Association between long-term exposure to particulate matter 2.5 and the risk of rapid renal function decline: a prospective study from CHARLS

ABSTRACT

This study investigated the association between particulate matter 2.5 (PM_2.5) exposure and rapid renal function decline in Chinese adults aged ≥45. We used data from the China Health and Retirement Longitudinal Study data from 2011 to 2015 (4-year follow-up period). The study outcome was rapid estimated glomerular filtration rate (eGFR) decline, defined as a decrease of ≥33% in eGFR from baseline. The associations between PM_2.5 and rapid renal function decline were analyzed by logistic regression models and logistic generalized additive models. Among 4365 participants, higher PM_2.5 exposure was linked to increased incidence of rapid eGFR decline (P = .037 for trend). Higher PM_2.5 concentrations, older age, a higher baseline eGFR, and elevated hemoglobin A1c and systolic blood pressure levels were identified as independent risk factors for rapid renal function decline (P < .05). Logistic regression demonstrated that higher long-term PM_2.5 exposure was associated with rapid eGFR decline (per 10-unit increase; odds ratio (OR) = 1.158 [95% confidence interval (CI) 1.025–1.308], P = .018), particularly in participants with eGFR <90 ml/min/1.73 m² (per 10-unit increase; OR = 1.298 [95% CI 1.066–1.582], P = .010). Moreover, effect modification indicated that the association between PM_2.5 and rapid eGFR decline was significantly stronger in participants with hypertension than in those without hypertension (P = .009 for interaction). An estimated 18.8% of rapid eGFR decline cases were attributable to PM_2.5. Furthermore, logistic generalized additive model analysis demonstrated non-linear associations between PM_2.5 and rapid eGFR decline, with significant associations observed at higher PM_2.5 concentrations (P < .05), especially in individuals with hypertension. Long-term exposure to high PM_2.5 is associated with an increased risk of rapid renal function decline, particularly in individuals with low baseline eGFR and hypertension. This finding highlights the need for public health policies to reduce PM_2.5 levels and protect vulnerable populations.

KEY LEARNING POINTS.

What was known:

• PM_2.5 exposure is linked to systemic diseases, with emerging evidence suggesting kidney damage, but evidence from high-pollution, non-Western settings remains limited.

This study adds:

• This large Chinese cohort demonstrates long-term ambient PM_2.5 exposure significantly increases rapid estimated glomerular filtration rate (eGFR) decline risk in adults ≥45 years.

• Hypertension critically modifies this risk: individuals with hypertension show significantly stronger susceptibility to PM_2.5-associated rapid eGFR decline.

• The exposure-response is non-linear: those with normal baseline eGFR (≥90) are susceptible only at very high PM_2.5 (>75 µg/m³), while those with lower eGFR (<90) show risk even at lower concentrations.

Potential impact:

• Clinicians should consider high PM_2.5 as an environmental risk factor warranting clinical awareness for kidney decline, especially in hypertensive patients or those with existing mild kidney impairment.

• Public health policies should consider prioritizing stricter air quality controls targeting PM_2.5 reduction below current standards (35 µg/m³), potentially preventing ∼19% of rapid eGFR decline cases attributable to PM_2.5 in populations with comparable exposure levels and demographic profiles.

• Enhanced kidney function screening may be warranted for hypertensive individuals residing in high PM_2.5 areas.

INTRODUCTION

Atmospheric fine particulate matter, with an aerodynamic diameter of <2.5 μm (PM_2.5), is detrimental to human health. PM_2.5 primarily includes elemental carbon, biological substances, inorganic and organic components, and trace elements [1]. PM_2.5 readily adsorbs harmful substances such as bacteria and viruses [2]. Upon inhalation, PM_2.5 crosses the blood–gas barrier, reaching distant organs via the bloodstream and causing various toxic effects [2]. Exposure to elevated PM_2.5 levels has been associated with an increased risk of hypertension [3], diabetes mellitus [4, 5], respiratory diseases [6, 7], cardiovascular diseases [8, 9], stroke [10], and mortality [11, 12].

