Fine Particulate Matter Exposure and Blood Pressure: Evidence from a Chinese Large Multiple Follow-Up Study

Long-term exposure to ne particulate matter (PM 2.5 or FPM) may cause adverse effects on cardiovascular diseases. However, evidence that whether improved air quality can decrease blood pressure (BP) in humans is still needed from a large population study. Our study aimed to investigate the association of population ambient PM 2.5 exposure with the blood pressure (BP) changes in China with implementing the Action Plan on Air Pollution Prevention and Control. A total of14,080 participants who had at least two valid visits were adopted from the China Health and Retirement Longitudinal Survey (CHARLS) during 2011–2015. Their long-term PM 2.5 exposure was assessed at the geographical level of a regular 0.1° × 0.1° grid over China. A mixed-effects regression model was used to assess their associations. The robustness and homogeneity of the association were tested via sensitivity analyses. The results revealed that each reduction of 10 µg/m 3 in the 1 year-mean PM 2.5 concentration (FPM 1Y ) was associated with a decrease of 1.24 (95% condence interval [CI]: 0.84–1.64) mmHg of systolic BP (SBP) and 0.50 (95% CI: 0.25–0.75) mmHg of diastolic BP (DBP), respectively. A robust association was observed between the long-term reduction of PM 2.5 and decreased BP in the middle-aged and elderly population in China. These ndings were further conrmed by a non-linear regression model. We concluded that air pollution health. Our study provided scientic the air pollution


Introduction
High blood pressure, also known as hypertension, is blood pressure that is consistently higher than what is considered normal. There are 2 types of blood pressure measures: systolic and diastolic. Systolic blood pressure is the pressure in the arteries when the heart beats, while diastolic pressure is the pressure in the arteries when the heart rests. Normal systolic blood pressure is less than 120 millimeters of mercury (mmHg), and normal diastolic blood pressure is less than 80 mmHg, together described as 120/80 mmHg (Desai et al., 2020). Hypertension has been well recognized as a major risk factor for cardiovascular diseases (CVDs) (Forouzanfar et al., 2017). Hypertension is also a leading risk factor of mortality and disability globally (Stanaway et al., 2018). In China, the prevalence rate of hypertension among adults over 35 years of age is 32.5%, resulting in various adverse health outcomes and a heavy nancial burden (Lewington et al., 2016). The prevalence of hypertension in Chinese adults has been increasing Bao and Wang, 2020;Ma et al., 2021). Previous studies reported that highdensity lipoprotein cholesterol, triglycerides, body mass index, alcohol dependence, insomnia, educational level, diabetes, smoking, stress, viral infection, and age are risk factors for developing hypertension (Hay et al., 2020;van Oort et al., 2020). It also has been reported that population blood pressure (BP) can be  (Rao et al., 2018). However, few studies have assessed whether blood pressure is decreased by a cleaner ambient environment. The Chinese population has been exposed to air that is severely polluted by PM 2.5 since the 2000s (Xue et al., 2019b). In 2013, the well-known Air Pollution Prevention and Control Action Plan (APPCAP) was released by the Chinese government as the rst national strategy on air pollution control (The State Council, 2013), which has markedly improved the air quality (Wang et al., 2019a). However, there is a lack of evidence to assess whether BP can be reduced by improved ambient air quality.  of air pollution prevention and control policies to improve air quality, which has created an observational quasi-experimental scenario to assess whether improving air quality is associated with a decrease in BP (Xue et al., 2019a). Therefore, identifying the relationship between long-term PM 2.5 exposure and BP would provide a policy-making reference for other countries to balance economic development with human health. We hypothesized that a reduction in the ambient PM 2.5 concentration would be associated with a decrease in BP. Herein, we performed a quasi-experimental study of the relationship between reduction in PM 2.5 and changes in BP based on a national survey conducted before and after the clean air policies in 2011 and 2015.

