Increased risk of emergency department presentations for bronchiolitis in infants exposed to air pollution

Air pollution has been linked to an increased risk of several respiratory diseases in children, especially respiratory tract infections. The present study aims to evaluate the association between pediatric emergency department (PED) presentations for bronchiolitis and air pollution. PED presentations due to bronchiolitis in children aged less than 1 year were retrospectively collected from 2007 to 2018 in Padova, Italy, together with daily environmental data. A conditional logistic regression based on a time‐stratified case‐crossover design was performed to evaluate the association between PED presentations and exposure to NO2, PM2.5, and PM10. Models were adjusted for temperature, relative humidity, atmospheric pressure, and public holidays. Delayed effects in time were evaluated using distributed lag non‐linear models. Odds ratio for lagged exposure from 0 to 14 days were obtained. Overall, 2251 children presented to the PED for bronchiolitis. Infants’ exposure to higher concentrations of PM10 and PM2.5 in the 5 days before the presentation to the PED increased the risk of accessing the PED by more than 10%, whereas high concentrations of NO2 between 2 and 12 days before the PED presentation were associated with an increased risk of up to 30%. The association between pollutants and infants who required hospitalization was even greater. A cumulative effect of NO2 among the 2 weeks preceding the presentation was also observed. In summary, PM and NO2 concentrations are associated with PED presentations and hospitalizations for bronchiolitis. Exposure of infants to air pollution could damage the respiratory tract mucosa, facilitating viral infections and exacerbating symptoms.


INTRODUCTION
Bronchiolitis is the first reason of hospitalization in children aged less than 1 year in Europe (Koehoorn et al., 2008). It consists of a lower respiratory tract viral infection, usually due to respiratory syncytial virus (RSV), which leads to expiratory wheezing and respiratory distress (American Academy of Pediatrics Subcommittee on Diagnosis & Management of Bronchiolitis, 2006;Meissner, 2016). The This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2022 The Authors. Risk Analysis published by Wiley Periodicals LLC on behalf of Society for Risk Analysis. smaller airways, the developing lungs, and immune system of children make them more vulnerable to the infection (Darrow et al., 2014). Furthermore, the development of bronchiolitis in infancy has been linked to an increased risk of asthma or lifetime wheezing later in life (Lee et al., 2018).
The time course of the disease illustrated by Karr and colleagues shows the beginning of symptoms 6-7 days after the infection and a peak in severity, compatible with the period of PED visits, on days 8-9 since infection (C. Karr et al., 2006). Moreover, the disease has a typical seasonality, with a highest number of cases between November and March in the northern hemisphere (Nenna et al., 2017).
Several studies have shown an association between air pollution and respiratory morbidity in children (Farhat et al., 2005;Garcia et al., 2021;Sram et al., 2013). Studies on the chronic and long-term exposure to air pollution showed that children living closer to the main roads have an increased risk of developing bronchiolitis (Girguis et al., 2017;C. J. Karr et al., 2009).
The short-term effect of air pollution on the risk of developing bronchiolitis has been mostly evaluated on single lags or on the moving average concentration of various pollutants during the days before the PED visit or the hospitalization. Hospitalization for bronchiolitis has been shown to be associated with the moving average concentration of PM10, PM2.5, and NO 2 in the week before (Carugno et al., 2018;Horne et al., 2018;Yitshak-Sade et al., 2017). However, lack of consistency has emerged for restricted time windows (Cheng et al., 2021;Ségala et al., 2008;Yitshak-Sade et al., 2017).
Studies on single lag reported on one side a nonsignificant association between hospitalization for bronchiolitis and PM2.5 on days 1 and 4 preceding the event (C. Karr et al., 2006), and on the other side a significant association between hospitalizations and the levels of PM10 in all the 11 days before the hospitalization (Carugno et al., 2018).
Recently, some researchers started to focus also on the cumulative effect of air pollution on children's lungs diseases. Leung and colleagues found a positive effect of NO 2 and a null overall effect of PM10 on bronchiolitis hospitalizations in the subtropical region of Hong Kong (Leung et al., 2021), whereas a positive 7-days cumulative effect was observed on lower respiratory obstructive diseases in children aged less than 4 years old (Schvartsman et al., 2017).
The present study aims to investigate the association between PED presentations for bronchiolitis and air pollution concentrations in the 2 weeks preceding the visit, with a particular focus on nonlinear and delayed effects in time.