Air pollution, particularly PM_2.5, may have detrimental effects on kidney function. PM_2.5 accumulates in the kidneys, causing endothelial dysfunction [13], abnormal renin–angiotensin system (RAS) activity [14], and immune complex deposition [15]. Mechanisms underlying PM_2.5-induced kidney damage include inflammation, oxidative stress, apoptosis, deoxyribonucleic acid damage, and autophagy [16–18]. These effects may manifest acutely or after long-term PM_2.5 exposure. Of note, the adverse renal effects of PM_2.5 are not limited to adults. Childhood exposure above the WHO annual guideline (5 μg/m³) has been associated with decreased estimated glomerular filtration rate (eGFR) in adolescence, and prenatal exposure to sulfate aerosols (a major component of PM_2.5) has been linked to a higher incidence of congenital anomalies of the kidney and urinary tract [19]. These life-course findings underscore the broad public health importance of PM_2.5 exposure.

Despite accumulating evidence linking PM_2.5 to renal outcomes, critical limitations in existing research necessitate further investigation. First, prior studies predominantly relied on Western cohorts [20–25], raising questions about the generalizability of findings to Asian populations with distinct genetic backgrounds, lifestyle factors (e.g. diet, pollution exposure levels, and pollution source patterns), and healthcare systems. Second, several methodological gaps persist: (i) exposure assessments often failed to account for residential migration history, increasing the risk of exposure misclassification; (ii) a lack of adjustment for key behavioral confounders—such as smoking and alcohol consumption, factors strongly associated with renal outcomes in developing nations—may introduce residual confounding; and (iii) few studies specifically examined rapid renal decline, a clinically critical endpoint, in middle-aged and elderly populations residing in high-pollution settings.

The China Health and Retirement Longitudinal Study (CHARLS) is a nationally representative longitudinal survey of Chinese adults aged ≥45 years and their families [26, 27]. Encompassing 150 counties and 450 communities across 28 provinces, CHARLS offers broad coverage. This cohort uniquely addresses the aforementioned gaps through three key advantages: (i) residence and migration history data from the 2014 Life History Survey, which mitigates exposure misclassification bias associated with population mobility; (ii) rich longitudinal covariate data, enabling dynamic tracking of individual-level confounders (e.g. comorbidities, smoking, alcohol consumption, BMI)—critical for avoiding bias from baseline-only covariates given long-term exposure nature of PM_2.5; and (iii) high contextual relevance to aging population of China, a demographic highly vulnerable to PM_2.5-associated nephrotoxicity [28]. Importantly, the renal vulnerability of this population under high pollution stress remains underexplored. Leveraging the CHARLS cohort, this study investigates whether long-term PM_2.5 exposure is associated with accelerated renal function decline in middle-aged and elderly Chinese populations.

MATERIALS AND METHODS

Study population and health data

The baseline CHARLS survey was conducted in 2011, with follow-up surveys in 2013, 2015, and 2018. The data collection was approved by the Institutional Review Board (IRB) of Peking University, and the IRB approval numbers were IRB00001052-11015 and IRB00001052-11014 for the household survey and biomarker collection, respectively. Detailed information regarding blood sample collection has been previously reported [26, 27]. This study utilized data from waves 1 (2011) to 3 (2015) of the Harmonized CHARLS version D. Although the full CHARLS cohort enrolled ∼17 708 participants at baseline, the analytic sample for this study was necessarily restricted by data availability. Specifically, cystatin C (used to calculate eGFR) was measured only in a subset of participants who provided blood samples at both baseline and follow-up (8878 at baseline and 5610 at follow-up). Additionally, complete 2014 Life History Survey residence and migration data were required to accurately assign annual PM_2.5 exposures accounting for residential mobility. The initial cohort comprised 5413 participants meeting all inclusion criteria: (i) harmonized CHARLS data from 2011 to 2015; (ii) cystatin C measurements from 2011 and 2015 blood tests; and (iii) complete 2014 residence/migration history. From this cohort, we applied the following exclusion criteria: (i) baseline eGFR <60 ml/min/1.73 m² (n = 955) to avoid confounding by pre-existing advanced chronic kidney disease (CKD); and (ii) age <45 years at baseline (n = 93) to maintain focus on the middle-aged and elderly population of interest. Despite the reduction to 4365 participants, the final analytic cohort retains broad geographic coverage, including participants from 417 communities in 121 counties across 28 provinces in China (participant flow detailed in Supplementary Figure 1). Trained CHARLS staff used standardized questionnaires to obtain sociodemographic and health-related data through in-person assessments. Self-reported hypertension or diabetes diagnosis, smoking and drinking status, and residence (rural or urban areas) data were obtained from wave 1 as baseline prevalence. Specifically, a current smoker was defined as a respondent who has current smoking habit in wave 1, while a current drinker was defined as a respondent who has consumed any alcoholic beverages within the past 12 months of wave 1. The mean systolic blood pressure (SBP), body mass index (BMI), and alcohol and cigarette consumption were calculated as the average of waves 1–3 owing to high variability over time. Mean alcohol consumption was calculated based on the frequency of drinking in the previous year from waves 1 to 3 and categorized into four levels: none, slight (less than once a week), moderate (once a week to less than daily), and serious (daily and more frequent). Baseline hemoglobin (Hb) A1c levels were obtained from wave 1. This study was reported following the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines. A completed STROBE checklist is provided as Supplementary File 2.