Population Recruitment
We obtained population health data from the China Health and Retirement Longitudinal Survey (CHARLS), which is publicly available at http://opendata.pku.edu.cn. The details of this project have been documented previously (Zhao et al., 2014). Brie y, to ensure the national representativeness of the project, the study population was selected from 28 provinces (150 counties or districts) via multistage probability sampling in China ( Figure S1). Face-to-face interviews were performed every 2 years using a standard questionnaire to collect basic information on socio-demographics (home address, age, gender, and educational level), energy-use characteristics for cooking and heating, health-related behaviors (smoking and drinking), and health status (self-reported general health and medicine usage). smoking and drinking history, marital status, disease history, cooking energy, indoor temperature, antihypertension drugs (medication) were extracted from the questionnaire, and utilized as covariates to control for confounders. Missing data in covariates included in our study were imputed by multiple imputation method. In our study, the population were chosen by following these criteria: complete information on BP and at least two valid records of BP. Finally, a total of 14,080 participants were included to con rm our hypothesis.

Ambient PM 2.5 Concentrations
Distribution of PM 2.5 were estimated using a hindcast approach based on a two-stage machine learning model. This approach integrated the data of historical emissions and satellite remote sensing measurements. This yielded daily PM 2.5 concentrations across a regular 0.1 × 0.1° grid over China, from 2000 to 2016. The detailed description of the estimate method has been published previously, with the validation results that the generated concentrations were highly correlated with the ground observations at the monthly (R 2 = 0.71) and annual (R 2 = 0.77) scales (Xue et al., 2017). The home address of CHARLS participants can be obtained only at the city level for reasons of con dentiality. Therefore, the PM 2.5 concentration data were rst converted into city-level and monthly averages, and then linked to the CHARLS respondents according to their spatiotemporal coordinates. The average PM 2.5 concentrations in 1 and 2 years before the visiting day were calculated as the long-term exposure value and denoted as FPM 1Y and FPM 2Y , respectively.

Statistics Analyses
We used the median, interquartile range (IQR), or standard deviation to describe the distributions. To explore the changes in BP before and after the policy intervention, we summarized the population level mean values between the CHARLS waves. Linear mixed-effects regression models with random effects were employed to investigate the associations between ambient PM 2.5 concentration and BP. The random terms were used to control for two clustering effects in individual, and community levels. We incorporated a spline term with three degrees of freedom into the regressions to describe the nonlinear effects of ambient temperature. We estimated the effects of a 10 µg/m 3 decrease in ambient PM 2.5 concentration on BP. To examine the robustness of the association between ambient PM 2.5 and BP (SBP and DBP), we controlled different groups of confounders resulting in the following four different models: (I) Model I: BP = β × PM 2.5 + β 1 × CF 1 + β 2 × CF 2 + γ 1 (S) + γ 2 (H). This model incorporated xed terms with the β coe cients of PM 2.5 , β 1 β 2 of potential confounders (CF 1−2 : city and medication), as well as a random intercept for each subject γ 1 (S) and each household γ 2 (H). We controlled for medication status and city location. The participants who were taking anti-hypertensive medicines or had ever taken them during the visiting period were classi ed into those taking medicines.
We conducted a strati cation analysis to examine whether the association between ambient PM 2.5 and BP was modi ed by the following factors: age, gender, education level, residence, marital status, smoking, alcohol drinking, and taking medicine. The statistical signi cance of the effect modi cation was tested by analysis of variance between the Model III and the modi ed model. The P-value was adjusted by the false discovery rate (FDR) method, which is a way to allow inference when diverse tests are being conducted. Compared with Bonferroni multiple testing method, FDR corrected the P-value in a milder way by means of controlling the proportion of false/true positives to a certain range. The above analysis assumed that BP and explanatory variables showed a linear correlation. To verify this, a generalized additive mixed model with a random intercept was established to explore exposure-response relationship between the ambient PM 2.5 and BP by replacing the linear term of PM 2.5 with a set of penalized spline functions: BP ~ g(PM 2.5 ) + β 1−10 × CF 1−10 + γ 1 (S) + γ 2 (H) where g is the smoothing spline term. All statistical analyses were conducted in R (version 3.5.3; The R Foundation for Statistical Computing, Vienna Austria). We used two-sided statistical tests, and a P-value < 0.05 was considered signi cant.

