Association of PM 2.5 Reduction with Improved Kidney Function: A Nationwide Quasiexperiment among Chinese Adults

Background. Increasing evidence from human studies has revealed the adverse impact of ambient fine particles (PM 2.5) on health outcomes related to metabolic disorders and distant organs. Whether exposure to ambient PM 2.5 leads to kidney impairment remains unclear. The rapid air quality improvement driven by the clean air actions in China since 2013 provides an opportunity for a quasiexperiment to investigate the beneficial effect of PM 2.5 reduction on kidney function.Methods. Based on two repeated nationwide surveys of the same population of 5115 adults in 2011 and 2015, we conducted a difference-in-difference study. Variations in long-term exposure to ambient PM 2.5 were associated with changes in kidney function biomarkers, including estimated glomerular filtration rate by serum creatinine (GFR scr) or cystatin C (GFR cys), blood urea nitrogen (BUN), and uric acid (UA).Results. For a 10  μg/m 3 reduction in PM 2.5, a significant improvement was observed for multiple kidney functional biomarkers, including GFR scr, BUN and UA, with a change of 0.42 (95% confidence interval [CI]: 0.06, 0.78) mL/min/1.73m 2, -0.38 (-0.64, -0.12) mg/dL, and -0.06 (-0.12, -0.00) mg/dL, respectively. A lower socioeconomic status, indicated by rural residence or low educational level, enhanced the adverse effect of PM 2.5 on kidney function.Conclusions. These results support a significant nephrotoxicity of PM 2.5 based on multiple serum biomarkers and indicate a beneficial effect of improved air quality on kidney function.


Introduction
Chronic kidney disease (CKD) has a significant and increasing impact on the world's population (1).In 2017, the recorded cases of CKD were 697.5 million with an average prevalence of 9.1% globally, resulting in 35.8 million disability-adjusted life-years (DALYs) (2).The symptoms of CKD are usually latent initially, before becoming more apparent manifesting as cardiovascular disease (CVD), edema, bone disease, anemia, and nerve damage.Finally, this leads to kidney failure, known as end-stage kidney disease (ESRD) when regular dialysis treatment or a kidney transplant is the only options for survival (1).Declining kidney function is a key characteristic and diagnostic indicator which identifies the onset and development of CKD.Traditional risk factors of CKD and kidney dysfunction, such as diabetes and hypertension (1), cannot fully explain the geographic heterogeneity of the disease, and it is suggested that other drivers including air pollution may account for the variation (3,4).
The air pollutant causing most health concerns is the particulate matter with aerodynamic diameter smaller than 2.5 μm (PM 2.5 ) which has been well documented for its adverse cardiorespiratory and metabolic effects (5).It has been proposed that the underlying biological mechanisms, such as systemic inflammation, oxidative stress, and vascular endothelial dysfunction, may potentially damage distant organs including the kidneys, which is a highly vascularized organ of the human circulatory system (6).
Increasing evidence from human studies has revealed associations between air pollution and declining kidney function indicators such as the prevalence (7)(8)(9), incident and progression of CKD (8,10), and decreased estimated glomerular filtration rates (eGFR) (7,8,(11)(12)(13); although, the findings remain inconsistent.A study which included 1.1 million adults from the US Medicare program reported that the annual concentration of county-level PM 2.5 was positively associated with the prevalence of diagnosed CKD (9).However, this conclusion has not been repeated in three other large cross-sectional studies (7,8,14).the evidence from cohort studies is consistent that long-term exposure to PM 2.5 is significantly associated with the incident and progression of CKD (8) and the risk of eGFR decline (10,12).To date, no evidence has been provided to examine if PM 2.5 reduction could beneficially impact kidney function in human.
The Clean Air Action Plan, initiated in 2013 by the Chinese government, was a bold nationwide policy aiming at tackling the severe air pollution problem in China that was leading to about 1 million premature deaths every year (15).The plan consists of ten key measures including some important policies as reducing emissions from industrial facilities and vehicles and improving the efficiency of fuel usage (16).As a consequence, annual PM 2.5 concentrations in mainland China have decreased significantly by 32%, from 67.4 μg/m 3 in 2013 to 45.5 μg/m 3 in 2017 (15).This rapid change provides an opportunity of designing quasiexperiments to investigate the beneficial effect of air pollution reduction on the human health in the Chinese population (17)(18)(19).Based on repeated nation-scale surveys launched before and after the Clean Air Action Plan in 2011 and 2015 on the same population of 5115 adults, we conducted a differencein-difference study to investigate the causal relationship between chronic exposure to ambient PM 2.5 and changes in kidney function.

