key: cord-0317904-1f3nqg6k authors: Chow, K.; Han, Y.; Li, V. O.; Lam, J. C.; Lee, S.-l.; Wong, W.; Lau, Y.-l. title: RGC-TBRS Personal Air Pollutant Exposure versus Health Condition and Perception Pilot Study for Young Asthmatics in Hong Kong (2019/2020) date: 2021-06-25 journal: nan DOI: 10.1101/2021.06.22.21259358 sha: 238c8aa736c00b64563645502ad2ab2db822bf90 doc_id: 317904 cord_uid: 1f3nqg6k This is a report of the RGC-TBRS funded observational pilot study which examines the effects of personal exposures to three types of air pollutants, namely, PM1.0, PM2.5, and PM10, on personal health condition and perception of young asthmatics (aged 12 to 15) in Hong Kong. This is the first study to investigate the relationship between PM1.0 and FEV1 and FVC of young asthmatics in Hong Kong, based on personal exposures obtained from portable sensors. Our preliminary results show that a higher level of PM1.0, PM2.5 and PM10 would deteriorate the health conditions of young asthmatics in HK. All correlations between particulates and lung functions are significant and negative, including PM1.0 exposure vs. FEV1 (R2=12%; p=0.023), PM1.0 exposure vs. FVC (R2=15%; p=0.010), PM2.5 exposure vs. FEV1 (R2=13%; p=0.019), PM2.5 exposure vs. FVC (R2=16%; p=0.008), PM10 exposure vs. FEV1 (R2=14%; p=0.012), and PM10 exposure vs. FVC (R2=18%; p=0.005). Moreover, after accounting for covariates, including age, gender, body mass index (BMI), temperature, and relative humidity, we found a significant relationship between PM1.0 exposure vs. FVC (Coefficient=-0.1224; p=0.032), PM2.5 exposure vs. FVC (Coefficient=-0.1177; p=0.021), PM10 exposure vs. FEV1 (Coefficient=-0.0703; p=0.019), and PM10 exposure vs. FVC (Coefficient=-0.1204; p=0.006). Further, using the pilot study data, we have performed a power analysis to estimate the sample size for our follow-up main study. Based on the primary null hypothesis that personal PM exposure would not change the FEV1 and FVC of young asthmatics in HK, the lowest sample size that gives 80% power at a 5% significance level is 107. Hence, the sample size (or the total number of participated asthma subjects) expected for the follow-up longitudinal clinical study should be 125 (after adjusting for the non-compliance and withdrawal of subjects). Our pilot study has demonstrated the feasibility of research into the effects of personal air pollutant exposure on health condition and health perception. Our follow-up study will address the challenges identified in the pilot study, based on the proposed follow-up actions for subject engagement, data collection, and data analysis. Previous air pollution and public health studies conducted internationally and in Hong Kong (HK) showed that exposures to outdoor air pollutants are correlated with the health conditions of young asthmatics. Panel studies provided the preliminary evidence for the cumulative effect of air pollutants on health conditions over time. A recent review of over 30 relevant panel studies showed that the adverse effects of air pollutant exposures on health conditions are more pronounced in asthmatic children than in healthy children [1, 2, 3, 4] . However, three major gaps have been identified from the previous studies. First, most of these panel studies are limited to the US and European contexts. Second, previous exposure assessment methods tend to use aggregate rather than personal air pollutant exposure data [5] . Further, the relationship between air pollutant exposures and health perceptions of the asthmatics is often overlooked. This study aims to fill the above research gaps and investigate how personal air pollutant exposures will affect health conditions and perceptions. A pilot study was conducted during Sep 2019 -Oct 2020. 51 asthma subjects were shortlisted, and nine subjects participated in the study and finished their pilot runs. This report summarizes the results of the pilot study. The report is organized as follows. Section 2 details the data collection methodology and statistical analysis. Section 3 presents the preliminary results based on the pilot run data, compares the results with the previous literature, and calculates the sample size for our follow-up study. Section 4 lists the major challenges of the pilot study. Section 5 proposes recommendations for our follow-up study. Section 6 concludes the pilot study. The pilot study was conducted on a rolling basis during Sep 2019 -Oct 2020. The subject recruitment was conducted by the Department of Pediatrics in Queen Mary Hospital (QMH). The pilot study targeted young asthma subjects aged 12 to 17. As of mid-Oct 2020, 51 asthma subjects were shortlisted, and nine asthma subjects (aged 12 to 15) participated in the study All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in 3 and finished their pilot runs. After subject enrolment, demographic information was collected through a one-off background survey (Table 1a). During the pilot study, for each subject, personal air pollutant exposure, health condition, and health perception data were continuously collected with a portable air pollution sensor, three portable/wearable devices (e-inhaler, e- spirometer, and Mi-Band), and a set of validated health-related surveys (Tables 