Population Characteristics
Their general characteristics of the included 14,080 participants for data analysis are summarized in Table 1. Approximately, 65% of the participants were from rural area and half of them were female. The proportions below the elementary and elementary and middle education levels were similar between the three visits. Overall, ~10% of them had education level above a middle education. More than 80% were married and lived together. It revealed that most participants did not frequently have tobacco smoking or drink alcohol. Most of the participants lived in their own houses and more than half lived in a multi-story building with a telephone facility, moderate household temperature, and moderate untidiness. Less than 30% of the participants took anti-hypertensive drugs.    In addition, SBP generally increased with age, and DBP increased at the beginning and then decreased with age ( Figure 1). Although the curves in different years were similar, that derived from the CHARLS 2015 (i.e., the wave after the clean air actions) was lower for most age groups. Because the curves were derived from cross-sectional information without adjustment for confounders, they were used to display the data and do not show the variation in BP with age among Chinese adults.  (Figure 2A). DBP increased approximately linearly with the increase of FPM 1Y without an obvious peak ( Figure 2B). The slope of the regression curve of SBP with increasing FPM 1Y was larger than that of DBP.

Strati ed Analysis
A strati ed analysis was performed to investigate the association between FPM 1Y and BP under different levels or grades of various confounders (i.e., medication, age, residence, gender, marriage, smoking, drinking alcohol, and education). The estimated associations between PM 2.5 and BP did not vary signi cantly on inclusion of most of them. For DBP, the association between FPM 1Y and DBP was greater in the urban residents than in the rural residents without adjusting P-value by the FDR method ( Figure 3). Also, a greater association was found in married participants than in single participants. These results suggest that residence and marital status may modify the association between FPM 1Y and DBP. However, no signi cant modi cation effects of various confounders were found in the association between FPM 1Y and DBP when adjusting P-value by the FDR method. For SBP, the association between FPM 1Y and SBP was greater in female than in male without adjusting by the FDR method ( Figure 3). Similar to DBP, no signi cant modi cation effects of various confounders were found when adjusting P-value by the FDR method. The details of the results are provided in Table S2 (SI).

Discussion
In this study, we investigated the effect of PM 2.5 exposure on BP using the on-going large population Overall, these cohort or cross-sectional studies provide certain evidences about the positive associations between PM 2.5 and BP. Overall, the ndings in our study were consistent with those in the previous reports. However, our study provided more valid evidence in consideration of the study method and exposure scenario. It has been well known that repeated-measurement studies have a stronger ability to verify causality than cross-sectional studies, which has been widely used as a special study design in environmental epidemiology. However, it is di cult to conduct such studies on large- We also adopted a nonlinear regression model to verify that BP increased linearly with the increase of . In our study, the average age, as well as the proportions of participants with high frequencies of tobacco smoking and drinking alcohol, were larger in single participants than those in married participants. In other words, age and living habits (smoking and drinking alcohol) may play a more important role in single participants than in married participants. It suggested that the BP of married population may be more sensitive to the PM 2.5 exposure. However, these results cannot be well explained using the current data and more evidence from additional studies are still needed.
Our study has two important limitations. First, the PM 2.5 exposure assessment was based on historical estimates; we did not conduct exact personal exposure measurements, nor did we have information on indoor air quality. This uncertainty in the PM 2.5 concentration could lead to exposure misclassi cations and bias the results. Similarly, coarseness in the exposure assessment due to the lack of addresses could also lead to exposure misclassi cation, despite that previous studies used a similar method, e.g., a six U.S. cities prospective cohort study measured air-pollution data in each community at a centrally located air-monitoring station (Dockery et al., 1993). Second, the underlying mechanisms for the modifying effects of population residence and marital status cannot be well explained using the current information. However, to the best of our knowledge, our study examined the largest population to investigate the effect of PM 2.5 on BP using the repeated-measurement study design conducted in China.
Particularly, our study results provided the direct evidences on the protective effects of the improved air quality on the blood pressure. Above all, our conclusion warrants further studies for con rmation.

Conclusions
We concluded that reducing long-term PM 2.5 exposure could decrease BP among middle-aged and elderly residents in China. Our ndings provide an important perspective of improve the cardiovascular health from the air pollution control. It will provide a reference for related air quality improvement.

Declarations
Funding: None Changes in PM2.5 and BP. Left panel, distribution of SBP, DBP, and PM2.5 by CHARLS wave; right panel, age-speci c distribution of the waves; smoothed curves for BP and age were derived using the spline approach.