Socioeconomic status

We considered four indicators of socioeconomic status (SES), including annual per-capita household expenditure, education, employment status, and medical insurance coverage. For annual per-capita household expenditure, employment status, and medical insurance coverage, we used the most recent available data from waves 3 to 1 for each participant. Annual per-capita household expenditure was calculated as total household consumption divided by total household residents. Participants were then categorized into per-capita expenditure above the median and below the median, respectively. Education was categorized into three levels: less than high school, high school or equivalent, and college or above. Employment status was grouped as employed (including participants with paid employment, self-employed, or retired) and unemployed. Medical insurance coverage was categorized into insured (covered by either public or private health insurance plans) and uninsured.

In line with previous studies, we used latent class analysis to generate an overall SES variable based on the above indicators. After assessing model fitting statistics using the Bayesian information criterion, a latent SES variable with two classes was identified as optimal. Based on item response probabilities, the two classes were labelled as lower SES and upper SES, respectively. The item response probabilities of the identified latent classes are presented in Supplementary Table 1. We also compared the distribution of the four SES indicators between the two latent classes using chi square tests, supporting the validity of the identified classes (Supplementary Table 2).

Ambient PM2.5 exposure assessment

We obtained PM_2.5 data from the global surface PM_2.5 dataset established by the Atmospheric Composition Analysis Group [29]. Goddard Earth Observing System-Chemistry (GEOS-Chem) models were built to measure near-surface PM_2.5 concentrations by acquiring aerosol optical depth (AOD) data from multiple orbiting satellites (NASA MODIS, MISR, and SeaWIFS) at a 0.01°× 0.01° resolution (∼1 km × 1 km). This was calibrated to global ground-based observations using geographically weighted regression to obtain ground-level annual PM_2.5 concentrations in China. The predicted annual mean PM_2.5 concentration had a high agreement with globally distributed ground monitoring data (R² = 0.81). County-level annual average PM_2.5 concentrations were calculated from grid cell (0.01°× 0.01°) averages within each county. To address exposure misclassification from intra-county heterogeneity, we stratified analyses by urban–rural residence. Residential mobility during follow-up was adjusted using 2014 migration history: for participants moving across counties, PM_2.5 exposure was reassigned based on relocation year and duration, calculated as a time-weighted average. Subsequently, we estimated the annual PM_2.5 exposure of study participants over the study period as the primary model. In the sensitivity analysis, we estimated associations using 3- and 5-year moving averages of PM_2.5 concentrations to account for potential lag effects and variability in exposure timing. The 3–year moving average was selected to reflect a moderate accumulation period that balances year–to–year variability, while the 5–year moving average captures a longer, more stable exposure window consistent with the chronic progression of kidney disease. Shorter–term windows (e.g. 1 year) were not examined because eGFR was measured only at baseline (2011) and at the end of follow–up (2015). Without intermediate blood collections, the study design was thus directed at assessing chronic, long–term effects rather than acute renal responses to transient pollution episodes. We categorized PM_2.5 intensity into three groups—low (<35 μg/m³), medium (35–75 μg/m³), and high (>75 μg/m³)—based on the National Ambient Air Quality Standard for annual mean PM_2.5 in China [30].

Study outcomes

A decline in GFR has been widely accepted as a suitable outcome for CKD clinical trials [31, 32]. Baseline and exit eGFR were calculated using the CKD epidemiology collaboration cystatin C equation [33], based on cystatin C levels measured in waves 1 and 3. The study outcome was rapid eGFR decline, defined as a decrease of ≥33% in eGFR from baseline.

