The study was conducted in an elementary school attached to Peking University [see Supplemental Material, (http://dx.doi.org/10.1289/ehp.1103461
)], which is 178 m and 300 m away from two busy roads. A continuous air pollution monitoring station is located about 650 m southwest of the school. A comparison of the air pollutant concentrations between the monitoring station and the elementary school showed good agreement (see Supplemental Material, ).
Figure 1 Average 0–24 hr concentrations for BC (A) and PM2.5 (B) and the corresponding averaged eNO concentrations for all subjects (C) during V1–V5. The dashed vertical line separates measurements taken before and during the Beijing Olympic Games (more ...)
Figure 2 Relationship of BC (A) and PM2.5 (B) with eNO by the GEE model using nonparametric smoothed function smoothing (degree = 1, span = 0.65) with 95% bootstrap confidence intervals (dashed lines) after controlling for BMI, asthma, temperature, RH, and time. (more ...)
Participants. At the beginning of the study, questionnaires were distributed to 734 students in grades 3 and 4 to obtain data about demographics, health status, and symptoms related to asthma, rhinitis, and eczema. Among the 437 students who returned the questionnaires, we recruited 8 students in grade 4 (6 boys, 2 girls) with doctor-diagnosed asthma. In addition, we randomly recruited 30 healthy children (12 boys, 18 girls) in grade 4 with no personal or family history of chronic respiratory disease or chronic inflammation. In total, we had 38 (18 boys, 20 girls) participants in the present study (mean age, 10.6 years).
The study included five observation periods, three in 2007 [visit 1 (V1), 11–22 June 2007; V2, 10–20 September 2007; V3, 10–21 December 2007] and two in 2008 (V4, 16–27 June 2008; V5, 1–12 September 2008). The last three observation periods each lasted 2 weeks and consisted of 10 weekday observations per subject. Only 8 days of observations were collected in June 2007 because of instrument malfunctions, and only 9 days of observations were collected in September 2007 because of a school holiday. During each period, we measured eNO for each subject every weekday (Monday through Friday), during the school lunch break. Data from one subject who withdrew after V2 in 2007, and one who withdrew before V5 in 2008 were not included in the analysis. Thus, we collected 1,581 valid observations from 36 subjects who each completed five visits.
The study protocol was approved by the Ethics Committee of the Centre of Health Sciences, Peking University. The guardians of all children gave their written consent for the children to participate in the research.
We collected the exhaled air samples from the subjects following the recommendations of the ATS/ERS (2005) for offline measurement. The exhaled air collection device was made of Teflon. A tube filled with activated carbon was used to filter out NO from the ambient air and was shown to remove up to 99.6% of NO in a laboratory simulation of the protocol used in the field. The device was equipped with a flow meter as a flow restrictor and a pressure indicator (Gong et al. 2006
Before sampling, each subject was asked to put the mouthpiece of the device tightly in his or her mouth, inhale deeply, and then exhale to wash the “dead space” from the device. This procedure was repeated twice. To sample the exhaled air, the subjects were instructed to inhale to tidal capacity and then exhale into a 4-L air-sampling bag made of aluminium foil, with a constant flow of 150 L/hr and a positive pressure of 13 cm H2O (to close the soft palate and prevent nasal exhalation) (ATS/ERS 2005).
We measured eNO concentrations in the 4-L air-sampling bag within 4 hr with a chemiluminescence NO–NO2–NOx (nitrogen oxides) analyzer (model 42i; ThermoScientific, Rockford, IL, USA), which had a detection limit of 0.4 ppb and a detection range of 0–100 ppb. The analyzer was calibrated every day with five concentrations (0–80 ppb) of NO (Beijing Huayuan Gas Chemical Industry Company Limited, Beijing, China) mixed in ultrahigh purity nitrogen (99.999%; Beijing Haikeyuanchang Practical Gas Co., Ltd.). We used a calibration curve to calculate the eNO concentrations in exhaled air of the subjects.
Ambient air pollutant measurement.
Hourly averaged concentrations of PM2.5
(tapered element oscillating microbalance, RP1400a; ThermoScientific), BC (multiangle absorption photometer, model 5012; ThermoScientific), NOx
, and CO (models 9841A, 9850A, 9830A; ECOTECH Pty Ltd., Knoxfield, Australia), and meteorologic parameters (Met One Instruments Inc., Grants Pass, OR, USA) were concurrently measured at the continuous monitoring site during V1–V5. The instruments for gaseous pollutants were automatically calibrated at 2300 hours every day. Additional details on methods and a comparison of PM2.5
and BC concentrations measured at the school and the fixed monitoring site are provided in the Supplemental Material (http://dx.doi.org/10.1289/ehp.1103461
). In this study, we only report the results associated with outdoor concentrations of PM2.5
, BC, CO, SO2
, and NO2
Statistical analysis. t
-Tests were used to derive p
-values for differences in mean air pollutants and eNO before and during the Olympic Games air quality intervention. Generalized estimating equations (GEE) (Liang and Zeger 1986
), which correct for repeated measurements within subjects, were used to estimate associations between pollutants and eNO. To check the validity of the correlations between repeated measurements on the same subject, we used the “quasi-likelihood under the independence model criterion” (Pan 2001
), which indicated that an autoregressive correlation matrix at lag 1 adequately fitted the data. The GEE calculations were conducted with the STATA statistical package (version 9.1; StataCorp LP, College Station, TX, USA) with robust standard errors, using the “force” option to allow for unequally spaced observations. The eNO data were transformed logarithmically. The model included adjustments for temperature, relative humidity (RH), body mass index (BMI), and asthma (yes or no). We also modeled interaction terms for asthma and each pollutant to evaluate differences in responses between nonasthmatic and asthmatic children.
To graphically analyze the exposure–response relationship between BC and predicted eNO, a nonparametric smoothed function (LOESS) was created with R statistical software (version 2.4.1; R Development Core Team, Vienna, Austria) (Cleveland 1979
). The degree and smoothing parameter of the LOESS model were optimized according to the residual plot with a horizontal line fit to zero (Jacoby 2000
). A LOESS model with a degree of 1 and a smoothing parameter of 0.65 was chosen for the analysis.
Two-pollutant models were used to investigate the stability of single-pollutant estimates and to check whether estimates for single pollutants might be biased because of confounding by other pollutants. We initially conducted the analysis including BC alone and PM2.5 alone and then considered the effects of SO2, NO2, and CO. These two-pollutant models provided the estimates of the individual effects of CO, SO2, and NO2 on eNO after adjustment for BC and PM2.5. All estimates were calculated with adjustment for temperature, RH, BMI, and asthma (yes or no).
To estimate effects of lagged BC exposures, we used a polynomial distributed-lag (PDL) model (Mar et al. 2005
; Schwartz 2000
). The PDL model allows effects at different lags to be estimated in the same model. To allow for autocorrelation, we adopted a third-degree PDL model with a maximum lag of 47 hr before eNO sampling based on the Akaike information criterion. The model equation can be written as follows:
ln(eNOi) = β0X0 + β1X1 + β2X2 + β3X3
+ β4temperature + β5RH
+ β6asthmai + β7BMIi,
n = 1, 2, 3, 
where eNOi is the eNO concentration of the subject i; BClagj is the BC concentration at the jth hour before the eNO measurement; asthmai is the subject’s asthma status; and BMIi is the individual subject’s BMI.
The individual hourly estimate effects were obtained using the following equation:
BClagj = β0 + jβ1 + j2β2 + j3β3. 
The PDL model was fit in STATA using commands developed by McDowell (2004)