Population Data.
The population in this study was based on the project "The China Health and Retirement Longitudinal Study (CHARLS)," which provided a public database (http://opendata.pku.edu.cn/) with a wide range of information from socioeconomic status to health conditions in Chinese people aged 45 and over (20).To ensure sample representativeness, ~20,000 participants were recruited based on a complex four-stage sampling approach across the country.A detailed introduction and protocol of the project is provided in the CHARLS document (21).Briefly, three waves of repeated surveys were launched in the years 2011, 2013, and 2015, and most of the surveys were conducted during summer season (Figure S1).The analysis of this study relied on the data collected from waves 1 and 3, because blood samples were not collected in wave 2. Demographic background and social economic information of all participants were collected using followup questionnaire qualified enumerators.Blood samples were also collected by well-trained staff.

Kidney Function Measurement.
The evaluation of kidney function was based on four clinical biomarkers measured in fast blood samples, namely, serum creatinine (SCR), cystatin C (CYS), blood urea nitrogen (BUN), and uric acid (UA).Samples were sent to a central national laboratory and stored -80 °C until assay.Strict standards were applied from sample transportation, storage, to the measurement and quality control, with details in supplement (S1.1).
Decline in glomerular filtration rate (GFR) was deemed as the golden standard for the clinical diagnosis of CKD (1).In this study, we applied the two equations to calculate eGFR by SCR (GFR scr ) or CYS (GFR cys ), based on age and sex (details in S1.2) (22).In a comparative study (21) for Chinese adults, the utilized equations were proofed to outperform other eGFR approaches suggested by the Chronic Kidney Disease Epidemiology Collaboration (23).BUN and UA can also be utilized as biomarkers of kidney impairment, because these two biomarkers reflect the concentrations of waste products generated from protein and purine metabolism, respectively.The elevated concentrations of BUN and UA are usually seen in patients with reduced GFR (24,25).Since BUN and UA may also be indicative for other diseases (e.g., gouty arthritis), in this study, they act as secondary biomarkers for kidney function.Major conclusions from our analyses should be drawn from the eGFR results.
2.3.Exposure Data.Similar to our previous studies, we evaluated the long-term exposure to PM 2.5 and temperature for CHARLS subjects according to the reanalyzed environmental database (26,27).Details on exposure assessment are documented in supplement (S1.3).

