Authors: Jelte Kelchtermans (1Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA; 2The Center of Applied Genomics, The Children’s Hospital of Philadelphia, Philadelphia, PA, USA; 3Division of Pulmonary Medicine, The Children’s Hospital of Philadelphia, Philadelphia, PA, USA), Frank Mentch (2The Center of Applied Genomics, The Children’s Hospital of Philadelphia, Philadelphia, PA, USA), Hakon Hakonarson (1Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA; 2The Center of Applied Genomics, The Children’s Hospital of Philadelphia, Philadelphia, PA, USA; 3Division of Pulmonary Medicine, The Children’s Hospital of Philadelphia, Philadelphia, PA, USA)
Categories: Article, Asthma, Pediatric, Ambient air pollution, Sensitivity
Source: Journal of exposure science & environmental epidemiology
Authors: Jelte Kelchtermans, Frank Mentch, Hakon Hakonarson
Ambient air pollution exposure increases the incidence and severity of pediatric asthma. Despite this, we lack effective therapies to protect patients from the impact of ambient air pollution exposure. A roadblock is the inability to identify patients that are affected by air pollution.
To examine the association between AAP sensitivity determined by individual exposure prior to asthma exacerbations and the severity of asthma in pediatric patients
We assess the association between spikes in ambient air pollution and asthma exacerbations. Patients were considered sensitive to a specific pollutant if they experienced an asthma exacerbation immediately following a spike in the concentration of that pollutant. Cut off values for these spikes were determined as two standard deviations above the mean concentration two weeks prior and two weeks post the days leading up to an asthma exacerbation.
We included 8,129 pediatric patients with over 34,346 associated asthma exacerbations. In a multinomial log-linear logistic regression model comparing patients with mild asthma to patients with moderate or severe asthma, sensitivity to Ozone, SO2, PM2.5 and PM10 was significantly associated to severe as opposed to mild asthma (OR 1.39 with CI 1.08-1.78, 1.58 with CI 1.12-2.23, 1.37 with CI 1.07-1.76, and 1.63 with CI 1.12-2.37 respectively). Furthermore, moderate as opposed to mild asthma was significantly associated with sensitivity to SO2 and PM2.5 (OR 1.24 with CI 1.06-1.44 and 1.26 with CI 1.12-1.43, respectively).
Our results show the impact of ambient air pollution sensitivity on asthma severity. To better understand air pollution sensitivity, follow up studies assessing biomarkers of exposure in patient sub-groups and comparing the genetic profile of patients sensitive to air pollution with non-sensitive patients are needed. We conclude that reliable assessments of air pollution sensitivity may lead to more successful mitigation of the effects of ambient air pollution.
There is a well-established link between ambient air pollution (AAP) exposure and asthma. (1, 2) Specifically, in pediatric asthma, up to 15% of exacerbations can be linked to AAP. (3) Furthermore, we now have longitudinal data showing that decreases in ambient air pollution are associated with decreases in asthma exacerbations and asthma incidence. (4, 5) Despite this, AAP concentration continues to exceed recommended standards. This excess exposure has been shown to lead to significant excess mortality and morbidity. (6) We currently have no effective therapies to protect pediatric patients with asthma from the impact of exposure to these high levels of AAP. A factor complicating the development of AAP specific asthma treatment strategies is that not all these patients are equally exposed or affected by AAP. Not only is there significant geographical and temporal variation in AAP, as race, ancestry, and familial atopic disease status are all associated with variability in the observed effect of AAP exposure. (7-9) In order to understand how our current therapies interact with AAP exposure and trial new interventions, we first need a way to identify and track both AAP exposure and sensitivity.
