Objectively Measured Association Between Air Pollution and Physical Activity and Sedentary Behavior Among College Students in Beijing

Background: Air pollution has become a major environmental health risk factor, notably in China. Air pollution potentially has the impact of human populations’ health behavior. Gaps in scientic literature remain regarding more accurately estimates the relationship between air pollution and sedentary behavior in China. The purpose of this study is to examine the association between hourly air pollution AP on hourly physical activity (PA) and sedentary behavior (SB) among college students in Beijing, China. The secondary aim was to examine such associations varied at specic times. Methods: A total of 340 participants were recruited from the Tsinghua University, in Beijing, China. Accelerometers provided PA measures, including moderate-to-vigorous physical activity (MVPA), walking steps, energy expenditure and sedentary time for 7 consecutive days. Corresponding AP data by the Beijing Municipal Ecological Environment Bureau in the closed site (Wan Liu site) at Tsinghua University were collected to include average hourly air quality index (AQI) and PM 2.5 (µg/m³). Associations were estimated using linear individual xed-effect regressions. Results: A one level increase in hourly air quality index (AQI) was associated with a reduction in one-hour MVPA by 0.083 minutes (95% CI =[-0.137, -0.029]), 8.8 walking steps (95% CI = [-15.0, -2.6]), and 0.65 kcals of energy expenditure (95% CI =-[1.03, -0.27]). A 10µg/m³ increase in AP concentration in hourly PM 2.5 was associated with a reduction in one-hour MVPA by 0.021 minutes (95% CI = [-0.033, -0.010]), 2.2 walking steps (95% CI = [-3.5, -0.9]), 0.170 kcals of energy expenditure (95% CI = [-0.250, -0.089]), and an increase in one-hour SB by 0.045 (95% CI = [0.005, 0.0845]). At a specic time, there are stronger negative associations of AQI and PM 2.5 to PA at 8 am, 4 pm, 5 pm and 7 pm. Similarly, stronger positive associations were found at one hour AQI and PM 2.5 with SB at 8 am, 9 am, 11 am, and 7 pm. Conclusions: AP may discourage PA and increases SB among freshman students living in Beijing, China. The impact of AP on PA and sedentary behavior at a specic time may be different.


Introduction
Air pollution (AP) has become a major environmental health risk factor for overall health worldwide. AP levels in China have increased rapidly due to the industrialization use of fossil fuel and population growth [1]. Previous evidences has showed that exposure to AP has been detrimental to various health outcomes e.g., cardiovascular disease, stroke, lung cancer, respiratory disease, all-cause mortality, sleep apnea, and depression [2][3][4][5][6][7][8][9][10].
Strong evidence indicates that engaging in regular physical activity (PA) has many health bene ts including increased effective weight management, reduced risk of all-cause mortality, and prevention and management of chronic diseases, e.g., cardiovascular diseases, diabetes, colon and breast cancer, hypertension, coronary heart disease, and osteoarthritis [11][12][13][14]. While there are several health bene ts to PA, performing PA under high levels of AP may accelerate the risks for adverse health effects such as asthma attacks and heart or lung pathologies [15][16][17]. Outdoor parks and playgrounds are common places to perform PA in China [18], and the presence of AP in China may further discourage young adults from engaging in regular PA and exercise [19].
Substantial research has measured the effects of AP on health outcomes, but few studies have determined its impact on health-related behaviors, speci cally PA and SB behavior. There are three major gaps in the scienti c literature that need for further investigation. First, existing research on the associations between AP, PA and SB are limited. Second, most previous studies have used a self-reported survey method [26], therefore, existing studies were subject to social desirability bias and limited by the frequency of PA performance and SB data (i.e., week by week, month by month, or year by year). This study is the rst to provide precise, objectively measured data through use of the digital accelerometers to measure PA and SB among university freshman students in China. A digital accelerometer omits the chance for social desirability bias while providing minute-by-minute measurements of PA and SB. Third, to date, all previous studies estimated the impact of AP on health-related behaviors week on a week, month by month, or year by year basis. There is currently no study which investigates the relationship between AP and PA and SB at a speci c time.
This study reported the associations between AP, PA and SB among university freshman students living in Beijing, China. Objectively measured data were collected by using of the digital accelerometers during this study. The nal sample size was 340 participants. Hourly AP data and minute-by-minute PA and SB data were measured. We hypothesized that in response to high levels of AP, university freshmen reduced their PA behaviors and increased their SB. We also hypothesized that such associations varied at a speci c time.