Statistical analyses

Continuous variables were presented as means with standard deviation and medians with interquartile ranges for normal and non-normal distributions, respectively. Categorical variables were expressed as frequencies and ratios. The P for trend was calculated using the Cochran–Armitage test for binary variables and the Spearman rank correlation for continuous or ordinal variables to evaluate the differences in clinical characteristics among the PM_2.5 groups [34]. Univariable and multivariable logistic regression models were used to evaluate the linear association between PM_2.5 exposure and the risk of rapid eGFR decline. Candidate variables with a P value <.1 on univariable analysis were included in the multivariable model. To account for the clustered nature of the CHARLS data, where participants are nested within households, all logistic regression models (both univariable and multivariable) included a random intercept at the household level. Additionally, stratified analyses provided insights into potential effect modifications. These analyses were performed based on eGFR (≥90 ml/min/m² and <90 ml/min/m²), urban or rural area residence, sex (male or female), age (<65 years and ≥65 years), normal weight (BMI < 25 kg/m²), current smoker (no or yes), and current drinker (no or yes). The influence of comorbidities, including diabetes mellitus and hypertension, was also assessed.

PM_2.5 concentration effects on rapid eGFR decline were estimated using a generalized additive logistic model (mgcv package version 1.8-40 in the R program, version 4.2.1; R Core Team). The R code implementing the generalized additive model (GAM) analysis is available in the Supplementary Methods. In this model, smoothing terms [denoted as s(·)] were used to flexibly model non-linear relationships between continuous predictors and the outcome. These terms utilize penalized thin-plate splines, allowing data-driven estimation of complex exposure–response patterns without assuming linearity. The model was adjusted for age, baseline eGFR and HbA1c, and mean SBP from waves 1 to 3 via respective smoothing terms [s(age), s(baseline_eGFR), s(HbA1c), s(SBP)], while also incorporating a Gaussian random intercept for household clustering. The differences between the smooth estimated for the stratified baseline eGFR and hypertension status subgroups were also modeled. The population attributable fraction of PM_2.5 on rapid eGFR decline was empirically estimated as the reduction percentage in rapid eGFR decline frequency under the GAM when PM_2.5 exposure was reduced to the grade-2 criteria set by the Chinese ambient air-quality standards (35 μg/m³). Statistical analyses were performed using SPSS Statistics (version 25.0; IBM, Armonk, NY, USA) and R software (version 4.2.1; R Core Team), specifically, the “mgcv” package version 1.8-40 for the GAM [35]. To address potential confounding by SES and regional disparities, we further adjusted the main models for SES and regional fixed effects. A series of sensitivity analyses were performed by removing covariates to evaluate the robustness of the PM_2.5–eGFR decline association. Statistical significance was set at two-tailed P < .05, and effect estimates were presented as confidence intervals (CIs).

RESULTS

Characteristics of the participants

Participants were categorized into three groups based on their 4-year mean PM_2.5 exposure concentrations (2011–2015) to evaluate the study population’s characteristics. The incidence of rapid eGFR decline significantly increased across PM_2.5 exposure cohorts, with 1.56%, 2.29%, and 3.33% rates in the low, medium, and high exposure groups, respectively (P = .037 for trend; Supplementary Table 3). Baseline age, hypertension and diabetes proportions, and mean SBP and BMI levels significantly increased, whereas baseline eGFR decreased with a higher PM_2.5 exposure cohort (P < .05 for trend; Supplementary Table 3). No significant differences were detected in sex, baseline HbA1c levels, baseline residence, baseline proportions of drinking and smoking, average alcohol consumption during the study period, or average daily cigarette intake among smokers throughout the study period.