The data stored in a patients’ Electronic Medical Record (EMR) allows us to do just that. The EMR provides information regarding the patients’ address, asthma severity and date of asthma exacerbation. Patients that are sensitive and exposed to AAP would be expected to have a spike in AAP exposure just preceding an asthma exacerbation. To assess the AAP exposure preceding exacerbation we can use the United States Environmental Protection Agency Air Quality System (AQS). (10) This system integrates AAP data from monitors across the United States. Their Application Programming Interface (API) allows researchers to obtain information covering highly specific time periods and geographical areas. In this study, we examined the association between AAP sensitivity determined by individual exposure prior to asthma exacerbations and the severity of asthma in pediatric patients
Subjects were drawn from The Children's Hospital of Philadelphia (CHOP) biorepository at the Center for Applied Genomics (CAG). All subjects have provided informed consent to both genomic analysis and EMR mining. (11) A previously validated, clearly delineated, and reproducible phenotype was used to define subjects with asthma (12). This phenotype defines asthma cases as patients above the age of 4 with at least two asthma related ICD-10 codes in at least two different and independent clinical encounters, and at least one asthma related prescription. Patients with significant other pulmonary diseases including, but not limited to, cystic fibrosis and bronchopulmonary dysplasia were excluded. A comprehensive overview of this phenotype is available online. (12)
Asthma exacerbations were defined as days on which patients were prescribed systemic steroids, had ED visits, or were hospitalized due to asthma. Records associated with an exceptionally high number of steroid prescriptions were reviewed by a study physician and removed if the patient caried non-asthma related ICD-10 codes for a disease that may require maintenance systemic steroid therapy. Finally, steroid prescriptions associated with anesthesia encounters were excluded as these likely represent preparation for the effects of general anesthesia as opposed to asthma exacerbations.
Based on EMR classification patients were classified as ‘Asian American’, ‘Black or African American’, ‘White’, and ‘Other or Missing (NA)’. Patients were classified as having mild-, moderate-, or severe-asthma based on the most severe ICD-10 code found in the EMR.
Household income in home ZIP code was estimated using the zipcodeR package. (13) Household income was classified as ‘low’ or ‘high’ if the median for the patients’ zip code fell in the highest or lowest quartile for all ZIP codes in our sample, respectively.
For particulate matter 2.5 (PM2.5) and Ozone (O3) daily county level Air Quality Index (AQI) data was obtained using the pre-generated Daily AQI by county files available on the AQS website. (14) Patient ZIP codes were matched to counties using the reverse_zipcode function available in the zipcodeR package. (13)
To obtain CO, NO2, PM10 and SO2 data, we created a program in RStudio that interacts with the AQS-API to download and process relevant air pollution data. (Figure 1) (15) This program looks up the central coordinates of the home ZIP code using the zipcodeR package. (13) It then calculates the coordinates for a 10 by 10-mile square centered on these coordinates. This choice was informed by the locations of air quality monitoring stations within Philadelphia and was a compromise between the need for a small search area to preserve data validity and the requirement for a large enough area to avoid biasing results due to differential data availability. (16) Next, it obtains air pollution data for active monitoring stations within this square during the specified date range. If data from multiple measuring stations is available, the program applies an inverse distance weighting formula. Due to significant regional variation in monitoring station density and ZIP code surface area, pollutant concentration data was only modelled for patients with a home ZIP code in Philadelphia County.
We employed a case-crossover approach to compare data leading up to an asthma exacerbation with baseline data. For baseline data, we calculated the mean and standard deviation of daily air quality index (AQI) for O3 and PM2.5, as well as daily pollution concentration for CO, NO2, PM10, and SO2. To minimize bias based on seasonal variation, we selected reference dates closely aligned with asthma exacerbation, namely two weeks prior and two weeks post the week leading up to an asthma exacerbation. (17) Additionally, control dates were chosen 14 days from the dates of interest to prevent bias based on day of the week and associated variability in traffic patterns. (18, 19) If the daily AQI based on a specific pollutant or mean pollution concentration of a specific pollutant was at least two standard deviations above this mean on the day of asthma exacerbation or one of three preceding days, the asthma exacerbation was marked as an Exacerbation associated with Ambient Air Pollution (EaAAP) for that pollutant. The patients that experienced EaAAP’s were considered sensitive to the pollutant in question.