Study Design and Statistical
Analyses.This study applied a difference-in-difference method to examine the causal effect of PM 2.5 on kidney function.The same approach has been applied in our previous analyses (17,19) and briefly introduced in supplement (S1.4).For this study, the difference-in-difference analysis was parameterized using the following equation: where i denotes the individual index; ΔPM 2.5, i denotes the temporal change in the exposure level from pre-to postclean air actions for the i -th adults; ΔBiomarker i denotes the corresponding temporal change in a targeted biomarker; Δx i denotes the temporal changes in the longitudinal covariates 2 Health Data Science n the other hand, O (i.e., inconstant variables, including marriage, smoking, drinking, indoor temperature maintenance, cooking energy type, and body weight, as shown in Table 1); x i denotes the baseline covariates (i.e., constant variables, including residence, sex, and education); and β and γ 1 and γ 2 denotes the corresponding regression coefficients.For categorical longitudinal variables, Δx i denotes a new categorical variable, coded by combination of measurements in the two waves (for example, between the two surveys, a person who quit smoking was coded as yes-no, and a person who started to smoke was coded as no-yes); for continuous longitudinal variables, Δx i denotes the difference between the two measurements.Given the possibility that the studied subjects were not completely randomly distributed along different levels of ΔPM 2.5 , we applied the inversed probability weights, which were derived using the R package ipw.The probability of ΔPM 2.5 was estimated using a regression model that incorporated covariates of residence, sex, education, age, BMI, temperature variation, body weight change, and an indicator of regional developmental level and gross domestic product per capita.The effect of PM 2.5 on kidney function was evaluated as (10 × β), i.e., the change in biomarkers for a 10 μg/m 3 increment of PM 2.5 in the longterm exposure.Δx i γ 1 aims to control for the potential confounding effects from the longitudinal changes in the study population, such as the lifestyle (drinking and smoking) changes.x i γ 2 aims to control for the heterogenous temporal trend in the kidney function, which is assumed to progress in temporal patterns different by demographic characteristics (e.g., educational level).Since ΔBiomarker i for each of the four biomarkers was distributed in bell shape (Figure S2), there was no clear evidence to model the outcomes in transformed scale.
In sensitivity analyses, we first examined whether the estimated effect varied with model settings: (1) different choices of adjusted covariates and (2) incorporation of the inversed probability weights or not.Second, we explored how the estimated effect was heterogeneous between different subpopulations using interaction analyses.Third, linearity of the effect was tested with details in supplement (S1.5).Fourth, we explored how the estimated effect of PM 2.5 was varied between subjects with different baselines of kidney function.The baseline-varying model, which is similar to the nonlinear model, has been applied in our previous study (28) and is documented in the supplement (S1.5).Fifth, we explored whether the estimated associations were sensitive to the potential heterogeneity between different survey waves or the skewed distributions of the biomarker levels.To test that, we first normalized the measurements (i.e., normalized biomarker i = ðbiomarker i − meanÞ/SD) within each wave and then utilized their differences as new dependent variables to reexamine the effect of PM 2.5 .Sixth, we conducted two sensitivity analyses to evaluate the exposure measurement errors caused by using city-level average to assess the PM 2.5 exposure.We applied a welldeveloped bootstrap method (Supplement S1.6), which has been applied in our previous studies, to quantify the exposure measurement error (17,29).We also further lowered the error by deriving city-level PM 2.5 concentrations specifically for urban or rural area (Supplement S1.7) and reestimated the associations.Finally, we also conducted post hoc analyses to examine whether the associations were sensitive to the choice of time window for exposure assessment.We repeated our main models using alternative city-level PM 2.5 concentrations averaged within 1, 2, 3, or 4 years preceding the survey time.All the statistical analyses were performed using R (version 3.4.1),and the significance level was set as p < 0:05.

Associations.
To illustrate the design of a difference-indifference study, we first conducted a preliminary analysis (Figure S2).The actual changes in PM 2.5 exposure (ΔPM 2.5 equals the concentration of PM 2.5 in 2015 minus the concentration in 2013) varied in different cities.Subjects were divided into two groups, namely, those who lived in the areas with a ΔPM 2.5 below its upper quartile (ΔPM 2:5 < −5:12 μg/m 3 ) as the treatment group, referring to a more efficient effect of the clean air actions and the rest as the control (Figure S2, upper panel).Because different magnitudes of PM 2.5 reduction were mostly driven by emission-control policies, the between-group difference in the temporal change in a biomarker (ΔBiomarker) could be utilized to reveal the policy's effect.The lower panel of Figure S2 suggests the intervention was associated to a positive change in GFR scr (0.22 mL/min/1.73m 2 ) and GFR cys (0.88 mL/min/1.73m 2 ) but a negative change in BUN (-0.14 mg/dL) or UA (-0.16 mg/dL).Because the reduction in GFR scr or GFR cys , and the increment of BUN or UA suggests a kidney impairment, the results consistently revealed the benefit of PM 2.5 reduction on kidney function.