Data collection
Data on children PED presentations from January 1st, 2007 to December 31st, 2018 in Padova, Italy, were retrospectively collected from the PED database. To more accurately identify all the children with a diagnosis of bronchiolitis, we reviewed the free text descriptive diagnosis field searching for the keyword root "bronchiol*" to maximize the sensitivity of our search. The PED of the University Hospital of Padova is a tertiary-care academic center, providing a 24/7 service to a catchment area of approximately 20 km, from the hospital. Data of children younger than 1 year of age, who received a diagnosis of bronchiolitis in the PED and lived within 20 km from the hospital, were abstracted from the database. Patient disposition data (discharge from the PED or hospitalization) were also retrieved.
Air quality and meteorological data were collected thanks to the Regional Agency for Environmental Prevention and Protection of the Veneto region (ARPAV). An urban background air quality monitoring station (located 5 km away from the hospital) ( Figure A1 in the Online Appendix) retrieves daily data on temperature, relative humidity, atmospheric pressure, and concentrations of PM10, PM2.5, , and NO 2 (Directive 2008/50/EC of the European Parliament & of the Council of 21 May 2008 on Ambient Air Quality & Cleaner Air for Europe OJ L 152, 11.6.2008).
The distribution of pollutants in this region has been evaluated by comparing levels of pollution of Padova with the ones of two other urban background monitoring stations, located in Treviso and Vicenza, at 46 and 31 km away from Padova hospital, respectively.

Statistical analysis
Continuous variables are represented as 1st quartile, median, and 3rd quartile. Categorical variables are shown as percentages (relative frequencies). The pairwise relations between NO 2 , PM10, and PM2.5 were assessed using the Spearman's rank correlation (ρ).
The association between air pollution concentration and PED presentations for bronchiolitis has been investigated using a multivariate conditional logistic regression based on a time-stratified case-crossover design (Levy et al., 2001;Maclure, 1991).
This approach allows to account for the possibility of confounding by seasonal patterns by selecting controls on the same day of the week of the same month of the same year of cases. For example, a case on the 8th of November 2017 will have the 1st, 15th, 22nd, and 29th of November 2017 as control days. In this way, every case will be matched with three or four controls. Therefore, subjects serve as their own control and known and unknown time invariant characteristics of all patients (e.g., parental smoking, wood burning habits that could produce indoor air pollution, preterm birth) are controlled by design.
The potential delayed effects in time of air pollution have been evaluated combining the conditional logistic regression with a distributed lag nonlinear model (DLNM) (Gasparrini, 2014). In this way, through a "cross-basis" function, it will be taken into account not only the potential nonlinear association between the exposure and the outcome, but also the nonlinear effect across lags.
Lags from 0 (the exposure on the day of PED presentation) to 14 (the exposure 14 days before the presentation) were considered for every pollutant to assess both the acute and the delayed effect of air pollution. The number of lags was chosen to evaluate the role of air pollution during the development of the disease and in the days preceding the infection.
Because of the strong correlation between the pollutants, three different models were performed, one for each pollutant. The pollutant entered into the model as a cross-basis matrix in which the exposure-response association has been linearly modeled, whereas the lag structure has been modeled with a natural cubic spline with 5 degrees of freedom.
Moreover, the potential strong and delayed effect of temperature on the development of bronchiolitis (Vandini et al., 2013) was taken into account by means of a cross-basis matrix for lags 0 to 14 of temperature. Natural cubic splines with 4 and 5 degrees of freedom were used to model the temperature-response association and the relationship between outcome and lag structure, respectively.
Bayesian information criterion (BIC) score was used to select the best cross-basis matrix.
The three multivariate models were adjusted for the abovementioned matrix of lagged temperature, public holidays, and the 3 days moving average values of atmospheric pressure and relative humidity.
A sensitivity analysis has been performed by temporally splitting the dataset into two parts (from 2007 to 2012 ad from 2013 to 2018). The analyses have been performed in both datasets to check the Bradford Hill criteria of strength (Hill, 1965).
Models were also applied only on children who were hospitalized, aiming at evaluating how the risk changes in a subcohort with more severe conditions.
Besides the daily effect of air pollution, the cumulative effect among the 2 weeks has also been assessed. Environmental missing data (2.1%) were imputed with univariate imputation using the robust nonlinear method implemented in the transcan function of Hmisc R package with default settings (Harrell Jr, 2019).
Results are shown as the odds ratio (OR) associated with a 10 unit increase in the selected pollutant.
All the analyses were performed using R statistical software (R Core Team, 2020) and dlnm (Gasparrini et al., 2019) package.

