2.1)
	College Station	0.6 (0.1 – 1.0)	0.4 (0.0 – 0.8)	0.5 (0.3 – 0.7)
	Dallas	0.2 (0.0 – 0.4)	0.1 (0.0 – 0.3)	0.2 (0.1 – 0.3)
	Flower Mound	0.5 (0.1 – 0.9)	0.3 (0.0 – 0.6)	0.4 (0.2 – 0.7)
	Houston	0.2 (0.0 – 0.3)	0.1 (0.0 – 0.2)	0.2 (0.1 – 0.2)
	San Antonio A	2.0 (1.0 – 3.0)	1.6 (0.9 – 2.3)	1.8 (1.2 – 2.3)
	San Antonio B	0.7 (0.2 – 1.3)	0.4 (0.1 – 0.7)	0.6 (0.3 – 0.8)
	Waco	0.6 (0.1 – 1.1)	0.4 (0.0 – 0.7)	0.7 (0.4 – 1.0)
Effect of trees during March	Austin/Georgetown	4.0 (0.6 – 7.2)	10.1 (8.0 – 12.2)	8.3 (6.8 – 9.7)
	College Station	6.8 (2.2 – 11.3)	12.0 (9.4 – 14.5)	10.8 (8.8 – 12.7)
	Dallas	3.1 (0.6 – 5.6)	7.6 (5.9 – 9.2)	6.1 (4.9 – 7.3)
	Flower Mound	5.4 (1.4 – 9.3)	11.6 (9.3 – 13.9)	9.4 (7.6 – 11.1)
	Houston	8.1 (1.3 – 14.2)	18.6 (14.8 – 22.2)	15.9 (13.0 – 18.6)
	San Antonio A	9.2 (3.7 – 14.3)	16.5 (13.2 – 19.5)	10.2 (7.4 – 12.8)
	San Antonio B	3.4 (0.3 – 6.4)	8.7 (6.9 – 10.5)	7.0 (5.7 – 8.4)
	Waco	3.2 (0.5 – 5.9)	7.3 (5.7 – 8.8)	5.7 (4.6 – 6.8)
Effect of trees across all months	Austin/Georgetown	0.6 (0.1 – 1.2)	2.0 (1.6 – 2.4)	1.4 (1.1 – 1.6)
	College Station	1.0 (0.2 – 1.7)	2.7 (2.1 – 3.2)	2.2 (1.8 – 2.6)
	Dallas	0.6 (0.1 – 1.1)	1.4 (1.1 – 1.7)	0.9 (0.8 – 1.1)
	Flower Mound	1.0 (0.2 – 1.9)	2.5 (2.0 – 3.0)	1.8 (1.4 – 2.1)
	Houston	1.0 (0.2 – 1.8)	2.8 (2.2 – 3.3)	2.3 (1.8 – 2.7)
	San Antonio A	1.6 (0.6 – 2.5)	3.5 (2.9 – 4.2)	2.2 (1.7 – 2.7)
	San Antonio B	0.6 (0.1 – 1.0)	1.6 (1.3 – 2.0)	1.1 (0.9 – 1.4)
	Waco	0.5 (0.1 – 1.0)	1.2 (0.9 – 1.5)	0.9 (0.7 – 1.1)

Viruses and pollen both contributed to explaining the observed seasonality of AREDV for school-aged children. For school aged children, high AREDV rates during September were attributable to rhinovirus (e.g., 52.7% in Georgetown/Austin in September), but peaks in early winter and spring were partly attributable to Cupressaceae and tree pollen, respectively. Omitting viruses from the model (SI 6) slightly but non-significantly decreased the effects of Cupressaceae for young children (0.5% vs 0.9%), school-aged children (0. 3% vs 0.7%) and adults (0.6% vs 0.8%); omitting viruses from the model significantly decreased the effect of tree pollen on young children (−1.0% vs 1.1%) and school-aged children (−0.5% vs 2.8%) but not adults (2.4% vs. 2.3%). Thus, analyzing the effects of pollen on AREDV was influenced by whether seasonal covarying viral prevalence was considered; however omitting pollen from the model had trivial effects on the risk attributable to viruses. Our results showed little sensitivity to which temperature metric was used (SI 7).

