Personal exposure measurements of school-children to fine particulate matter (PM2.5) in winter of 2013, Shanghai, China

Objective The aim of this study was to perform an exposure assessment of PM2.5 (particulate matter less than 2.5μm in aerodynamic diameter) among children and to explore the potential sources of exposure from both indoor and outdoor environments. Methods In terms of real-time exposure measurements of PM2.5, we collected data from 57 children aged 8–12 years (9.64 ± 0.93 years) in two schools in Shanghai, China. Simultaneously, questionnaire surveys and time-activity diaries were used to estimate the environment at home and daily time-activity patterns in order to estimate the exposure dose of PM2.5 in these children. Principle component regression analysis was used to explore the influence of potential sources of PM2.5 exposure. Results All the median personal exposure and microenvironment PM2.5 concentrations greatly exceeded the daily 24-h PM2.5 Ambient Air Quality Standards of China, the USA, and the World Health Organization (WHO). The median Etotal (the sum of the PM2.5 exposure levels in different microenvironment and fractional time) of all students was 3014.13 (μg.h)/m3. The concentration of time-weighted average (TWA) exposure of all students was 137.01 μg/m3. The median TWA exposure level during the on-campus period (135.81 μg/m3) was significantly higher than the off-campus period (115.50 μg/m3, P = 0.013 < 0.05). Besides ambient air pollution and meteorological conditions, storey height of the classroom and mode of transportation to school were significantly correlated with children’s daily PM2.5 exposure. Conclusions Children in the two selected schools were exposed to high concentrations of PM2.5 in winter of 2013 in Shanghai. Their personal PM2.5 exposure was mainly associated with ambient air conditions, storey height of the classroom, and children’s transportation mode to school.

Introduction assessment. This indirect exposure assessment method has been reported more than twenty years ago and has been used for measurement of many air pollutants such as benzene exposure, carbon monoxide (CO) exposure, and particle exposure [16,17], which used validated microenvironment models and human activity pattern data obtained from questionnaires to predict exposure levels in a certain population [18].
Given that there were few studies which have carefully looked at the relationship between personal activity and exposure levels to PM 2.5 with respect to time and real-time pollutant monitoring, especially in China, we conducted personal exposure level assessments of PM 2.5 among school-children in two schools in Shanghai. Particularly, this research was conducted in the episode of 2013 exactly and this article intends to highlight the personal exposure level of PM 2.5 among school students in this period by means of micro-environmental monitoring and personal sampling, and to explore influence of potential sources of PM 2.5 exposure in both indoor and outdoor environments.

Study participants and design
This study was designed to continuously monitor changes in the PM 2.5 level, while simultaneously recording activities among school-children at school and home. Based on air pollution and human health monitoring area data in Shanghai and the cluster sampling principle, two primary schools were selected. School A is located in the downtown Shanghai and situated 1 km west of the monitoring spot at the Shanghai Normal University, Xuhui district, while School B is located in the suburban area of Baoshan district and situated 4.7 km north of the monitoring spot at Hongkong Liangcheng area. Following the principles of gender equity and voluntary participation, 57 children of the 3 rd , 4 th and 5 th grade from the two primary schools were recruited. All students lived in two communities close to the two schools. Our study was approved by the ethics committee of Shanghai Municipal Centre for Disease Control and Prevention (IORG No: IORG0000630). Written informed consent was obtained from the parent or guardian of each participant before the study was initiated. The protocol included four parts.
Questionnaire survey. We conducted a cross-sectional survey on indoor and outdoor environment among these children. The main variables included general information, living condition, and lifestyle information. PM 2.5 measurements inside the campus microenvironment. Continuous PM 2.5 measurements were collected simultaneously from several micro-environments inside the school where students stayed (e.g. classrooms, main corridors, and the playgrounds). TSI DUST-TRAK TM DRX (Model 8533, TSI Inc. Paul, MN, USA) apparatus was set up in each environment from about 8 a.m. to 4 p.m. (Investigations in Schools A and B were taken between November 20th to 28th and from December 18th to 26th in 2013, respectively) and calibrated with the manufacturer's high-efficiency particulate filter before sampling. One sampler was placed in the playground and the other was placed in the main corridor, respectively for 6 days during the working days. Two samplers were placed in each classroom for continuous two days based on the diagonal distribution principle, with one in the front of the classroom and another in the rear. All sampling instruments were placed approximately 1 meter away from walls and other barriers, and were at a height of 1.2 meters above ground level, which was approximately at the height of the breathing zone of children. The sampling flow rate was 3 l/ min, and values were recorded continuously over 8 hours with time intervals of 1 min. Recorded values were corrected to 1 μg/m 3 , and the limit of detection (LOD) was 1μg/m 3 .
During the sampling period, all classrooms retained ventilation habits as usual, including the opening and closing of windows and doors.
Personal measurements outside campus. During the off-campus period (from 4 p.m. to 6 a.m. the next day), PM 2.5 concentrations were measured by a set of real-time laser diode photometers (model no. SidePak TM AM510, TSI Inc, USA), which were placed in small bags. A sampling air inlet was fixed in the vicinity of the students' breath zone. All participants were asked to carry the sampling bags from about 4 p.m. to 8 a.m. the next day, although this excluded the sleeping and showering periods. When the monitor was taken off, it was placed within the breathing zone of the subjects. All students returned sampling bags on the next morning. Each instrument was cleaned, greased and batteries were replaced after daily sampling. All instruments were reset daily before sampling, according to the manufacturer's instructions. After school the next day, the same sampling bags were re-distributed to the same students who took the sampling bags on the first day. All participating students were required to finish personal sampling for a continuous 2 days period. Meanwhile, ambient air pollutants concentration (e.g. PM 10 , PM 2.5 , NO 2 and SO 2 ) and meteorological indicators (e.g. temperature, humidity, and wind speed) in the two monitoring spots closed to the schools were collected from the Real-time Air Quality Reporting System of Shanghai Environmental Monitoring Centre (http://www.semc.com.cn/aqi/home/Index.aspx) and Shanghai Meteorological Service, respectively.
Time-activity diary investigation. Each student was asked to finish a diary questionnaire that included time, activities, and location. The on-campus part of the time-activity questionnaire was based on the school timetable, and the off-campus part was recorded every 30 minutes according to individual schedules by participants.