Health Data Science
Considering the PM 2.5 reduction as continuous, the between-group comparison of ΔBiomarker could be converted into a regression analysis (i.e., the difference-indifference model), which enable a quantitative examination on the association between ΔBiomarker and ΔPM 2.5 , as shown in Figure S3.According to the fully adjusted models (Table 2), an increment of 10 μg/m 3 in PM 2.5 was associated to a change of -0.42 (95% CI: -0.78, -0.06) mL/ min/1.73m 2 , 0.02 (-1.16, 1.20) mL/min/1.73m 2 , 0.38 (0.12, 0.64) mg/dL, and 0.06 (0.00, 0.12) mg/dL for GFR scr , GFR cys , BUN, and UA, respectively.The estimated effects of PM 2.5 were not sensitive to different model settings.The findings based on GFR scr , BUN, and UA consistently suggested a significantly adverse effect of PM 2.5 exposure on kidney function.The large uncertainties in the estimated effect on GFR cys might be due to its relatively small sample size (Table 1).

Sensitivity Analyses.
Figure 1 presents the results from subgroup analyses.We found that the estimated effects were not significantly varied within the most of subpopulation indicators, except for the variables related to the socioeconomic status and age.Compared to the urban residents, rural people were more susceptible to the effect of PM 2.5 on GFR scr (p = 0:06), GFR cys (p = 0:01), and BUN (p = 0:04 ).The association between PM 2.5 and GFR scr was also varied with education level at marginal significance (p = 0:05), with enhanced susceptibility in the lowest educated subpopulation.We also observed a trend that aging could enhance the associations between PM 2.5 and GFR scr (p = 0:11) and GFR cys (p = 0:05).Additionally, directions of the estimated associations reported by the nonlinear models were in consistent with the results from the main (linear) models (Figure S4).Furthermore, the baseline-varying effect models suggested that the adults with normal kidney function (e.g., GFR scr > ~80 mL/min/1:73m 2 or BUN < 20 mg/dL), who might be exposed to less competing risk factors (e.g., alcohol usage), could be more susceptible to the toxic effect of PM 2.5 , compared to those with poor kidney function (Figure S5).Finally, after normalizing the biomarkers, we still observed significant effect of PM 2.5 on GFR scr , BUN, and UA (Figure S6).
We utilized a bootstrap method to evaluate how the exposure errors influenced the association estimates.The bootstrapped results were presented in Table 2. Generally speaking, the bootstrapped results were statistically comparable with the estimates before correcting the measurement errors.We also reevaluated the associations by urban-orrural-specific exposures, and the reestimated results were statistically comparable with those from the main models for all biomarkers, except for UA (Figures S3-4).After incorporating the type of residential community into the exposure assessment, the association between PM 2.5 and UA was no longer statistically significant (Figure S3b).Given that, the effect estimations from our major outcomes, i.e., GFR scr and BUN, were not significantly changed by the limitation of using city-level PM 2.5 data.Additionally, the bootstrapped method tends to report weaker associations between PM 2.5 and kidney function biomarkers, which suggests that the measurement errors in exposure may lead to underestimated uncertainties embedded in the associations.Finally, we also found for exposure in a longer term, the effects for per-unit change in PM 2.5 tend to show larger point estimates but with wider uncertainty ranges (Figure S7).Using a longer timewindow for exposure might be more representative for the chronic effect of PM 2.5 on kidney function, but less representative to show the improved air quality.(10).The result from our study further confirmed the potential kidney toxicity of PM 2.5 ; although, the estimated effect [-0.42 (-0.78, -0.06) mL/min/1.73m 2 change in GFR src per an increment of 10 μg/m 3 in PM 2.5 ] was smaller compared to the findings from previous study (12).Apart from the decline in eGFR, evidence using other outcomes was also supportive for an adverse effect of PM 2.5 on kidney function.Significant associations between chronic exposure to PM 2.5 with the prevalence and incident of CKD and progression of ESRD were suggested from several nationwide cross-sectional and cohort studies (7)(8)(9)(10), in which the definition of CKD and ESRD was largely based on the value of eGFR by SCR.Additionally, a unique study based on an 11-year collection of 71,151 native kidney biopsies from 938 hospitals across 282 cities in China observed that long-term exposure to PM 2.5 was associated with an increased risk of membranous nephropathy, the second leading type of glomerulopathy that contributed to 23.4% of all cases (31).Besides, the observational studies, a few biological evidences also indicate an association between PM 2.5 exposure and impaired kidney function (Supplmental S1.8).

