Association between wildfire smoke exposure and Seattle, Washington Pediatric Hospital services, 2006–2020

Pacific Northwest wildfire smoke events have been increasing in prevalence and severity over the past three decades, resulting in documented negative health outcomes in adults. However, there is less evidence demonstrating the effect of wildfire smoke in pediatric populations. To evaluate the association between wildfire smoke exposure and healthcare utilization in a pediatric tertiary medical center in Seattle, WA. We utilized a case–crossover study to determine the odds of pediatric emergency department (ED) visit/ hospital admission at Seattle Children’s Hospital on wildfire smoke days versus non-wildfire smoke days during wildfire season (June to September), 2006–2020. The health outcomes dataset reports hospital encounters in two categories: ED visits or admissions that are for inpatient or observational purposes. The health outcomes dataset reports hospital encounters in two categories: ED visits or admissions that are for inpatient or observational purposes. The reported encounter types are mutually exclusive. We stratified analyses by individual-level characteristics and examined associations for lagged exposures 0–7 d prior to admission. In adjusted analyses, smoke exposure was associated with a 7.0% (95% CI: 3.0%–12.0%) increase in odds of all-cause hospital admissions and a 0.0% (95% CI: −3.0%, 3.0%) change in odds of all-cause ED visits. We also observed increases in the odds of all-cause hospital admissions ranging from 4.0% to 8.0%, for lagged exposure on days 1–7. When stratified by health outcomes, we found a 9.0% (95% CI: 1.0%–17.0%) and an 11.0% (95% CI:1.0%–21.0%) increase in the odds of ED visits for respiratory and respiratory infection-related concerns, respectively. Our results demonstrate associations between wildfire smoke and negative health effects in children. Similar to other studies, we found that wildfire smoke exposure was associated with an increase in respiratory-related ED visits and all-cause hospital admissions in a pediatric population. These results will help inform patient education and motivate interventions to reduce pediatric morbidity during wildfire season.