Factors related to rapid eGFR decline

Univariable logistic regression analysis revealed that the average PM_2.5 concentration was associated with a rapid eGFR decline (per 10-unit increase; odds ratio [OR] = 1.124 [95% CI 1.008–1.252], P = .035). The OR for rapid eGFR decline significantly increased with higher age, eGFR, mean SBP, and HbA1c, and combined hypertension or diabetes mellitus (all P < .05; Table 1). Continuous variables with P < .1 in the univariable analysis were included in the multivariable model, whereas categorical variables, including hypertension and diabetes, were excluded owing to their relationship with SBP and HbA1c, respectively. Multivariable analysis revealed that higher PM_2.5 concentrations, older age, elevated eGFR, HbA1c, and SBP levels were independent risk factors for rapid renal function decline (P < .05; Table 1). In sensitivity analyses that either sequentially added socioeconomic/regional covariates or removed HbA1c/mean SBP from the main model, the PM_2.5–rapid eGFR decline association remained robust; the odds ratio for a 10 μg/m³ increase in PM_2.5 changed by <3% in all cases, and 95% confidence intervals consistently excluded the null (Supplementary Table 4). We categorized the 4-year mean PM_2.5 concentration into three groups to ensure the robustness of PM_2.5 intensity. The high PM_2.5 group consistently exhibited a significantly increased risk of rapid eGFR decline compared with the low PM_2.5 group across all multivariable models, with adjusted odds ratios ranging from 2.851 to 3.348 (all P < .05). For instance, after full adjustment for age, baseline eGFR, HbA1c, mean SBP, region, and SES, the OR for the high–exposure group was 3.348 (95% CI 1.278–8.775, P = .014). No significant association was observed for the medium PM_2.5 group in any model (Supplementary Table 5).

Table 1:

Logistic regression analysis of factors associated with rapid renal function decline among older adults in China (2011–2015).

	Univariable analysis		Multivariable analysis
Variables	OR (95% CI)	P value	OR (95% CI)	P value
4-year mean PM2.5	1.124 (1.008–1.252)	.035	1.158(1.025-1.308)	.018
Age	1.086 (1.062–1.111)	<.001	1.145 (1.115-1.175)	<.001
Male	1.039 (0.698–1.547)	.849	/	/
eGFR	1.047 (1.036–1.058)	<.001	1.073 (1.060–1.087)	<.001
Hypertension	2.221 (1.488–3.316)	<.001	/	/
Mean SBPa	1.256 (1.134–1.392)	<.001	1.144 (1.017–1.287)	.025
Diabetes	2.862 (1.651–4.958)	<.001	/	/
HbA1c	1.391 (1.215–1.593)	<.001	1.203 (1.028–1.407)	.021
Mean BMIa	1.036 (0.979–1.096)	.221	/	/
Current drinking	0.819 (0.529–1.267)	.369	/	/
Mean alcohol consumptiona	/	.554	/	/
None	1	/	/	/
Slight	0.954 (0.594–1.531)	.844	/	/
Moderate	0.575 (0.261–1.267)	.170	/	/
Serious	0.805 (0.396–1.636)	.550	/	/
Current smoking	0.849 (0.537–1.344)	.485	/	/
Mean cigarette consumption among 1235 smokersa	0.978 (0.952–1.005)	.113	/	/
Urban	0.913 (0.590–1.413)	.682	/	/
Region	/	.320	/	/
Central	1	/	/	/
East	1.366 (0.801–2.330)	.252	/	/
West	1.202 (0.279–5.176)	.805	/	/
South	0.642 (0.288–1.434)	.280	/	/
North	0.925 (0.549–1.560)	.771	/	/
Higher SESb	0.914 (0.397–2.108)	.833	/	/

The 4-year mean PM_2.5 is reported as a per 10-unit increase; 4-year mean SBP is reported as a per 10-unit increase.

Abbreviations: PM_2.5, particulate matter 2.5; eGFR, estimated glomerular filtration rate; SBP, systolic blood pressure; HbA1c, hemoglobin A1c; BMI, body mass index; SES, socioeconomic status.

^aVariables are averages of data from waves 1 to 3.

^bVariable is from the most recent available data from waves 3 to 1 for each participant, and others are obtained at baseline in wave 1.

Stratified analyses

Our stratified logistic regression analysis revealed a significant multiplicative interaction between PM_2.5 and self-reported hypertension regarding the risk of rapid eGFR decline (P = .009 for interaction; Table 2). This suggests that the effect of PM_2.5 on rapid eGFR decline is exacerbated by hypertension. Participants with hypertension had a stronger association between the 4-year mean PM_2.5 level and rapid eGFR decline (per 10-unit increase; OR = 1.388 [95% CI 1.149–1.676], P < .001) than those without hypertension (per 10-unit increase; OR = 0.991 [95% CI 0.837–1.172], P = .915), irrespective of baseline age, eGFR, HbA1c, and mean SBP from waves 1 to 3 (Table 2). Furthermore, logistic regression demonstrated that a 4-year mean PM_2.5 concentration was a risk for rapid eGFR decline in participants with eGFR <90 ml/min/1.73 m² (per 10-unit increase; OR = 1.298 [95% CI 1.066–1.582], P = .010), indicating a marginal quantitative interaction between eGFR (<90 ml/min/1.73 m²) and increased 4-year mean PM_2.5 (P = .070 for interaction; Table 2). No interactions were observed between baseline residence, sex, age, BMI, smoking, drinking, diabetes mellitus status, and PM_2.5 in predicting rapid eGFR decline (P > .1 for all interactions; Table 2). Stratified analyses were not expanded to include SES or regions because of the low number of events in certain subgroups (e.g. only 6 events among 280 participants in the upper SES category), which precluded reliable estimation.