Statistical analysis
Real-time PM 2.5 concentrations were calculated by taking the average PM 2.5 mass concentration obtained from the gravimetric method for the sake of reliability. Gravimetric PM 2.5 was measured according to the method stipulated by the Environmental Protection Agency of United State (U.S. EPA) and research conducted by Jiang et.al. [19]. In this investigation, linear regression analysis indicated that the PM 2.5 concentration could be measured by two laser diode photometer measurements with linear regression models of y = 0.602x+0.774 (n = 21, R 2 = 0.799, P <0.001) and y = 0.577x-1.467 (n = 21, R 2 = 0.769, P <0.001). On average, the PM 2.5 concentrations reported by the two laser diode photometer monitors in our research were approximately 2.25-and 2.59-fold as high as the gravimetric measurements (Please refer to Supporting information file S1 Protocol).
The exposure level (E total ) was calculated as the sum of the partial exposure levels according to the relationship described by Eq 1 [18](where Ci is the concentration in the ith microenvironments, Ti is the fractional time spent in the ith microenvironment, and N is the number of microenvironments.) TWA is the time-weighted average exposure, which means the average concentrations weighted on the integration period described by Eq 2.
Principal regression analysis was used to explore indoor and outdoor impact of ambient PM 2.5 pollution, meteorology and individual activities on personal PM 2.5 exposure levels. Students' PM 2.5 total exposure concentrations (E total ) over the two days were treated as the dependent variables, while other potential factors were treated as the independent variables. The latter included ambient air pollutant concentrations, meteorological factors, school position, sex, age, height, weight, family member status, mode of transportation to school, exercise, second-hand smoke exposure, vehicles near houses and other variables. All data were examined for validity and complied with our standard operating procedures. Flagged data were removed when battery failure or disconnected power supply was detected. All statistical analysis was performed by SAS for Windows (version 9.4; SAS Institute Inc., 2003) and the level of significance was defined as P <0.05 (2-tailed).