Introduction
In the Pacific Northwest, climatic changes in patterns of extreme heat, precipitation, and wind have led to an increase in the frequency, duration, and severity of wildfire smoke occurrence over the past 50 years (Abatzoglou and Williams 2016, Turco et al 2018, Xu et al 2020. Wildfires consumed an estimated 4.2 million hectares of area in the western U.S. during the years of 1984-2015 (Abatzoglou and Williams 2016). Paired with factors such as forest management practices, deforestation, and increases in the interface between the human residence and forest, wildfires and their resulting smoke is a growing concern for the western U.S. communities (Adetona et al 2016). These more severe and longer-lasting fires expose people to higher levels of smoke than historically experienced (Balmes 2018). Wildfire smoke events significantly increase levels of pollutants, such as particular matter (PM), both locally and regionally, often exceeding regulatory limits (Liu et al 2015). Emissions from wildfires have a negative effect on annual measures of air quality, erasing much of the improvements in air quality during wildfire seasons that have been made over the last several decades (Balmes 2018).
Wildfire smoke is comprised of a heterogeneous mixture of gaseous, liquid, and solid particles, with the toxic composition varying dependent on the fuel being burned (Evans et al 1977, Naeher et al 2007, Finlay et al 2012, Balmes 2018. The primary pollutant of concern within wildfire smoke is particulate matter with an aerodynamic diameter less than 2.5 µms (PM 2.5 ). A study observing smoke events in the Western U.S. from 2004 to 2009 noted that over 70% of total PM 2.5 emissions on days exceeding the 35 µg m −3 (24 hour average) regulatory standard can be attributed to wildfires (Liu et al 2016). Due to its small size, PM 2.5 can penetrate deep into the lungs and the smallest particles may enter systemic circulation (Reid andMaestas 2019, Aguilera et al 2021). Children and adolescents comprise some of the most vulnerable populations to PM 2.5 and the neurological impacts of air pollution as their brains are still in development (Balmes 2018, Roberts et al 2018. For children living in certain parts of the U.S., up to 20% of the PM they are exposed to comes from wildfires (Holm et al 2021). Their vulnerability can be attributed to increased exposure to PM from lifestyle factors (often spending more time outdoors in comparison to adults), inhaling more air relative to their body size, higher proportions of particles penetrating deeply into their lungs (due to less nasal deposition), and the ongoing developing of their respiratory and immune systems (Bennett et al 2007, Vanos 2015, Holm et al 2021. Around the U.S., there are approximately 7.4 million children annually impacted by wildfire smoke (Rappold et al 2017). Health impacts from wildfire exposure can vary in severity, ranging from mild exacerbation of underlying health conditions (Black et al 2017) to an increase in emergency department (ED) and hospitalization service utilization (Reid and Maestas 2019) to increased mortality (Doubleday et al 2020). There is documented evidence of the association between PM 2.5 and adverse respiratory outcomes in children, particularly asthma exacerbations (Leibel et al 2019, Oliveira et al 2019, Aguilera et al 2021. Similar to the findings in adults, children that are closer (or directly downwind) from the fire are at highest risk, and their respiratory symptoms can manifest as upper respiratory irritations in the eyes, nose, and throat, increased visits to the hospital, and increases in infections (such as pneumonia) (Leibel et al 2019, Holm et al 2021 Children from 0-5 years old are likely to be the most at risk as seen by their age group having the highest incidence of hospital admissions during smoke events (Leibel et al 2019).
Though there are recent studies regarding wildfire smoke that have focused on Washington-state specific outcomes (Gan et al 2017, DeFlorio-Barker et al 2019 and some that have specifically examined pediatric populations (primarily in California) (Leibel et al 2019, Aguilera et al 2021, additional research is needed to characterize the risk faced by children in Washington. When a Washington-based wildfire smoke study by Gan et al (2017) stratified by patients younger than 15, they found a 7% increased association between smoke and all respiratory-related hospital admission. We know children in this state are at risk, and these effects can be realized decades into the future. Studies indicate that increased exposure to PM results in poor respiratory outcomes in children that can contribute to adverse, lifelong chronic conditions such as decreased lung function (Black et al 2017, Kim et al 2017, increased likelihood of mental illness in adulthood (McFarlane andVan Hooff 2009, Brown et al 2019), and a decrease in academic performance (Gibbs et al 2019). However, when exposure to pollutants is decreased, growth in lung function improves Marty 2010, Gauderman et al 2015). These findings suggest that characterizing the risk between pediatric health and wildfire smoke events and developing interventions to decrease PM exposure could protect children during smoke events and maintain respiratory health into adulthood. Based on prior work, we hypothesized that wildfire smoke would be associated with increased ED visits and hospitalizations among children.

Health outcome data
The health outcomes in this study were sourced from the Pediatric Health Information System (PHIS) dataset (www.childrenshospitals.org/phis). The PHIS dataset is comprised of deidentified patient data from 51 children's hospitals in the U.S and includes data on encounter diagnoses, length of stay, demographic information, and billing (www.childrenshospitals.org/phis). For the purpose of this study, however, only hospital encounters from Seattle Children's Hospital were included. Our study period was confined to 1st June 2006-30th September 2020. The utilized variables included date of the medical visit and patient diagnosis (reported as an All Patient Refined Diagnostic Related Group (APR-DRG)). We utilized the broad APR-DRG grouping strategy because it allowed us to easily compare cases before and after the switch from International Classification of Diseases-9 (ICD-9) to ICD-10, and it increased our study power by allowing more cases in our groupings that may be influenced by PM 2.5 -related effects. Individual-level variables included age, sex, race (either assigned by hospital staff or reported by the patient/patient guardian during the clinical encounter), and insurance type. Children less than or equal to 19 years of age and with a Washington state residential zip code were included in the study. We excluded any children who came to the hospital due to COVID-19 because at the time of methods development, the impact of wildfire smoke on COVID-19 was uncertain and the health records did not specify whether access to COVID-19 testing was the primary purpose for the healthcare visit. We also excluded individuals admitted on the fourth or fifth of July due to confounding levels of air pollution resulting from fireworks. The University of Washington's institutional review board (IRB) reviewed this study and determined it exempt from further review on June 7, 2021. The Seattle Children's IRB board approved this study on 3 June 2021.
This study examined both ED visits and hospital admissions (which includes inpatient admissions and observational stays). These types of hospital services are mutually exclusive-a patient could not fall into both categories. For each service type, we stratified the data by age categories (0-5, 6-12, 13-19) derived from prior studies (Aguilera et al 2021), sex (male or female), race group as reported in PHIS ('Non-Hispanic White' , 'Non-Hispanic Black' , 'Hispanic' , 'Asian' , 'Other'), insurance type as reported in PHIS ('Government' , 'Private' , or 'Other'), and cause of clinical encounter (respiratory (not infectious), respiratory infections, dermal conditions, trauma, and mental health). We defined the cause categories a priori based on established literature regarding the impacts of wildfire smoke on all aged populations and from physiologically plausible links suggested by practicing pediatricians and pulmonologists. The full-list of APR-DRG conditions included in each clinical encounter cause category can be found in the supplemental information (table S1).