Participants and Sampling Procedure
The study was conducted from the period of November 2017 to April 2018. All freshmen students are required to take the same physical education class in Tsinghua University. 344 freshmen students who were enrolled in the required physical education class were recruited as participants. Participants were . Upon acceptance, the subjects were asked to visit our lab to get a wGT3X-BT device tted. Height and weight were measured to the nearest 0.1 kg using a Seca beam scale (Tsinghua Tongfang, S5000, China). Each participant completed one paper-pencil based health survey on demographic status (age, gender, ethnicity, lifestyle smoking, drinking, and mental and physical health conditions). Of the 344 recruited, 340 subjects had completed the study (4 participants' data were excluded due to 2 devices errors, and 2 missing accelerometers that were not returned). All participants gave informed consent, and the study was approved by the Tsinghua University Institutional Review Board (IRB #2017DX02_11).

PA and SB measurement
The participants were instructed to wear a wGT3X-BT accelerometer over the right hip on a waist band for at least 10 hours a day (excluding sleep time) over 7 consecutive days. The participants removed the device it only during showers, bathing, swimming or other water activities. The recording epoch was set to record by one minute. Hourly PA and SB were calculated using the minutes' data. For all subjects, absolute time in moderate-to-vigorous physical activity (MVPA), walking steps, kcals in energy expenditure and absolute time spent in SB were estimated using the device. Nonwear hour periods were de ned as 60 consecutive minutes of zero activity intensity count at 1 METs and 0 kcals in one hour [27].

Environmental measures
AP data and other environmental measures were selected including-air quality index (AQI), PM 2.5, and average daytime temperature (°C). Hourly AQI and PM 2.5 data were provided by the Beijing Municipal Ecological Environment Bureau. The data collection site is Wan Liu was the Hai Dian district site, which is approximately 5 km from Tsinghua University. The air quality measurement data we collected are concurrent with PA measurements. Daytime temperature was retrieved from the China Meteorological Administration.
The AQI is an index for reporting daily air quality [28]. It indicates you how clean or polluted the surrounding air is, and which associated health effects might be of concern for you. The AQI focuses on health effects that may be experienced within a few hours or days after breathing polluted air. The China Meteorological Administration calculates the AQI using the following ve major air pollutants: groundlevel ozone, particle pollution (also known as particulate matter), carbon monoxide, sulfur dioxide, and nitrogen dioxide.

Statistical Methods
Descriptive statistics were computed including means, SD, and percentages. All descriptive statistics were compared for the characteristics of the overall sample. Chi-square tests were conducted to compare categorical variables. ANOVA tests and t-test were used to compare for continuous variables. One-way repeated measures ANOVA tests were conducted to compare the differences in hourly time between 7 am and 11 pm. Adjusted linear individual xed-effect regressions were performed to reveal the associations between the accelerometer data and AP data. The data of participant controlled for age, gender, BMI, selfrated physical health, self-rated mental health, smoking, drinking, temperature, and temporal order for participants.
The key independent variables were AQI and PM 2.5 during this time. We performed six levels (1)(2)(3)(4)(5)(6) in AQI and used 10 µg/m³ in PM 2.5 in the model analysis. Individual-level time-variant covariates and environmental measures including average daytime temperature were controlled for the aforementioned.
Each outcome variable was then analyzed using separate regression and were strati ed by speci c time (from 7 am to 11 pm).
Compared to the conventional pooled cross-sectional regression, individual xed-effect regression was selected due to its used of within-individual variations in hourly PA and SB to identify the impacts of AP concentration, thus removing potential omitted variable bias due to differences in time-invariant individual characteristics such as habits, and personal preferences. Table 1 presents the characteristics of the participants. Among the 340 student participants, accounting for more than two-thirds (70.6%) were male. Participants complied to wearing a wGT3X accelerometer for 42291 hours over 7 continuous days. The mean age of the participants was 18.4 (SD = 1.0). The mean participants' BMI was 21.6 kg/m 2 (SD = 3.1) where male BMI was signi cantly higher than females' (p < 0.001). Only 0.88% of participants were reported smokers and 2.9% was reported drinkers. The mean selfrated physiological health score was 5.5 (SD = 1.8) and the mean self-rated mental health score was 6.5 (SD = 1.9).    The PA and SB Variations. Table 3 presents the mean variations of participants' PA and SB in the study. As illustrated, there are large variations in PA and SB. For example, the mean minutes of the participant's one-hour MVPA was 3.5 (SD = 4.5). One-hour mean minutes of MVPA ranged largely from 1.6 (SD = 2.7) at 11 pm to 6.1 (SD = 8.3) at 5 pm in the participants (p < 0.001). The mean steps of the participant's one-hour walking was 394 (SD = 627). Table 1 shows that one-hour mean steps of walking ranged largely from 147 (SD = 303) at 11 pm to 710 (SD = 947) at 5 pm in the participants (p < 0.001). Similarly, the mean participant's energy expenditure in one-hour was 19.8 (SD = 39.1). One-hour mean energy expenditure in kcal ranged largely from 9.2 (SD = 19.5) at 11 pm to 38.0 (SD = 66.2) at 5 pm among the participants (p < 0.001). Whereas the mean minutes of participant's one-hour SB was 31.5 (SD = 19.7). One-hour mean minutes of SB ranged largely from 38.6 (SD = 18.2) at 11 pm to 25.4 (SD = 17.1) at 12 pm among the participants (p < 0.001). Impact of AQI on PA and SB. Table 4 shows the estimated effects of air quality index (AQI) on individual-level outcomes of hourly PA and SB using linear individual xed-effect regressions. AQI was found to be signi cant and negatively associated with one-hour PA among participants. A one level increase in AQI was linked to a signi cant reduction in minutes of one-hour MVPA, in steps of one-hour walking, and in kcals of one-hour energy  Table 1 (i.e., age, gender, BMI, drinking status, drinking status, self-rated physical health, self-rated mental health, temporal order for participants) and environmental variables (average temperature). * P<0.05; ** P<0.01; *** P<0.001