RESULTS
Overall, 42,543 infants in the first year of life presented to the PED between the 1st of January 2007 and the 31st of December 2018 and 2215 (5.2%) received a final diagnosis of bronchiolitis. Overall 62% of infants with bronchiolitis were male and the median age was 3.6 months (Table 1). Air pollution concentrations in the period of the admissions, mostly between November and March, are shown in Table 2. Levels were high, especially for PM2.5, which exceeded the World Health Organization (WHO) thresholds in 71% of days, whereas for PM10 and NO 2 exceeding occurred in 49% and 51% of days, respectively (World Health Organization, 2006).
As an example of the distribution of air pollution across the Veneto Region, Figure A2 shows the pollutants concentrations in Padova, Treviso, and Vicenza during the cold TA B L E 1 Presentations characteristics. Data are median [IQR] or n (%). PM2.5 and PM10 = particulate matter of less than 2.5 and less than 10 μm in aerodynamic diameter, respectively. NO 2 , nitrogen dioxide; WHO, World Health Organization As expected, PM10 and PM2.5 were the most strongly correlated pollutants (ρ = 0.95), whereas the correlation between NO 2 and particulate matter was slightly weaker (ρ = 0.70 with PM10 and ρ = 0.73 with PM2.5). A detailed description of correlations by year is presented in Supporting Information (Table A1).
The association between presentations for bronchiolitis and 10 g/m 3 increase in both PM2.5 and PM10 is statistically significant from lag 0 to lag 4 ( Figure 1).
The highest risk due to PM10 is on lag 0, lag 1, and lag 2 with an OR (95% CI) of 1.13 (1.03-1.24), 1.15 (1.07-1.23), and 1.13 (1.07-1.20), respectively. Moreover, from lag 2, the risk starts decreasing until it becomes nonsignificant from lag 5. The risk of PED presentation associated with PM2.5 shows a similar descending trend, in this case starting from the first lag and becoming nonsignificant from lag 5.
The risk associated with NO 2 starts to increase from lag 2, has a peek on lag 4 with an OR (95% CI) of 1.34 (1.21-1.49), and decreases progressively. The association is significant from lag 2 to lag 12 for a 10 μg/m 3 increase F I G U R E 1 Odds ratio (OR) and 95% confidence interval for the association between pediatric emergency department (PED) presentations for bronchiolitis and particulate matter of less than 10 μm (PM10) and less than 2.5 μm (PM2.5) in aerodynamic diameter resulted from the multivariate model. The OR was calculated for a 10 unit increase in PM10 and PM2.5 on lags from 0 to 14.