Exposure classification
We assigned exposure to pollution, temperature and humidity based on residential zip code. Using a method described in Doubleday et al (2020), we assessed exposure for cases using an exposure grid that contains 24 h average PM 2.5 and daily average maximum Humidex (a measurement of heat) values by zip code. We then overlaid the exposure grid with population data to derive a grid with population-weighted zip code-level daily average PM 2.5 . For grid cells missing exposure data, we utilized the nearest grid cell values as a proxy for exposure (Doubleday et al 2020). We joined the exposure dataset with the PHIS dataset using R 4.0.2 (R Core Development Team 2020) and assigned each clinical encounter a corresponding 24 h average PM 2.5 concentration and average maximum humidex value.
This study uses a binary 'smoke day' classification based on meteorological threshold conditions from Doubleday et al (2020). Briefly, a wildfire smoke day is defined as a day 'with a 24 h average PM 2.5 concentration greater than 20.4 µg m −3 ' per zip code. Previous work recognized that in some areas of our state, wildfires contribute to lower 24 h average PM 2.5 concentrations; therefore, an additional set of criteria are applied for the smoke day classifications. The criteria are as follows: 'For the [zip code] days between 9 and 20.4 µg m −3 : 1. The day must be part of an event in which two of three consecutive days are greater than 9 µg m −3 2. One of the days in the event window must be greater than 15 µg m −3 3. For urban areas (Seattle, Tacoma, Spokane), at least 50% of the air monitors in those areas must be greater than 9 µg m −3 ' (Doubleday et al 2020).
This classification was chosen because this study evaluates patient residential exposure across the state, including major urban areas (such as Seattle, Tacoma, and Spokane) as well as rural areas. Therefore, this state-wide definition best minimized false positives in urban areas with high background levels of PM as well as false negatives in rural areas with lower levels of background PM.

Statistical approach
This study utilized a time-stratified case-crossover design to determine whether wildfire smoke exposure is associated with increased risk of pediatric hospital encounters. In the case-crossover design, each case serves as their own control by comparing the 'at risk' exposure period with the level of exposure during a referent period (i.e. a period during which the patient was exposed but did not result in a hospital visit) (Levy et al 2001). These 'referent windows' create within-subject comparisons that control for time-invariant confounders (Janes et al 2005). To adequately control for time-dependent confounders (long-term time trends, seasonality, and day-of-week impacts), we matched the referent days by the same day of the week as the 'case' day within the same month and year (Levy et al 2001, Janes et al 2005, Hutchinson et al 2018. This matching method yields multiple referent days within the same month. We removed any referent days that overlapped as a case day if a patient was admitted multiple times within the same month and year. We also removed any referent days on the fourth or fifth of July due to possible air pollution resulting from fireworks. We adjusted for humidex using a natural cubic spline with three degrees of freedom (Doubleday et al 2020); all other confounders were controlled for by design. We used conditional logistic regression adjusting for humidex to estimate the odds of pediatric hospital encounter on wildfire smoke vs non-smoke days. To further characterize the association between wildfire smoke and pediatric health, we stratified the results by age group, sex, race/ethnicity category, insurance type, and cause of hospital encounter. We examined single day lagged effects for lag days 0-7. For same-day exposures, we reported our findings by age group and sex within the cause of admission categories (respiratory, respiratory infections, dermal conditions, trauma, and mental health).
All statistical analyses were performed in R (R Core Development Team 2020). We define statistical significance as any result that has a p-value of less than or equal to 0.05.