Implications.
Long-term exposure to ambient PM 2.5 has been known to cause premature deaths by increasing the risks of cardiorespiratory diseases.GBD study in 2016 estimated the global deaths and DALY attributed to PM 2.5 was about 4.1 million and 105.7 million, respectively (32).However, these numbers could underestimate the health impacts from air pollutants, because a recent study on the association between all-cause mortality and PM 2.5 showed that cardiorespiratory effects might not be the only explanation behind the mortality burden attributable to PM 2.5 (33).One recent study estimated that the annual global toll of CKD attributable to ambient PM 2.5 exposure is significant with 6.9 million incident cases of CKD and 11.4 million DALYs (3).representing about 10% of burden of disease reported from GBD 2016.In accordance with other evidence (7-10), our findings further reveal a causal linkage between PM 2.5 exposure kidney impairment and suggest that kidney disease could be a nonnegligible outcome of the poor air quality.Future evaluations should incorporate the effects of PM 2.5 on kidney function and diseases, to accurately quantify the risks from nonoptimal air quality.4.3.Limits.First, this study is based on all-available observations from a series of preestablished surveys with many general aims, which were not focused on kidney function specifically.Its representativeness depends on the missing patterns in the dataset.However, from the publicly available version of CHARLS, information on the missingness is limited, which can introduce potential bias into (e.g., survival bias), or lower the representativeness of our findings.Second, we reasoned the estimated effect might be biased due to exposure misclassification by the usage of monthly and city-level averages of PM 2.5 concentrations; although, this is the best dataset available for analysis because the CHARLS did not release the population data with a higher spatiotem-poral resolution, in order to protect personal privacy.Finally, although the difference-in-difference analysis controls for some unchanged risk factors on kidney function (e.g., generic defects) by the study design itself, our findings could be undermined by the unmeasured longitudinal confounders.Such confounders can be environmental factors (i.e., lead) that shared the common emission sources of the ambient PM 2.5 .The study design also controlled for the longitudinal factors that were progressed with time in the same pattern.Age is an example of such factor, if we assume it affects kidney function linearly.However, the design cannot fully control for nonlinear risk factors.For instance, the effect of 4-year's aging could be varied between individuals of different generations.Compared to conventional crosssectional analyses, our difference-in-difference design is more capable to reveal a causal effect of PM 2.5 , but the causality of our findings is not conclusive and should be reexamined in future.

Conclusion
Based on a quasiexperiment design, this study provides a strong evidence for the linkage between chronic exposure to ambient PM 2.5 and kidney impairment.The findings suggest clean air actions applied in China brings rapid improvement in air pollution and can lead to beneficial health effect by reducing the impact of kidney diseases.

Table 1 :
Statistics of the population characteristics.

Table 2 :
The association between PM 2.5 and kidney function biomarkers, estimated directly from the fully adjusted regression or after a further correction using the bootstrap method.The estimated associations, which are statistically significant (p < 0:05), are highlighted by bolded numbers.*Thepositive/negativechange means increase/ decrease in a biomarker.For GFR scr or GFR cys , a negative association indicates for the toxic effect of PM 2.5 ; For BUN or UA, a positive association indicates for the toxic effect of PM 2.5 .Figure1: The subpopulation-specific effects of PM 2.5 on the biomarkers of kidney function.The fully adjusted model incorporated both the baseline covariates and the longitudinal covariates.The dots denote significant associations, and the circles denote nonsignificant ones.Longitudinal covariates denoted temporal changes in the inconstant variables (i.e., body weight, marriage, drinking, smoking, cooking energy type, and indoor temperature maintenance); the baseline covariates denoted values of the longitudinal variables in the baseline wave and the constant variables (i.e., residence, sex, education, age at 2011, and average BMI).Along x-axis, the positive/negative change means increase/decrease in a biomarker.For GFR scr or GFR cys , a negative association indicates for the toxic effect of PM 2.5 ; for BUN or UA, a positive association indicates for the toxic effect of PM 2.5 . #