Basic participant characteristics
57 students participated in the investigation with 30 (52.6%) children in school A and 27 (47.4%) children in school B. Table 1  In spring and winter, 51 (91.1%) homes tended to report opening the window(s) every day. 47 (83.9%) children often exercised. 48 (85.7%) children were from families that used smoke exhaust ventilator in their kitchen every day. Children's mode of transportation to school varied, which was 24 (42.9%) on foot i.e. walking, 20 (35.7%) by bike and 12 (21.4%) by bus, car or subway. Table 2 summarizes the real-time concentration (calibrated results) of PM 2.5 in different microenvironments in the two schools. Median and inter-quartile ranges (P 25 -P 75 ) were reported together with the arithmetic means and these were used to describe the distribution due to the right-skewed distribution of all measurements. All the median personal exposure and microenvironment PM 2.5 concentrations greatly exceeded the secondary standard of daily 24-h PM 2.5 Ambient Air Quality Standards of China (75μg/ m 3 ) [20], the current U.S. EPA daily ambient standard of 35 μg/m 3 [21], and the WHO Air Quality Guideline of 25μg/m 3 [22]. The median PM 2.5 level in the rear classroom microenvironments in school A was 139.84 μg/ m 3 . In comparison, the median PM 2.5 level in school B was 144.65 μg/m 3 , which was significantly higher than in school A. However, the PM 2.5 levels in other microenvironments in school B were not significantly different than in school A. Moreover, school B seemed to have an extremely high concentration of PM 2.5 at the personal exposure level, which was 131.65 μg/ m 3 , while the exposure level in school A was much lower, with a median value of 103.25 μg/m 3 (P < 0.001). Table 3 summarizes the ambient fixed monitoring data obtained from Shanghai Environmental Monitoring Centre and Shanghai Meteorological Service for the same period of exposure measurements. Ambient PM 2.5 and SO 2 concentrations in the fixed monitoring spot and some meteorological indicators such as relative humidity (RH) and wind speed (WS) close to school B were significantly higher than for school A (P < 0.001), while temperatures close to school B were significantly lower than for school A (P < 0.001).

Time-activity pattern survey
During the on-campus period, the constituent ratio of time-activity patterns in the different micro-environments in the two schools were significantly different (χ 2 = 26.988, P <0.001). Students in school A spent much more time in classroom, which concomitantly resulted in less time in the playground, corridors and other places compared with those students in school B. In particular, students in school A spent duration of 377 mins in the classroom compared with students in school B who spent 355 mins in classroom. We calculated a constituent ratio for the average time spent in different micro-environments such as the bedroom, dining room, bath room, kitchen, being on the road and other places. Students in school A spent 960 mins during the off-campus period in particular micro-environments, which was 706 mins (73.5%) in the bedroom, 113 mins (11.8%) in the dining room, 36 mins (0.3%) in the bath room, 3 mins (0.3%) in the kitchen, 51 mins (5.3%) on the road, and 51 mins (5.3%) in other places, respectively. Students in school B spent 930 mins during the off-campus period in the following micro-environments: 701 mins (75.3%) in the bedroom, 103 mins (11.1%) in the dining room, 35 mins (3.8%) in the bath room, 9 mins (1.0%) in the kitchen, 33 mins (3.5%) on the road, and 50 mins (5.4%) in other places, respectively. The constituent ratio for the time-activity patterns in different micro-environments during the off-campus periods between two school students were not significantly different (χ 2 = 6.919, P = 0.227).  PM 2.5 exposure concentrations during the on-campus period were all much higher than ambient PM 2.5 concentrations in closed environment monitoring spots during the investigation period. Two peaks were detected at 8 a.m. and 3 p.m. or 2 p.m. during the on-campus period. During the off-campus period, the variations in personal exposure of students to PM 2.5 between the two schools were different. Students in school A seemed to have a similar exposure level to that of ambient PM 2.5 exposure. Furthermore, students' personal exposure level peaked at 9 p.m. and started to decline from 9 p.m. to 5 a.m. the next day. Students' personal exposure level in school B peaked at 6 p.m. and started to decline from 6 p.m. to 1 a.m. the next day.

PM 2.5 exposure level of school students
Based on micro-environmental PM 2.5 monitoring results, personal sampling and students' time-activity patterns, the PM 2.5 exposure levels for all students during the on-campus and offcampus school periods were calculated ( Table 4). The median E total of all students in a 22-hr period was 3014.13 (μg.h)/m 3 . The median E total of school A and school B were 2783.77 (μg.h)/ m 3 and 3404.60 (μg.h)/m 3 respectively. The median TWA in a 22-hr period of all students was 137.01 μg/m 3 . The median TWA of school A and school B were 126.53 μg/m 3 and 154.75 μg/ m 3 respectively. No significant differences in the median E total and TWA in a 22-hr period were observed between the two schools. During the on-campus period, the median TWA of all students was 135.81 μg/m 3 , which was significantly higher than the off-campus period value  (115.50 μg/m 3 , P = 0.013 < 0.05). As for school A, the median TWA during the on-campus was also significantly higher than the off-campus period value (P = 0.017 < 0.05). No significant differences in median TWA were observed between the two schools.

Correlation between ambient air pollutants, meteorological data and exposure levels
Spearman correlation coefficients were used to compare ambient air pollutants, meteorological data and exposure levels ( Table 5). The high correlations were observed between on-campus PM 2.5 exposure levels (E total _on-campus) and ambient PM 2.5 (r = 0.769), SO 2 (r = 0.709), NO 2 (r = 0.306), and PM 10 (r = 0.281). Similarly, the off-campus PM 2.5 exposure level (E total _ off-campus) correlated with ambient air PM 2.5 (r = 0.361) and NO 2 (r = 0.543). Moreover, a strong correlation between air pollutants and meteorological indicators were observed. Hence, there was strong co-linearity between exposure levels and ambient air pollutant concentrations, and meteorological indicators.