To assess the impact of AAP sensitivity on asthma severity, a multinomial log-linear logistic regression model was created in R using the purposeful selection method published by Zhang. (20) In brief, we completed a univariable analysis and included variables that met a p-value of 0.25 in the multivariable model. Next, variables that did not meet a p-value significance threshold of 0.05 in the multivariable model were removed in a stepwise fashion, ensuring that removal of variables did not meaningfully change (>20% change) the coefficients of the remaining variables in the model. After removal of a variable, the fit of the new model and the old model was using a partial likelihood ratio test. If removal of the variable did not meaningfully change the fit, the new model was preferred. Since modelled AAP data was only created for patients with a ZIP code in Philadelphia County, two models were created, one incorporating sensitivity based on the AQI data (PM2.5 and Ozone) and one incorporating the sensitivity based on the modelled data (PM10, NO2, SO2, and CO). Both models used asthma severity as the outcome variable.
A total of 8,129 pediatric patients with asthma were enrolled in the study. (Table 1) This population was associated with 34,364 asthma exacerbations. Most patients were male (4,624 patients or 56.9%) and had at least one documented allergy (6,632 patients or 81.6% of total cohort). Asthma severity was classified as mild for 4,624 patients (56.9%), moderate for 2,271 patients (31.3%), and severe for 410 patients (5.7%). The most common racial classification in our cohort was ‘Black or African American’ (4,983 patients or 61.3%), followed by ‘White’ (2,438 patients or 30.0%), ‘Other or NA’ (592 patients or 7.3%) and ‘Asian American’ (116 patients or 1.4%). The mean number of asthma exacerbations per patient was 4.23 exacerbations with a standard deviation (Sd) of 5.9 exacerbations.
We were able to obtain data for the AQI based on O3 for 28,214 exacerbations and based on PM2.5 for 31,357 exacerbations. The mean number of measuring stations per observation was 3.86 (Sd 2.63) and 4.79 (Sd of 2.49) for O3 and PM2.5 respectively. The mean AQI was 49.51 (Sd 25.87) based on O3 and 53.3 (Sd 16.10) based on PM2.5. (Figure 2) The concentration of CO was available for 10,937 exacerbations, NO2 for 15,989 exacerbations, PM10 for 8,728 exacerbations and SO2 for 16,464 exacerbations. The mean number of measuring stations contributing to the modelled air pollution was 1.69 (Sd 0.80) for CO, 1.44 (Sd 0.66) for NO2, 1.28 (Sd 0.66) for PM10, and 1.44 (Sd 0.70) for SO2. The mean modelled AAP concentration was 0.33 parts per million (ppm) (Sd 0.21) for CO, 16.52 parts per billion (ppb) (Sd 8.30) for NO2, 23.3 (Sd 11.50) microgram per cubic meter for PM10 and 1.97 ppb (Sd 2.05) for SO2. (Figure 3)
We were able to define EaAAP for both the AQI values and the modelled air pollution concentrations. Using the AQI data there were 3,013 EaAAP for O3 (11% of exacerbations with data for O3) and 4,248 EaAAP for PM2.5 (14%). Using modelled data there were 2,055 EaAAP for CO (19%), 2,322 EaAAP for NO2 (15%), 1,016 EaAAP for PM10 (12%), and 3,287 for SO2 (20%).