F I G U R E 2
Odds ratio (OR) and 95% confidence interval for the association between nitrogen dioxide (NO 2 ) and PED presentations for bronchiolitis resulted from the multivariate model. The OR was calculated for a 10 unit increase in NO 2 on lags from 0 to 14. of NO 2 (Figure 2). The OR for each lag is reported in the Supporting Information (Table A2) The three-dimensional plot in Figure 3A-B shows the joint relationship between OR, lags, and pollutant concentrations. The exposure to higher concentrations of all the pollutants is associated with a higher risk of presenting to PED for bronchiolitis and the trend along lags remains constant.
The sensitivity analysis performed in the two different time periods shows that for both PM10 and PM2.5, the trend and the estimated risks remain constant with higher OR observed in the 4 days before the presentation. Since the sample size halved, wider CI are observed ( Figure A3a-d in the Online Appendix).
The impact of NO 2 on both cohorts shows a similar trend with a peak on lag 4 ( Figure A3e-f in the Online Appendix).
Analysis on hospitalized children (707 infants) confirmed the results obtained on the whole cohort for all pollutants. Nevertheless, wider confidence intervals were observed because of the lower sample size. The OR of the models shows that the risk of hospitalization associated with F I G U R E 3 (A) 3D graph of PM10 effect. (B) 3D graph of NO 2 effect. Odds ratio (OR) of presentations by daily levels of particulate matter of less than 10 μm in aerodynamic diameter (PM10) and nitrogen dioxide (NO 2 ) air pollution is higher than the risk associated with PED presentation (Figure 4).
The overall cumulative effect of air pollution infants has been exposed to in the 2 weeks before the PED presentation shows a positive effect of NO 2 , whereas PM10 and PM2.5 do not seem to significantly affect the risk of being admitted ( Table A2 in the Online Appendix). A 10-unit increase in the overall exposure to NO 2 is associated with an OR of 1.47 (95%CI: 1.24-1.74) of being admitted to PED.

DISCUSSION
The present study found an association of PM10 and PM2.5 concentrations with PED presentations for bronchiolitis in the 4 days before the presentation. According to what was published by Karr and colleagues, this period mostly corresponds to the first appearance of symptoms, thus suggesting PM to play a role in the onset of symptoms (C. Karr et al., 2006).
On the other hand, NO 2 is associated with presentations for bronchiolitis in a range of days consistent with the incubation period before symptoms onset. This finding suggests that NO 2 could impair the functional integrity of the respiratory mucosa, facilitating infection from viral agents and progression of the disease. Moreover, NO 2 also demonstrated a cumulative effect over the 2 weeks preceding presentation to the ED.
Evidence of the effects of air pollution on the immune system has been recently summarized by Glencross and colleagues (Glencross et al., 2020), whereas the NO 2 role in the incubation period of the infection was studied by Becker and Soukup (1999). The exposure of RSV-infected bronchial epithelial cells to different concentrations of NO 2 showed a decrease in interleukin (IL)−6 and IL-9 production (Becker & Soukup, 1999). On the same line, increased exposure to NO 2 was associated with a deficient production of the type I interferon-(IFN-β) from bronchial epithelial cells exposed to other respiratory viruses, such as RV (Bonato et al., 2021). Activation of the interferon and inflammatory pathways in the airway epithelium represents the first line of defense to prevent the spread of the virus; thus, NO 2 by impairing innate responses at the epithelial surface may facilitate the progression of the infection.
Not only NO 2 may affect the innate immune response to viral infections, but it can also enhance the epithelial expression of the intracellular adhesion molecule-1(ICAM-1) (Ayyagari et al., 2007), a protein required for RSV adhesion to epithelial cells (Behera et al., 2001). The observation of an effect of cumulative exposure to NO 2 in the few days before the infection and during the period of infection could be related to sustained upregulation of ICAM-1 expression, which facilitates entry of RSV into epithelial cells and its spread through the respiratory tract.
The exposure of human bronchial epithelial cells to PM2.5 has been demonstrated to cause an increase in production of Reactive Oxygen Species (ROS) and cell apoptosis (Zhu et al., 2018). This damage to the epithelial cells of the airway mucosa could lead to a higher symptoms severity in case of concurrent viral infection. The risk of hospitalization in our sample was approximately 20% higher than the risk of presenting to the PED based on PM exposure. This was not true for NO 2 , thus supporting the hypothesis that NO 2 plays a role in increasing the probability of infection, but not in increasing symptoms severity, which is instead more closely related to PM exposure.
Other studies focused on the association between air pollution and bronchiolitis. An association was found between hospitalization for bronchiolitis and the moving average concentration of NO 2 , PM10, and PM2.5 one week before the hospitalization in the south of Israel (Yitshak-Sade et al., 2017). A similar result was found on the 5 days moving average concentration of NO 2 and PM10 in Paris (Ségala et al., 2008). Moreover, Carugno and colleagues evaluated the weekly lags of PM10 and found an association with hospitalization for bronchiolitis in the 2 weeks before (Carugno et al., 2018). In the same study, single lags were also F I G U R E 4 Odds ratio (OR) and 95% confidence interval for the association between air pollutants and hospitalizations for bronchiolitis resulted from the multivariate model. The OR was calculated for a 10 unit increase in particulate matter of less than 10 μm (PM10) and less than 2.5 μm (PM2.5) in aerodynamic diameter and nitrogen dioxide (NO 2 ) on lags from 0 to 14.
analyzed and results showed an association in the first nine lags. Their findings are partially different from the results of the present study. One possible reason is that Carugno et al. did not model lag in a constrained framework, thus resulting in a possible collinearity in the model (Bhaskaran et al., 2013).
A correlation between the number of RSV-positive patients who accessed the hospital and the mean PM10 concentration in the preceding week was also measured by Vandini and colleagues (Vandini et al., 2013). On the other side, the study published by Karr and colleagues did not find an association between PM2.5 and hospitalization for bronchiolitis. However, the data retrieved by the monitoring stations in the study by Karr et al. were recorded every 3 days (C. Karr et al., 2006). Focusing on the models proposed in the present study, different lags have been shown to variably impact on the risk of PED presentation; thus, the effect of lag 1 could be quite different from the effect of lag 3 on the risk of hospitalization.
Another research conducted in the subtropical region of Hong Kong found a positive overall effect of NO 2 on hospitalization for bronchiolitis, whereas for PM10, they only observed an immediate but not long lasting effect, similarly to what was found in the present study (Leung et al., 2021).
Most of the published cited studies focused on the moving average concentration of pollutants in the days preceding the hospitalizations. Nevertheless, the moving average concentration approach does not allow one to estimate the single lag effect and it only focuses on the average atmospheric conditions the patients were exposed to.
Otherwise, DLNM can estimate the potential effect of air pollution on all days of the disease progression and can reveal the time at which air pollution can have a higher impact on favoring the onset or worsening the symptoms of bronchiolitis in infants.