Principal component analysis (PCA)
PCA was used to eliminate co-linearity between exposure levels and ambient air pollutant concentrations, and monitored meteorological indicators [23]. We estimated a correlation matrix incorporating a total of 7 variables including PM 2.5 , PM 10 , SO 2 , NO 2, temperature, RH, and WS. Two PCs were extracted to interpret the original datasets after the varimax rotation. The two PC S could explain approximately 72% of the total variances noted for the original data concourses. PC1 mainly reflected the information that was associated with ambient air pollution (i.e. PM 2.5 , PM 10 , SO 2 , and NO 2 ). PC2 reflected the information related to meteorological conditions (i.e. temperature, RH, and WS).

Multiple linear regression
We explored influence factors of children's PM 2.5 total exposure by multiple linear regression (Table 6). Children's personal PM 2.5 exposure in the two schools was associated with various factors. During the on-campus period, ambient air pollution (PC1) and meteorological conditions (PC2) were strongly correlated with E total (P <0.0001). Storey height of the classroom was negatively correlated with students' on-campus personal PM 2.5 E total levels (P = 0.0316). The regression equation was statistically significant (adjust R 2 = 0.585 F = 50.33, P < 0.001).
During the off-campus period, ambient air pollution (PC1) and children's age were all positively correlated with students' E total exposure level (P = 0.0027 and P = 0.0260). However, students' mode of transportation to school was negatively correlated with students' E total level (P = 0.0266). The regression equation was statistically significant (adjusted R 2 = 0.168, F = 6.46, P = 0.0005).