Patients were considered sensitive to a short-term change in a specific pollutant if at least one of their asthma exacerbations was classified as an EaAAP for that pollutant. Using AQI data, 2,057 patients met this criterion for O3 (28.36% of patients with O3 data available for at least one of their asthma exacerbations) and 2,694 patients for PM2.5 (35.52%). Using modelled data, 1,283 patients were sensitive to CO (39%), 1,472 to NO2 (36%), 672 to PM10 (27%) and 1,906 to SO2 (45%). In a multinomial log-linear logistic regression model being sensitive to ozone was associated with an Odds Ratio (OR) of 1.39 with confidence interval (CI) of 1.08-1.78 of having severe as opposed to mild asthma. In the same model, sensitivity to PM2.5 was associated with higher odds of having moderate or severe as opposed to mild asthma. (OR 1.26 with CI 1.12-1.43 and 1.37 with CI 1.07-1.76 respectively). In the second multinomial log-linear logistic regression model, sensitivity to SO2 was associated with higher odds of moderate as opposed to mild asthma (OR 1.24 with CI 1.06-1.44) and sensitivity to PM10 or SO2 was associated with higher odds of severe as opposed to mild asthma. (OR 1.63 with CI 1.12-2.37 and OR 1.58 with CI 1.12-2.23 respectively)). Covariables included in both initial models Personal allergic status, number of asthma exacerbations, history of documented antibiotic usage, history of antibiotic usage prior to the first asthma exacerbation, documented obesity, documented race, history of RSV infection, history of rhinovirus infection, documented sex, documented family history of asthma or atopic disease, documented gestational age and median household income by ZIP. (Table 2 and 3)
Despite the high global burden of AAP exposure in pediatric asthma, we currently have no effective therapies to protect our patients. Understanding and predicting individual AAP sensitivity is a prerequisite for developing these therapies. The existence of variable AAP sensitivity is supported by numerous studies highlighting genetic variants associated with AAP sensitivity. (21-23) Furthermore, prior studies on the effect of AAP have shown significant variability in the specific pollutant with the highest impact, the relative risk assigned to exposure and the time lag between exposure and effect. (18, 24-28)
In the current study we set out to identify pediatric patients with AAP sensitive asthma. To this end, we used EMR and EPA data to define EaAAP. For both O3and PM2.5 we were able to use daily AQI data to model AAP exposure. For CO, NO2, SO2 and PM10 this was not available, therefore we relied on modelled exposure data using ZIP code centroids, to avoid introducing bias, we limited modelling data to patients living in Philadelphia County. While there was remarkable consistency between the logistical regression models based on AQI and modelled data in AAP sensitivity effect size and directionality, the lack of correction for topography or meteorological factors is a limitation of the modelled exposure data. It is worth noting that despite the inexact nature of this data, prior studies using monitoring data from central measuring stations have shown clear associations between AAP exposure and healthcare utilization in pediatric asthma. (18, 26) Additionally, it is worth highlighting that assigned AAP sensitivity status refers to a temporal association between a spike in a pollutant and asthma exacerbation. Given the high correlation between various gaseous pollutants, it is possible that this association is driven by other pollutants associated with the pollutant of interest. Recognizing the limitations of both AQI and modelled air pollution concentration our study avoids using absolute values and instead relies on relative changes in either AQI or modelled pollutant concentration. Furthermore, as the EPA data is critical in setting national ambient air quality standards, the association between data from these monitoring stations and asthma exacerbations has unique public health implications.
Our study has several unique strengths. By leveraging the CAG biobank and its EMR integration we were able to select a large cohort of pediatric patients with asthma and obtain patient level data. This, taken together with the available AQI data and program we created to rapidly query and access specific pollutant data, allowed us to look at a patient-specific level at an unprecedented scale.
Overall, exacerbations occurring because of changes in AAP concentration would be expected to occur just after a significant spike in AAP concentration. In our dataset, depending on the pollutant, 11-20% of exacerbations were temporally related to a spike in AAP concentration. This number seems consistent with prior estimates attributing around 15% of exacerbations in pediatric asthma to AAP exposure. (3) We found significant associations between more severe asthma and sensitivity to Ozone, PM2.5, PM10 and SO2. (Table 2 and 3) The observed clinical impact of AAP sensitivity highlights the importance and urgency of targeted mitigation strategies.
We demonstrated that not all pediatric patients with asthma have exacerbations after spikes in AAP exposure, but those that do have more severe asthma. These findings highlight the importance of further study in variability in both AAP exposure and sensitivity. Understanding the interplay between environmental exposure and predisposition may allow us to individualize our treatment and achieve better asthma outcomes in the future. Follow-up studies leveraging (ideally) direct exposure measurements or alternatively more refined extrapolation models that incorporate topography and meteorological factors are needed to validate our results. Furthermore, a key next challenge will be the development of sensitive and specific biomarkers for both AAP exposure and the consequences of this exposure. First, such biomarkers could support the biological plausibility of causation as opposed to association. Second, if causation is supported, they could allow us to rapidly identify and target those patients most at risk for developing AAP related morbidity. Finally, they could be incorporated in future clinical trials to guard against inadvertently biassing results.