Population
Between June and September 2006-2020, there were a total of 194 887 pediatric hospital encounters at Seattle Children's Hospital. Among these, 132 408 (67.94%) were ED visits, and 62 479 (32.06%) were inpatient or observational admissions to the hospital (see table 1 for patient demographics). About 56.9% and 46.4% of patients were in the youngest age category (0-5) for ED and hospital admissions, respectively. The most common race reported in our data was Non-Hispanic White, followed by Other and Hispanic. The most common insurance type for ED visits was government (49.3%) and for inpatients/observational admissions, private (48.8%). Trauma was the most common reason for ED visits (19.2%) while respiratory-related conditions were the most frequently noted cause (10.1%) for hospital admissions (table 1). Most patients originated from King, Snohomish and Pierce Counties (figures S1 and S2).

Exposure
For ED visits, the average PM 2.5 concentration on both case and referent days was 6.67 µg m −3 (figure 1, table S2). For hospital admissions, the 24 h average PM 2.5 concentration on case days was 6.73 µg m −3 while the 24 h average PM 2.5 concentration on referent days was 6.58 µg m −3 ( figure 1, table S2). There were 32 950 zip-code days that met our criteria for 'wildfire smoke'; averaging 29.78 µg m −3 (SD: 30.58) and 31.30 µg m −3 (SD: 35.89) PM 2.5 for ED visits and hospital admissions, respectively.

All-cause and lag analysis
In adjusted analysis, we found a 0.0% (95% CI: −3.0%-3.0%) change in odds for all-cause same-day ED visits (table 2) and a 7.0% (95% CI 3.0%-12.0%) increase in odds for all-cause hospital admissions on wildfire smoke days versus non-wildfire smoke days. When modeling lagged exposure 1-7 d before a case encounter, we did not observe statistically significant changes in odds for ED visits (figure 2). However, when evaluating the hospital admissions, we observed statistically significant increases in odds across all lag days (figure 2 and table S4). The strongest association observed (8.0% (95% CI: 4.0%-12.0%)) was for wildfire smoke exposure one day prior to hospital admission. The statistically significant increase in odds continued through lag day 7 (figure 2 and table S4).

Stratification by cause
This study looked at respiratory, respiratory infections, dermal conditions, trauma, and mental health causes of ED visits and hospital admissions (table 2). Most notably, we observed a 9.0% (95% CI 1.0%-17.0%) and 11.0% (95% CI 1.0%-21.0%) increased odds of an ED visit for all respiratory-related outcomes and respiratory infections, respectively. Smoke exposure was also associated with a 44.0% (95% CI: 3.0%-102.0%) increase in odds of trauma-related hospitalization. We conducted a sensitivity analysis removing all encounters in the year 2020 (table S5), and we found no significant changes in the ORs when stratified by cause.