1b-1d) . * The e-inhaler records the following events, including "inhaler installed", "inhaler removed", and "medication". ** Each participant is required to perform three or more repeated tests during a short period in a day. Each spirometry test is marked with a quality assessment label, including "Good blow", "Blow out faster", "Blow out longer", "Don't hesitate", "Abrupt end", and "Don't start too early". The PM exposure and meteorology data were sampled every minute by the portable sensor. In the raw tabular data, 19% of the rows consisted of at least one missing PM value (PM1.0, PM2.5, or PM10). We used an iterative imputer (Bayesian Ridge Regression) [6] to fill in the missing PM values based on the relationship between one feature and the others, utilizing the available All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in Moreover, spirometry tests were performed by the study participants daily (at least three repeated tests during a short period). We only selected spirometry tests with "good blow" labels to reduce the potential measurement errors. Spirometry values were aggregated into daily maximum values. Further, we calculated the mean and standard deviation of the daily PM and spirometry values. Any values out of the normal range (mean ± 3 × standard deviation) were considered outliers and removed. After data pre-processing, we performed a preliminary analysis to examine the relationship between personal air pollutant exposures and personal health conditions and health perceptions. The key variables are listed as follows. Health conditions refer to objective assessments of one's own health. For health conditions, the primary outcomes are the two lung function indicators that have been widely used in previous medical studies: FEV1, which measures how much air one can exhale during a forced breath during the first second, and FVC, which measures the total amount of air exhaled during the FEV test [7] . In addition, the secondary outcome is the number of asthmatic medications received by the subject as recorded by the e-inhaler. Health perceptions refer to subjective assessments of one's own health and well-being. For health perception, the primary outcomes are the ACT and AQLQ scores measured weekly. The ACT score is calculated using a standardized self-reported questionnaire that measures how well the symptoms of asthma are controlled. The AQLQ score is calculated using a standardized self-reported questionnaire that evaluates the quality of life of asthmatic patients. For personal air pollutant exposures, PM2.5 and PM10 were selected due to their welldocumented adverse health impacts in HK [8] . Moreover, PM1.0 was included in our analysis, given that the adverse health impacts of PM1.0 may be stronger than PM2.5 [9] . Further, demographic information, including age (number of years), gender (male or female), and body mass index (BMI), and meteorology information, including temperature (degree Celsius) and relative humidity (%), were included as covariates [10] [11] [12] . BMI (kg/m 2 ) is calculated as the body weight divided by the square of the body height. All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in For each health condition and perception outcome, a univariate linear regression analysis was performed to examine the relationship between the personal air pollutant exposure variable and the outcome variable. A multivariate linear regression analysis was also performed to examine the statistical relationship after controlling for covariates. All variables were measured at the daily level (except for the three demographic variables). Single-pollutant models were used due to the collinearity between PM1.0, PM2.5, and PM10. Two-sided p-value < 0.05 was considered statistically significant. After data pre-processing, 44 daily observations were obtained. Table 2a shows the characteristics of the study subjects. Table 2b shows the descriptive statistics of the exposure and outcome variables after removing outliers, including the mean, standard deviation (SD), median, and interquartile range (IQR) values. The level of average PM exposure during the study period is relatively low, given that the study subjects often stayed at home (especially during the Covid-19 period) and indoor PM exposure is lower than outdoor levels in homes without strong PM sources. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in Our preliminary results show that a higher level of PM1.0, PM2.5 and PM10 would deteriorate the health conditions of young asthmatics in HK, as shown by a lower FEV1 and a lower FVC value. This is the first study to investigate the relationship between PM1.0 and FEV1 and FVC of young asthmatics in Hong Kong, based on personal exposures obtained from portable sensors. Results of our pilot study also provide insights on the proper effect size of our followup study. Following the previous studies on personal air pollutant exposure and lung function [10] [11] [12] , we have adjusted the regression coefficients based on one IQR change in PM2.5 exposure level. For one IQR change in personal PM2.5 exposure, the change in FEV1 and FVC was -0.137 and -0.260, respectively. We found that the effect size in our study, as represented by