4. Discussion

4.1. Pollen

Although viruses explain a large portion of asthma-related ED visits, pollen can account for up to 18.6% of visits for certain age groups in the highest risk cities during peak pollen seasons. This is a meaningful percentage, especially considering that these attributable risk estimates are population-wide and do not account for pollen sensitization; the risk for sensitized individuals is presumably substantially higher. This study’s population of 8.8 million had approximately 1,151 pollen-induced ED visits per year that could be targeted by interventions (e.g., increasing controller medications before exposure or public health messaging about asthma and allergy medications) (Castillo et al., 2017; Giles et al., 2011). Better pollen forecasts provided by process-based models of pollen at granular scales (e.g., Katz et al., 2023) could potentially help people to reduce their own exposures or could be paired with precision digital health interventions for asthma (Mosnaim et al., 2021).

Spatial variation in regionally important Cupressaceae pollen concentrations in differences in attributable risk among cities. Few studies have demonstrated how attributable risk due to pollen varies between cities (but see(Dales et al., 2008), because most epidemiological analyses of airborne pollen concentrations investigate associations between airborne pollen and a health outcome in a single city or region (Erbas et al., 2018; Idrose et al., 2022; Shrestha et al., 2021). Although the effect of Cupressaceae pollen was minor in aggregate (~0.8% of total AREDV throughout the study area), it was associated with up to 9% of AREDV during peak season near the NAB station with the highest J. ashei abundance (see map of J. ashei basal area and sampling locations in SI 1 and relative pollen abundance at each sampling location in SI 2). There can also be substantial differences in pollen within cities; Cupressaceae pollen concentrations at the two San Antonio stations averaged 638 and 108 grains/m³, likely contributing to variation in intra-municipal AREDV. There were also important differences in tree pollen attributed AREDV between cities: for example, for school-aged children in March, attributable risk in Houston (18.6%) was double that of Dallas (7.6%). Thus, this study highlights how differences in plant composition among metropolitan areas can affect the timing and rates of AREDV.

Sensitization to pollen is low during early childhood (Calabria & Dice, 2007; Sheehan et al., 2010; Wong et al., 2012), so our finding of small but statistically significant effects of Cupressaceae and tree pollen on AREDV for young children was somewhat surprising. While tree pollen had substantially lower effects for young children than on school aged children, the effects of Cupressaceae were similar across age groups. One potential explanation is IgE independent pathways (Traidl-Hoffmann et al., 2009), which have been demonstrated for Juniperus ashei pollen and are linked to the generation of high levels of reactive oxygen species (Endo et al., 2011). Additionally, pollen can be a route or proxy for bacterial exposures (Oteros et al., 2019), which could offer another conceivable explanation for correlations between pollen and AREDV for young children. Another age-stratified study also found positive associations between tree pollen and AREDV for infants and preschool children, although neither result was statistically significant (Lee et al., 2019).

4.2. Viruses

Viruses were the largest source of attributable risk for young children (51.2%), school-aged children (57.0%), and adults (25.4%). Our estimates are within the range reported in the literature (Busse et al., 2010; Johnston et al., 2005; Murray et al., 2006; Tan et al., 2020) including in Texas (Eggo et al., 2016); a review of studies reported that for those with asthma exacerbations in North America, 44% tested positive for rhinovirus, 14.2% tested positive for coronavirus, 17.5% tested positive for influenza virus, and 10.6% tested positive for RSV (Zheng et al., 2018).

Differences in the proportion of positive tests for rhinovirus and RSV contributed at times to substantial spatiotemporal differences in our model of AREDV. Our results also show that seasonal covariation between airborne pollen and certain viruses could lead to incorrect attribution of the risk of pollen and implies that epidemiological analyses of AREDV should ideally include both types of variables. However, we did not detect meaningful interactions between pollen and viruses, in contrast to some others (Damialis et al., 2021) but in accordance with another similar analysis (Lee et al., 2019); considerable uncertainty about these interactions persists (Martikainen et al., 2023). Overall, the large number of virus attributed AREDV (~10,452 annually in our study population) represents a substantial opportunity for interventions including pharmaceutical approaches (Pike et al., 2018; Teach et al., 2015), vaccinations (Vasileiou et al., 2017), and non-pharmacological interventions (Khanolkar et al., 2022).