Limitations
Linking individuals to the nearest monitoring station may arise some issues on the accuracy of the measured exposure. However, the uniform distribution of air pollution in the Veneto region partially reduces potential inconsistency of measured exposure. The geological characteristics, including a flat land with mountains on three sides and a shallow sea on the fourth with an average wind speed of 4 km/h in a year, favors stagnation of air pollutants in the study area. For this reason, concentrations provided by the monitoring station are representative of a quite wide area.
The correlation between the pollutants did not allow to perform a model with both NO 2 and PM simultaneously, so it was not possible to disentangle the effect of one single pollutant (adjusting for the other). As so, uncertainty remains as to whether there is confounding by correlated pollutants or NO 2 itself represents a pollution mixture (e.g., from traffic vehicles).

CONCLUSIONS
Results shown in this study underline the great impact air pollution has on the respiratory system of infants. The higher the concentration of air pollutants they are exposed to, the higher the risk of presenting to the PED for bronchiolitis. Moreover, the novel approach used shows new insights: different air pollutants have a different time-related impact during the development of the disease. This could also be very interesting to investigate at a pathophysiological level and useful for implementing prevention strategies.

A C K N O W L E D G M E N T
The authors would like to acknowledge the University Hospital of Padova and the ARPAV for the resources provided, and the University of Padova for funding the Ph.D. position of Elisa Gallo to work on this subject. Open Access Funding provided by Universita degli Studi di Padova within the CRUI-CARE Agreement.

D I S C L O S U R E
The present research has been presented at the 2019 SRA annual meeting.