Discussion
The importance of indoor air quality (IAQ) in school environments has been globally highlighted [24,25]. In our investigation, the PM 2.5 concentration of all school micro-environments greatly exceeded the secondary standard of daily 24-h PM 2.5 Ambient Air Quality Standards of China (75μg/ m 3 ) [20], the current U.S. EPA daily ambient standard of 35 μg/m 3 [21], and the WHO Air Quality Guideline of 25μg/m 3 [22]. Moreover, the average PM 2.5 concentration in the classrooms was higher than in the corridors and the playgrounds. Furthermore, it was obvious that the average PM 2.5 concentration was higher in the front of classrooms than in the rear of classrooms, which was probably caused by proximity to the blackboard position and more exhaled aerosols from teachers and students in this direction. Moreover, the corridor was a relatively semi-closed micro-environment, which was partly exposed to external ambient environment but without intensive or long-term human activities. Therefore, PM 2.5 concentrations in the corridors are lower than those in the classrooms. The playgrounds were open-air microenvironments, and the PM 2.5 concentration was equivalent to ambient PM 2.5 levels except for those during centralized physical activities. These included the morning exercise period and physical activity times. Rovelli et al. claimed that one possible reason for PM 2.5 pollution in the classroom was the re-suspension of settled particles due to insufficient ventilation, frequently cleaned indoor surfaces, and the presence of children and their movement [25]. Fromme et al. also found a higher average classroom PM 2.5 concentration compared with the corresponding outdoor level at a German primary school [26]. Therefore, classroom air quality is one of the key aspects in exposure analysis and assessment of students' PM 2.5 exposure, air pollution control and health precautions. As reported by de Oliveira et al., children between the ages of 6 and 14 were exposed to a higher PM 2.5 dose during the dry season than during the rainy season [27]. Winter in Shanghai is generally considered to be the dry haze season because there is less rain and relative humidity was <80% often [28]. In our study, children's daily PM 2.5 exposure level reached a median value of 137.01 μg/m 3 . According to monitoring data from the Shanghai EPA, air quality during December in 2013 in Shanghai was rated 'extremely serious', which may consequently lead to higher PM 2.5 exposure in primary school students. According to Brown et al., ventilation  may have resulted in significantly higher personal exposure to particles originating from ambient sources [29]. Our study also showed the same results where students' personal PM 2.5 exposure levels were more strongly associated with ambient PM 2.5 air pollution. It is supposed that ventilation hobby of Chinese people either in summer or winter (91.1% children family had the hobby of ventilation every day even in winter) and less residential building air tightness in China are the probably reasons for this problem. According to a report by Wang et al., for an apartment with normal air tightness and without any HVAC-filter system, most indoor PM 2.5 was originated from outdoor-generated particles and closed windows can only play a very weak role on the decline of indoor PM 2.5 concentrations [11].Several studies have demonstrated that central HVAC (heating, ventilating, and air-conditioning) in residences, and children's gender were likely to have a significant impact on exposure to particulate matter [13,[30][31][32][33]. However, in our research, no significant impact was observed for central HVAC or gender differences to personal PM 2.5 exposure levels. Therefore, we consider these results in terms of the limitations of the research design and perform further research if necessary. Our investigation was performed in winter of 2013, during which an extremely rare air pollution event occurred in Shanghai because of excessive emissions and unique meteorological conditions. Although students probably reduced the duration of their outdoor activities during this period, these students were still exposed to high PM 2.5 ambient contamination. In addition, the duration of sports activities for these children was significantly insufficient both at home and school, according to our research. One probable reason was that these children preferred to stay indoors during the extremely rare air pollution event but another problem that surfaced was the heavy burden of schooling in China, which is of significant concern. This issue should be considered by government authorities. Increasingly, scientists have reported that personal exposure to air pollutant concentrations was strongly associated with the daily activity patterns, lifestyle, and the different microenvironments in which they frequently occurred [5,34,35]. Based on our results from PCA regression, children's personal PM 2.5 exposure was associated with various factors. For example, ambient air pollution, meteorological conditions and children's age were strongly correlated with personal PM 2.5 exposure levels. This result demonstrated again that PM 2.5 from ambient origins predominantly contributes to personal PM 2.5 exposure in China [36]. Multiple linear regression result showed that children's age was strongly correlated with personal PM 2.5 exposure levels, which was probably because of the fact that an elder child had longer time of and more intense outdoor activities, so that their chance of exposure to outdoor PM 2.5 is higher compared with a younger child. Nonetheless, storey height of classroom and students' transportation mode to school was negatively correlated with students' exposure levels. The PM 2.5 exposure levels in students were different depending on the different storey height of classroom and modes of transportation to school. Therefore, we recommend that children should take cars, buses or the subway to school as much as possible, especially during periods of haze or serious pollution.
An important advantage of real-time continuous instrumentation is the ability to determine short-term temporal and spatial variation. Although real-time sampling has greater measurement error than the US federal reference method (FRM) gravimetric PM 2.5 samplers, it can provide useful real-time information. As reported by Jeff D. et al., the 24-h average DustTrack levels are well correlated with FRM levels with a slope of 2.57 and an R 2 of 0.859 (P <0.0001) [37]. In our research, the two laser diode photometer monitors were approximately 2.25-and 2.59-fold as high as the gravimetric measurements with the R 2 of 0.799 and 0.769 (P <0.001), respectively. Therefore, Such data are of use especially if any biases are consistent and when statistical adjustment can be made and these types of continuous measurements could be accepted and used in many studies [38][39][40]. However, we had to admit that statistical power of the multiple linear regression results in our research was low because of a small sample size. Therefore, we should consider further studies with a larger sample size, while also performing the study throughout all seasons and improving our monitoring method in the future.
We made use of personal sampling and micro-environmental monitoring to explore children's PM 2.5 exposure levels. Meanwhile, we analyzed children's real-time exposure variation in a day and explored the probable impact of daily exposures. The method we used can be used as a reference, especially for susceptible populations such as the elderly, pregnant women, and other groups. In contrast, children's exposure levels reported here were monitored during an extremely polluted episode in Shanghai, which could help us understand children's exposure to PM 2.5 in China more generally. Our study findings should be of interest to the government or relative institutions so that appropriate policies to protect children can be considered.

Conclusions
These primary school students in Shanghai were exposed to high concentrations of pollutants during a serious haze period in winter of 2013. All the median personal exposure and microenvironment PM 2.5 concentrations greatly exceeded the daily 24-h PM 2.5 Ambient Air Quality Standards of China, the USA, and the World Health Organization (WHO). The concentration of time-weighted average (TWA) exposure level of all the students was 137.01 μg/m 3 . Based on our research, children's personal PM 2.5 exposure was associated with various factors. Besides ambient air pollution and meteorological conditions, storey height of the classroom, and mode of transportation to school were all significantly correlated with personal PM 2.5 exposure levels. Since outdoor air pollution has been an important source of PM 2.5 exposure among children, the Chinese government must take strong environmental actions to combat haze. On the other hand, school children and families also should take measures to prevent personal exposure of pollution to children. What is equally important is that we need to pay more attention to the problem of insufficient physical activity time for primary school students, as this can result in public health issues later in life. Only through the joint efforts of governments, schools, and families can we minimize the health risks of air pollution to children.