4.3. Study limitations

Exposure measurement error is a major issue for epidemiological studies of allergenic pollen because airborne pollen concentrations can vary considerably within even a single city (Katz & Batterman, 2020). Distance thresholds create a trade-off between exposure assignment error (e.g., due to differences in environmental variables and therefore the timing of pollen release over space (Katz et al., 2019) and sample size. Indeed, when our epidemiological analysis was restricted to populations within 10 km and 50 km of each pollen monitoring station (SI 8; SI 9), we noted reduced attributable risk due to pollen at longer cut-off distances (e.g., attributable risk due to tree pollen for school-aged children at 10 km was 3.4%, at 25 km was 2.8%, and at 50 km was 2.6%). To address this limitation, and to better account for differences even within cities (e.g., as seen for San Antonio) we advocate for the development of comprehensive spatially-oriented pollen models. Intra-municipal differences in airborne pollen are often closely associated with urban plant composition at local and neighborhood spatial scales (Charalampopoulos et al., 2018; Katz et al., 2019; Katz & Carey, 2014; Werchan et al., 2017), allowing for the creation of airborne pollen models within cities (Katz et al., 2023). We hope that future epidemiological analyses will pair more spatially granular airborne pollen and health outcomes data, which could better link plants to their pollen-mediated health effects and inform decision making about urban plants and green spaces (e.g., species selection choices).

There are limitations to the viral surveillance data used here, as the actual number of tests administered varied by time and location and did not include all potentially relevant viruses. The statistically significant effects of the low concentrations of other pollen on attributable risk for young children (−4.8% of cases [CI: −6.8 – −3.0]) and school aged children (−4.4% of cases [CI: −5.7 – −3.0]) has also been noted in other studies on ragweed pollen (De Roos et al., 2020; Gleason et al., 2014; Héguy et al., 2008) which peaks as the school year begins and viral prevalence spikes. This presumably non-causal association between ragweed and asthma-related emergency department visits may be due to incompletely accounting for viruses. We conducted further exploratory analyses focusing exclusively on the time period when this category of pollen is high (August – November), but this did not substantially change risk attributable to other pollen (data not shown). Additionally, viral testing may better reflect the prevalence of viruses in children than in adults due to a higher frequency of testing in children with respiratory virus symptoms (Tran et al., 2022). Finally, the school vacation term affected the estimates of viral and pollen effects (e.g., for school children, omitting the school vacation term from the model increased the risk attributable to Rhinovirus from 42.1% to 56.4%; SI 10).

Air pollutants can also contribute to AREDV (Fan et al., 2016; Orellano et al., 2017) but were not included in this study; however, the magnitude of their effects are generally low and they are unlikely to be confounded with upper-respiratory viruses or pollen (Goodman, Loftus, et al., 2017; Strickland et al., 2010), including in Texas (Goodman, Zu, et al., 2017). The one potential exception is a temporal correlation between grass pollen and ozone concentrations, but grass pollen was generally low in the cities included in this study.

4.4. Conclusions

This study represents one of the few multi-city epidemiological analyses of airborne pollen concentrations and a respiratory health related outcome. The results presented here demonstrate the public health importance of spatio-temporally variable environmental triggers of asthma including pollen and viruses and show how their effects vary substantially over geography due to differences in plant composition. This study also illustrates the need for understanding the dynamics of regionally important pollen types, such as Cupressaceae in central Texas. Together, these results highlight how environmental triggers of asthma vary over time and space, which could help inform the development and testing of public health and health care delivery interventions.

Highlights.

• There is substantial spatio-temporal variation in asthma-related emergency department visits

• Most risk is attributed to viruses (25 – 57%, depending on age group)

• In certain cities during peak season, up to 18% of visits are due to pollen

• Plant composition causes substantial variation among cities in risk due to pollen