the adjusted regression coefficient, is larger than the previous studies (Table 4 ). Table 3 and (2) the adjusted R 2 value of the corresponding regression model without the PM exposure variable. We found that the Cohen's f 2 value ranges from 0.107 to 0.200. Then, using the effect size of 0.107, we performed a two-sided F test under a multivariate linear regression setting (three predictors to be tested and ten predictors in total) [14] to calculate the sample size that gives 80% power at a 5% significance level. We found that the lowest number of asthmatic subjects is 107. After accounting for a potential non-compliance and dropout rate of 10% to 15% (given that children in HK normally show much higher compliance), we determined that 125 asthmatic subjects are sufficient for our main study. Moreover, to obtain more rigorous statistical findings by comparing the difference between the asthmatic subjects and healthy subjects, 125 healthy subjects will also be enrolled in our main study in a 1:1 ratio. Our follow-up main study will collect more longitudinal health condition and perception and air pollutant exposure data from 125 asthmatic subjects and 125 healthy subjects. This longitudinal dataset will help us better investigate the statistical relationship between personal air pollutant exposure and health condition and perception, by utilizing more advanced statistical modelling to account for (1) the repeated measures of health outcomes and All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in (2) the lag effects of personal air pollutant exposure. We expect to obtain more rigorous statistical findings in our upcoming main clinical study. As hospitals and clinics have focused on combatting COVID-19, the recruitment of subjects has been adversely affected. The Hospital Authority has required all non-urgent cases to stay away from hospitals to minimize risks. Additional challenges in data collection and statistical analysis are listed below (See solutions to overcome such challenges in Section 5). Notable missing values have been observed in the data collected from different types of sensors. During the pilot study, the missing data can be due to human errors (e.g., the subject may forget to perform the required actions) or data transmission errors (e.g., the sensor/phone may fail to connect to the internet). The data collection process is yet to be fully automated and monitored. In the pilot study, the data collected from the e-inhaler, e-spirometer, and Mi-Band were downloaded and processed manually before further analysis. Moreover, the questionnaires are yet to be automated and conducted online. Due to the small number of data points, advanced statistical models for longitudinal data, such as linear mixed-effect models, are yet to be utilized. The major challenges of the pilot study are interlinked. By improving our data collection procedure and engaging with more participants, we expect to establish a high-quality longitudinal dataset for advanced statistical analysis. More specifically, to overcome these challenges, our follow-up work can be divided into two main directions: (1) improving data collection and analysis and (2) increasing and mobilizing subject engagement. All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in 13 We are building an integrated data management platform to automatically collect, process, monitor, and manage the data across all sources, utilizing a cloud computing server equipped with strong measures for user privacy and data security. So far, the portable sensor data collection procedure has been successfully integrated into the data management platform. Moreover, an alert system, as a part of the data management platform, will be deployed to identify and fix data synchronization and transmission issues, and issue missing data alerts to (1) inform the study team to take action and (2) remind the subjects to follow the data collection procedure closely. Calibrating air pollution data can reduce the measurement errors and allow the subjects to better understand the risks of his/her exposure. In the portable sensor, PM pollutants are measured by a laser particle counter. The PM sensor can be calibrated based on the raw data (bin counts for a range of particle size bins) while accounting for meteorology factors [15] . Our sensor calibration can be performed seasonally to capture the seasonal variation of air pollution in HK (for more details, please refer to our sensor calibration report). We are integrating the e-surveys into the UMeAir App and the data management platform, based on a separate online e-survey system developed by us previously. The UMeAir App will motivate the subjects to fill in the surveys on time with gentle musical alerts. We will also look for more incentive and empowerment measures to our young participants, along the line of good citizenship recognition/differentiation; such as running smart competitions for our participants with chances of visiting our new HKU innovation facilities online, opportunities of visiting online new high-tech start-ups in HK or Israel, inviting our participants to our UMeAir Club as VIPs, and awarding certificates of green, healthy and smart citizenship to our VIPs, in