Discussion
This case-crossover study of close to 200 000 hospital encounters from Seattle Children's Hospital found that wildfire smoke exposure was associated with an increased odds of respiratory-related ED visits and all-cause hospital admissions. Although wildfire smoke exposure was not associated with all-cause pediatric ED visits, in analyses stratified by health outcome, we found a 9.0% (95% CI: 1.0%-17.0%) increase in the odds of presenting to the ED for a respiratory-related concern on wildfire smoke days versus non-wildfire smoke days. When focusing on respiratory infections, we found an 11.0% (95% CI: 1.0%-21.0%) increase in odds under the same conditions. These findings expand on other studies in both pediatric and adult populations, which report associations between wildfire smoke and adverse respiratory health outcomes (Liu et  Our results also align with a study examining the health impacts of the 2012 wildfire smoke season in Washington state which also reported an 8.0% (95% CI: 2.0%-14.0%) increase in risk for all-age asthma-related hospital admissions for a 10 µg m −3 increase in wildfire smoke (Gan et al 2017). When looking at the pediatric literature, our results also mirror the general trends in findings where younger ages (especially ages 0-5) report an increase in odds of admissions (Hutchinson et al 2018, Leibel et al 2019, Aguilera et al 2021. With all-cause inpatient/observation hospital admissions, we observed a 7.0% (95% CI: 3.0%-12.0%) increase in odds for same day admissions and 5.0%-8.0% increase in odds for lagged exposure on days 1-7. The increase in odds of hospital admission during lag days was also echoed in other studies (Johnston et al 2014, DeFlorio-Barker et al 2019, and Washington-specific studies reported similar lag findings for all-cause respiratory admissions (Gan et al 2017) and mortality (Doubleday et al 2020).
Our stratified results need additional study as there were significant confidence interval overlaps between most groups, requiring caution for any interpretations or conclusions (table 2). However, some findings pose interesting areas of future study. When stratifying by cause of hospital encounter, dermal-related cases (defined in table S1) had a 10.0% (95% CI: −5.0%-27.0%) increase in ED visit odds, and they made up 4.1% of all ED cases. Hospital admission odds were much higher at a 61.0% (95% CI: −12. 0%-194.0%) increase, and they made up only 0.6% of inpatient/observational cases. Despite the wide confidence intervals in both cases, there is evidence in the literature that frequent and chronic exposure to high levels of PM can have negative effects on skin cells, manifesting in mild to severe symptoms (Okada et al 2013, Oudin et al 2016, Bonamonte et al 2019, Roberts 2021. When looking specifically at conditions affecting pediatric populations, atopic dermatitis (which often starts during infancy or childhood) can be further irritated by higher levels of pollutants such as volatile organic compounds and PM (Bonamonte et al 2019, Kim et al 2019, which are common components of wildfire smoke. For our study, any interpretations about dermal conditions must be made cautiously, but further study with a more granular focus on dermal conditions could further elucidate the relation between wildfire smoke and dermal health. When stratifying by trauma-related cases, we found a −1.0% (95% CI: −8.0%-5.0%) decrease in odds of an ED visit. Though trauma is broadly defined (table S1), Gan et al (2017) also found a null association between broken arm admissions and wildfire smoke exposure. However, there was a 44.0% (95% CI: 3.0%-102.0%) increase in odds of a trauma-related hospital admission, but the smaller sample size (only 1.6% of all hospital admissions) and wide confidence interval invite further study into the association between trauma and smoke exposure.
When stratifying by race/ethnicity, we found a 30.0% (95% CI: 11.0%-53.0%) increase in odds of hospital admissions for Non-Hispanic Black patients, the largest increase observed across any of our racial/ethnic strata. Despite the wide confidence interval, this finding leads to a critical, complex discussion on the relationship between air pollution, race/ethnicity, and health. Though the literature specifically related to wildfire smoke and racial/ethnic disparity is currently limited, established research shows us that there is a link between race/ethnicity, socioeconomic status (SES), and air pollution. Air pollution and income are significant correlates of mortality (Finkelstein et al 2003), and lower SES communities in the United States, often comprised primarily of Black and other minority populations (such as Hispanic), have a higher likelihood of poor air quality along with disease prevalence (Servadio et al 2019). This trend between lower SES, increased air pollution, and higher prevalence of negative health outcomes is echoed across most of North America, Asia, New Zealand, and Africa (Hajat et al 2015). However, other studies support that institutional racism also accounts for the health disparities observed in minority groups, especially for Black populations. Historical practices in Washington state and across the United States, such as redlining, are linked to poorer quality urban housing, and these affected areas can report higher rates of adverse health outcomes like asthma-related ED visits (Nardone et al 2020, Bose et al 2022, Lane et al 2022. Urban development trends can perpetuate already present inequalities by selectively placing pollutive and industrial infrastructure in low-income, often minority-occupied neighborhoods (Bolin et al 2005). A study conducted in Phoenix examining the relationship between some criteria air pollutants and sociodemographic variables found that race/ethnicity, independent of SES, served a predictor of increased pollutants for their Native American and Latino residents (Grineski et al 2007).
Looking specifically at wildfire events, Méndez (2022) captured how extreme events tend to exacerbate already present inequalities, such as a lower accessibility to resources. Since Latin and indigenous migrant communities face disproportionate exposure to poor air quality, they can develop underlying chronic respiratory conditions that increase susceptibility during smoke events (Méndez 2022). A similar mechanism may apply to Black and other minority communities in low SES neighborhoods that are more likely to be exposed to poor ambient air quality, which could result in the increased association we observed for adverse health outcomes during wildfire smoke events. However, the wide confidence interval in our study means that any theories must be rigorously examined before drawing any interpretations. Further research between the nexus of race/ethnicity, SES, pre-existing health conditions, and wildfire smoke could shed vital insight on this complex, crucial topic.