Descriptive Statistics
The impact of AQI on individual-level one-hour PA at a speci c time was different. AQI was found to be more negatively associated with participants' one-hour PA at 8 am, 4 pm, 5 pm and 7 pm. Speci cally, a one level increase in AQI was linked with a signi cantly reduction in steps of one-hour walking at 8 am, , respectively. However, AQI was found to be positively associated with participants' one-hour PA at 10 am and 3 pm. There was no signi cant relationship between AQI and one-hour SB among participants.
Impact of PM 2.5 on PA and SB.     Table 1 (i.e., age, gender, BMI, drinking status, drinking status, self-rated physical health, self-rated mental health, temporal order for participants) and environmental variables (average temperature). * P<0.05; ** P<0.01; *** P<0.001 The impact of PM 2.5 on individual-level one-hour PA at speci c time was also different. PM 2.5 was found to be more negatively associated with participants' one-hour PA at 8 am, 4 pm, 5 pm and 7 pm.
The impact of PM 2.5 on individual-level one-hour SB at a speci c time was also different. PM 2.5 was found to be more positively associated with participants' one-hour SB at 9 am, 11 pm, 5 pm and 7 pm. Speci cally, a 10 µg/m³ increase in PM 2.5 was linked with a signi cantly increase in minutes of one-hour

