Within-microenvironment exposure to particulate matter and health effects in children with asthma: a pilot study utilizing real-time personal monitoring with GPS interface

Background Most particulate matter (PM) and health studies in children with asthma use exposures averaged over the course of a day and do not take into account spatial/temporal variability that presumably occurs as children move from home, into transit and then school microenvironments. The objectives of this work were to identify increases in morning PM exposure occurring within home, transit and school microenvironments and determine their associations with asthma-related inflammation and rescue medication use. Methods In 2007–2008, thirty Denver-area schoolchildren with asthma performed personal PM exposure monitoring using a real-time sensor integrated with a geographic information system (GIS) to apportion exposures to home, transit and school microenvironments. Concurrently, daily monitoring of the airway inflammatory biomarker urinary leukotriene E4 (uLTE4) and albuterol usage was performed. Results Mean PM exposures each morning were relatively well correlated between microenvironments for subject samples (0.3 < r < 0.8), thus limiting use of this exposure metric to attribute health effects to PM exposure in specific microenvironments. Within-microenvironment increases in exposure, such as would be characterized by one or a series of transient spikes or a sustained increase in concentration (exposure event), however, were not strongly correlated between microenvironments (|r| < 0.25). On days when children were exposed to a ≥ 5μg/m3 exposure event during transit, they demonstrated a 24.0 % increase in uLTE4 (95 % CI: 1.5 %, 51.5 %) and a 9.7 % (−5.9 %, 27.9 %) increase in albuterol usage compared to days without transit exposure events. Associations between exposure events and health outcomes in home and school microenvironments tended to be positive as well, but weaker than for transit. Conclusions School children with asthma moving across morning microenvironments experience spatially heterogeneous PM exposures with potentially varying health effects. Electronic supplementary material The online version of this article (doi:10.1186/s12940-016-0181-5) contains supplementary material, which is available to authorized users.

standard deviations of the log-transformed stratified data subsets were then calculated. These distribution parameters were used to impute values for the original zero readings via a probability integral transformation (2). The imputed, or modeled, values were then substituted for the original zero readings. These data were then log-transformed and aggregated up to the microenvironmental level. Multiple imputation procedures were performed and the outputs were evaluated for consistency; the average change in the mean values of the full dataset was less than 0.4%.

Time-Activity Apportionment
The microenvironmental apportionment methodology has been published (3), however, a brief description follows. Collected exposure data was post-processed using a temporal spatial algorithm that apportioned morning data into pre-determined location-activity categories (e.g., home, transit, school). Each 10-second data point was assigned a specific location-activity category (home, morning transit, school) using geographic proximity analyses supported by time-based rules and temperature measurement -the latter used to confirm when a subject passed from indoors to outdoors. The geographic proximity analysis determined if a recorded point lied within a predefined, two-dimensional area (i.e., a home boundary). The time-based rules further supported the proximity analysis by establishing expected times for the individual to be in the home or work/school area. For example, if the recorded position of a sample was within a certain radius of the work/school position (e.g., 50 m) during expected work/school hours then the exposure was assigned to the work/school category. Similarly, the home category was assigned if the recorded position of the sample was within the defined home area during expected home hours. If the recorded position of the sample was neither at home or work/school, the sample was considered in-transit.

Models and interpretations of effect estimates
Models for LTE 4 take the form where Y is LTE 4 , x is the pollutant variable of interest, and i and j index subject and time, respectively. In the model, it is assumed that Additionally, errors ij e are assumed to follow a spatial power structure (i.e., an AR(1) structure that accounts for intermittent responses) within subjects, and independent between subjects;. Due to taking the natural log of LTE 4 , 1 e b is the multiplicative increase in the mean of Y for a 1-unit increase in x, while where ij µ is the mean albuterol use for subject i at time j, Friday is an indicator for that day of the week. (On Monday through Thursday children received an albuterol pre-treat before exercising, but not on Fridays because they did not have a physical education class that day.) The model was fit using generalized estimating equations, using an AR(1) working correlation structure to account for repeated measures. Due to the use of the natural log link, 1 e b will also have a multiplicative increase interpretation, although in this case the natural log is applied to the mean of outcome rather than the outcome itself.

Construction of increase and exposure event variables
Initially, exposure patterns within each sample-day (n=125) were examined individually by 2 of the coauthors (MS and JV) to determine, qualitatively, whether a noticeable increase in PM exposure occurred within a given microenvironment. We then more formally constructed increase and exposure event variables to capture concentration spikes within microenvironments. The initial qualitative approach was very consistent with the more quantitative metrics (92% agreement with exposure events of at least 5µg/m 3 and 88% agreement with exposure events of at least 10µg/m 3 ). Due to the better fits in health outcome models for 5µg/m 3 , we chose to primarily focus on that cut-point in the manuscript. 12.6 (-5.0, 33.6) 2-day 6.5 (-7.0, 22.0) 9.6 (-2.1, 22.7) For 2-day moving average estimates, records were weighted by number of records used in the moving average (1 or 2). For Albuterol usage models, temperature and Friday indicator covariates were used; for LTE 4 models, temperature and cold indicator were used (the lag for cold was set as the same for the air pollution variable). IQR for 0 and 1 day lags were 1.84, 1.39 and 1.54 for home, school and transit, respectively (for 2-day average they were 1.72, 1.36 and 1.47). 30 subjects were available for analysis; number of records used for model fits were: 114, 80 and 143 for 0, 1 and 2-day moving averages for albuterol usage, respectively, and 80, 58 and 111 for LTE 4 . For more detail, see the text. Figure S2: Percent increase in uLTE 4 per IQR increase in mean personal exposure to PM 1.5 by home, a.m. transit, and school microenvironments for same day (lag 0), one-day lag, and a 2-day moving average. Error bars represent 95% confidence intervals. Figure S3: Percent increase in school-day albuterol usage per IQR increase in mean personal exposure to PM 1.5 by home, a.m. transit, and school microenvironments for same day (lag 0), one-day lag, and a 2-day moving average. Error bars represent 95% confidence intervals.

Outcome
Micro