Limitations and future study recommendations
The strengths of this study include the datasets as well as the study design. Our large health outcome dataset (194 887 encounters) contains accurately captured health data with corresponding demographic information, increasing the power of the study to make significant conclusions. The air pollution dataset, created with Washington's Department of Ecology, also grants accurate exposure data down to the zip code level. Finally, the case crossover study allows us to use individuals as their own control, and we were able to use the accurate exposure dataset to adjust for the confounding impacts of temperature and humidity, yielding more accurate association estimates. Overall, this study has allowed us to further elucidate the relationship between wildfire smoke and vulnerable pediatric populations.
However, when assessing exposure, this analysis was limited by the assumption that individual exposure for the entire zip code is represented by the recorded PM 2.5 level at the nearest monitor, resulting in potential exposure misclassification. This concern is particularly applicable for rural communities in which there are fewer monitors per geographic area (Yao et al 2016). Additionally, the classification of wildfire smoke day (using the 20.4 µg m −3 threshold) can be challenging due to the lack of available methodology to distinguish between anthropogenic and wildfire related PM 2.5 . However, the tailored methodology reduced misclassification from anthropogenic PM 2.5 sources such as ambient PM pollution or event-based from fireworks.
This study was conducted with electronic medical records from a single, tertiary care pediatric hospital, so findings may not be generalizable to the underlying population. In particular, the study hospital has a referral base with more complex medical conditions compared to other pediatric hospitals, and it is uncertain whether this could have influences effect estimates.
There are many directions future research can take to address these limitations and further explore the associations relayed in this study. Regarding exposure, it could be modeled as a continuous, non-binary variable, and additional lag analyses could be explored, such as a distributed lag model or observing lag impacts beyond 7 days. Second, as mentioned previously, underlying conditions within the patient population can affect associations between pollutants admitted from wildfires and adverse health outcomes (Bernstein et al 2022, Méndez 2022. Conducting an analysis that examines the interaction between certain pre-existing conditions (such as diabetes) and wildfire smoke exposure would shed additional insight on vulnerability in sub-populations.

Conclusions
To our knowledge, this case-crossover study is the first to examine the associations between pediatric hospital admissions and wildfire smoke exposure in Washington state. Our findings confirm the increased odds of respiratory-related ED visits and all-cause hospital admissions during wildfire smoke events, and we have documented how lagged exposure to wildfire smoke may elevate the odds of inpatient and observational hospital admission from days 0-7. This work also explores how individual-level characteristics, such as age and race/ethnicity, and specific health outcomes may be associated with smoke events. The results can be used to inform pediatric practitioners, public health officials, and community members about the impacts that patients and their families can expect during smoke seasons. As wildfires continue to intensify, we hope these findings will encourage further study about the interaction between underlying health impacts, demographic factors, and wildfire smoke.

Data availability statement
The data cannot be made publicly available upon publication because they contain sensitive personal information. The data that support the findings of this study are available upon reasonable request from the authors.