order to improve the quality of questionnaire data collection. After collecting more longitudinal data with higher data integrity and fewer missing values, we aim to improve the current statistical methodology. We will adopt advanced statistical models (such as the linear mixed-effect models) to account for the repeated measurements and the lagged effects of air pollutant exposure, while controlling for important confounders and autocorrelation in longitudinal data. We will also address non-linearity and multicollinearity if All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted June 25, 2021. ; https://doi.org/10.1101/2021.06.22.21259358 doi: medRxiv preprint any are found in our follow-up study. We expect to obtain more rigorous statistical findings in our follow-up study. During the pilot study, considerable data gaps for most of the sensors and measurements were identified. While data collection automation will help, increasing the accessibility of the sensors and reducing technical barriers of the subjects can improve the engagement of the subjects and thus the integrity of the data. We are working closely with QMH staff to produce a series of instructional videos for the usage of different sensors for the study and provide timely technical assistance to both the staff and subjects. In order to provide an effective control group, we are currently proceeding with the recruitment of healthy subjects for the pilot study. However, due to the suspension of schools as a result of the ongoing COVID-19 pandemic, there are difficulties in the recruitment of students. We are working closely with schools and starting to install air pollution sensors in schools. Meanwhile we shall recruit the healthy subjects from schools. This report provides a summary of the current progress made in the TBR Air Pollution Clinical and Quality of Life Study. We determine that all correlations are significant and negative, including PM1.0 exposure vs. FEV1 (R 2 =12%; p=0.023; Figure 1a Figure 1f ). Moreover, after accounting for covariates, including age, gender, BMI, temperature, and relative humidity, we found a significant relationship between PM1.0 exposure vs. FVC (Coefficient=-0.1224; p=0.032; ; Table 3d ). Further, using the pilot study data, we have performed a power analysis to estimate the sample size for our follow-up main study. Based on the primary null hypothesis is that personal PM exposure would not change the FEV1 and FVC of young asthmatics in HK, the lowest sample size that gives 80% power at a 5% All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted June 25, 2021. ; https://doi.org/10.1101/2021.06.22.21259358 doi: medRxiv preprint significance level is 107. Hence, the sample size (or the total number of participated asthma subjects) expected for the follow-up longitudinal clinical study should be 125 (after adjusting for the non-compliance and withdrawal of subjects). Our pilot study has demonstrated the feasibility of research into the effects of personal air pollutant exposure on health condition and health perception. Our follow-up study will address the challenges identified in the pilot study, based on the proposed follow-up actions for subject engagement, data collection, and data analysis. Association between air pollution and lung function growth in southern California children Residential traffic-related pollution exposures and exhaled nitric oxide in the children's health study Determinants of the spatial distributions of elemental carbon and particulate matter in eight southern Californian communities Association of improved air quality with lung development in children Clearing the air: a review of the effects of particulate matter air pollution on human health mice: Multivariate imputation by chained equations in R All rights reserved. No reuse allowed without permission. perpetuity preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted Using the lower limit of normal for the FEV1/FVC ratio reduces the misclassification of airway obstruction Health Effects of Air Pollutants -Respirable and Fine Suspended Particles (PM10 and Is PM1 similar to PM2.5? A new insight into the association of PM1 and PM2.5 with children's lung function Personal exposure to fine particulate matter, lung function and serum club cell secretory protein Personal and ambient air pollution exposures and lung function decrements in children with asthma Personal Exposure to PM2.5 Oxidative Potential in Association with Pulmonary Pathophysiologic Outcomes in Children with A practical guide to calculating Cohen's f2, a measure of local effect size, from PROC MIXED Statistical power analyses using G* Power 3.1: Tests for correlation and regression analyses Characterising low-cost sensors in highly portable platforms to quantify personal exposure in diverse environments We acknowledge the assistance of Miss Jennifer Lam, Research Nurse, Department of Pediatrics, Queen Mary Hospital, Hong Kong, in training recruited asthma subjects in using the medical and air pollution sensor devices, and in collecting clinical trial data. We also thank Mr. Andong Wang for his research assistance in this clinical trial project. All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in