Discussion
The purpose of this study was to examine the impact of AP level on PA and SB among university freshman students in Beijing, China from November 2017 to April 2018 using objectively-measured PA and SB. Our study found a signi cantly negative relationship between AP and PA and a positive relationship between AP and SB. With a one level increase in AQI and a 10 µg/m³ increase in PM 2.5 , hourly total minutes of MVPA, walking steps and kcals of energy expenditure were signi cantly reduced. A 10 µg/m³ increase in PM 2.5 was associated with a signi cantly increase in SB among participants. The impact of AP on individual-level one-hour PA and SB behavior at a speci c time was different. To our best knowledge, this is the rst study to use objective methods to determine the effect of hourly AP on PA and SB. In addition, this is the rst study to estimate the impact of AP on PA and SB at a speci c time.
Our ndings on the negative relationship between AP and PA are consistent with existing literature [19,[30][31][32][33]. In our study, we found that a one-hour AQI increase one level was associated with a decrease by 9 walking steps in one hour. This study additionally found that a 10 µg/m³ increase in PM 2.5 was linked with reduction by 2 walking steps in one hour. Two previous U.S. studies linked one unit (< 10 µg/m³) monthly average PM 2.5 increase of AP to be associated with decreasing 0.46% leisure time PA using a cross sectional study from the US Behavioral Risk Factor Surveillance System (BRFSS) survey [31,34]. Evidence from our previous follow-up studies also found a one unit (44.72-56.6 µg/m³) increase in PM 2.5 to discourage outdoor PA 110.67 PASE scores among older adults in China [30] and to reduce 32.45 weekly MVPA among Chinese college students [19,35]. However, these previous studies were limited by the potential social bias of self-report measures of PA and therefore were have not been able to examine the hourly effects of AP on objective PA. Only two studies with objectively measured data have reported that the association between AP and PA. Consistent with this study, a study of 153 middle-age adult users of an exercise app reported that AQI increase was associated with participates' reduction in outdoor PA, such as running, biking, and walking [33]. With this current analysis, we can more precisely use accelerometers to estimate the impact of AQI and PM 2.5 on MVPA, energy expenditure, and steps rather than use an exercise app associated with PA. Inconsistent with this study, another study showed that PM 2.5 increase had no impact on PA in Beijing among 40 Han Chinese participants in the mean age of 31 years using GT3X accelerometers [36]. A possible explanation for this difference could be that the study had a relatively small sample size and could not account for differences among participants. Based on 340 participants' wGT3X accelerometers data, we can more con dently suggest that an increase in AQI and PM 2.5 increase was associated with a reduction of PA in MVPA, energy expenditure, and walking steps.
This study con rmed ndings from previous studies regarding the positive correlation between AP and SB [37][38][39]. This nding suggests that a one-hour 10 µg/m³ PM 2.5 increase was associated with an increase in SB by 0.045 minutes in one hour. Consistent with our previous research, an increase in AP concentration in PM 2.5 by one unit (81.16 µg/m³) was associated with an increase in total weekly hours of SB by 6.24 hours among a large sample (12,174) of university freshmen in China based on a cohort study survey [37]. To our knowledge, this is one of the rst studies to investigate the impact of AP on SB by hourly use objectively measured GT3X accelerometers. Because of con icting results in this emerging area and somewhat preliminary of this nding, additional investigations are necessary to fully explore the effect of AP on SB among different groups.
The impact of AP on individual-level PA and SB at a speci c time was different. This study is the rst to examine the association between AP on PA and SB at a speci c time. Stronger negative associations of AQI and PM 2.5 AP with MVPA, walking steps and energy expenditure in the morning before 8 am, at 4 pm, at 5 pm and at 7 pm were found. Similarly, stronger positive associations of one hour AQI and one hour PM 2.5 on SB at 8 am, 9 am, 11 am, and 7 pm were found. In this study, all participants were recruited from freshmen. Typically, a substantial proportion of freshmen do not have class before 8 am in the morning, and/or after 4 pm. Participants in this study could choose PA or SB behavior according to their own preferences of activities during leisure time. Similar time-speci c relationships between built environment and PA were found in the previous studies [40,41]. However, it is interesting that positive associations of AQI and PM 2.5 were found with MVPA, walking steps and energy expenditure in the morning at 10 am and 3 pm. Yet, negative associations of AQI and PM 2.5 were found with SB. This could be explained by the fact that there are classes, including physical education class, for a substantial proportion of a freshman students' schedule between 10 am in the morning and 3 pm in the afternoon. Freshmen often engage in PA when traveling to class (e.g. walking or bicycling to or from class) and may perform more exercise in physical education class, regardless of AP. Therefore, 10 am in the morning and 3 pm in the afternoon associations between AP and PA or SB observed in this study are logical. Thus, this study con rms that the impact of AP on PA and SB are different from patterns of in time-speci c associations. This nding is closer to the 'true' potential effects of AP on PA and SB.
First, the strengths of this study reside in its objectively measured PA-related behavior and precise reporting of data. Most existing studies on the impact of AP on PA and SB have used subjective methods, allowing for uncontrolled confounding bias due to self-report and limited frequency of PA data. Second, this is one of the rst studies to measure the impact of AP on PA and SB by one hour using objective methods. Third, this is the rst study to examine time-speci c results on the relationship between AP and objectively measured PA and SB. However, a few limitations to this study should be noted. First, on the one hand, we didn't monitor indoor air pollution and may have bias in this study. A subject's physical activity participation may not vary by air pollution if an individual is active indoors. On the other hand, all freshmen (the subjects of this study) in Tsinghua University live in the same campus, live in the similar dormitories, take similar transportation (e.g., bicycle), and have similar classrooms (6th teaching building in Tsinghua). The subjects sharing similar classrooms, transportation means, and dormitories may partially offset the disruption the in uence of indoor air pollution levels. Second, we could not identify the speci c types of PA and SB through using accelerometers to assess PA and SB. Third, all participants were recruited by a convenience sampling. Freshman students from one university cannot represent all university students in Beijing, China or nationwide therefore limiting the generalizability of the study's ndings. Future studies are warranted to produce more generalized estimates.

Conclusions
In conclusion, the present study aimed to examine the relationship between AP and hourly behavioral modi cation to relate objectively-assessed PA and SB among university freshmen in Beijing, China. A negative association between hourly AQI and PM 2.5 and hourly PA in minutes of MVPA, walking steps and energy expenditure among study participants were found. A positive relationship between hourly AQI and Trend for walking steps in speci c time in one day among the freshman.