Table 2:

Stratified analyses for rapid eGFR decline associated with each 10 μg/m³ increase in 4-year mean PM_2.5 levels (2011–2015) in CHARLS.

Stratified characteristic	Events/N (%)	OR (95% CI)	P value	P for interaction
eGFR	/	/	/	.070
≥90 ml/min/m2	65/1570 (4.1%)	1.092 (0.932–1.278)	.276	/
<90 ml/min/m2	35/2795 (1.3%)	1.298 (1.066–1.582)	.010	/
Residence	/	/	/	.667
Urban	29/1346 (2.2%)	1.045 (0.807–1.352)	.739
Rural	71/3014 (2.4%)	1.195 (1.041–1.372)	.011
Sex	/	/	/	.491
Male	46/1967 (2.3%)	1.230 (1.029–1.471)	.023	/
Female	54/2398 (2.2%)	1.105 (0.936–1.305)	.239	/
Age	/	/	/	.961
<65 years	51/3388 (1.5%)	1.136 (0.967–1.335)	.121	/
≥65 years	49/977 (5.0%)	1.169 (0.965–1.414)	.110	/
BMI	/	/	/	.165
≥25 kg/m2	36/1502 (2.4%)	1.293 (1.060–1.577)	.011	/
<25 kg/m2	60/2849 (2.1%)	1.024 (0.868–1.208)	.778	/
Smoking status	/	/	/	.478
Non-current smoker	73/3073 (2.4%)	1.088 (0.948–1.249)	.230	/
Current smoker	25/1235 (2.0%)	1.352 (1.017–1.799)	.038	/
Drinking status	/	/	/	.899
Non-current drinker	71/2911 (2.4%)	1.191 (1.026–1.383)	.022	/
Current drinker	29/1446 (2.0%)	1.104 (0.889–1.370)	.370	/
Diabetes	/	/	/	.360
No	82/4027 (2.0%)	1.148 (1.002–1.315)	.047	/
Yes	16/269 (5.9%)	1.255 (0.920–1.712)	.151	/
Hypertension	/	/	/	.009
No	56/3185 (1.8%)	0.991 (0.837–1.172)	.915	/
Yes	44/1151 (3.8%)	1.388 (1.149–1.676)	<.001	/

Adjusted for age, eGFR, HbA1c at baseline, and mean SBP from waves 1 to 3. The 4-year mean PM_2.5 concentrations were used in the models. The analysis was performed for an increase of 10 μg/m³ in PM_2.5.

Our analysis revealed a relationship between the 4-year mean PM_2.5 and rapid eGFR decline across the eGFR and hypertension groups. The adjusted associations between PM_2.5 and rapid eGFR decline were insignificant overall or within any eGFR subgroup for participants without hypertension (Supplementary Table 6). However, among participants with hypertension, the associations were significant overall and within both eGFR subgroups of ≥90 ml/min/m² (OR = 1.334 [95% CI 1.023–1.740], P = .033) and <90 ml/min/m² (OR = 1.505 [95% CI 1.128–2.007], P = .005) (Supplementary Table 6).

Logistic GAMs for multiple factors associated with rapid eGFR decline

After adjusting for confounders such as baseline age, eGFR, HbA1c, and average SBP through waves 1 to 3, a non-linear association was observed between higher PM_2.5 exposure levels and an increased risk of rapid eGFR decline (Fig. 1). Assuming a causal relationship, 18.8% of the rapid eGFR decline in China could be attributable to PM_2.5 air pollution exposure. The exposure–response curve derived from GAM remained generally consistent across sensitivity models, though the linear constraint in the model adjusting for baseline age, eGFR, HbA1c, average SBP through waves 1 to 3 and SES produced an implausible negative slope at low concentrations that was not replicated in models with additional regional adjustment (Supplementary Figure 2). Subgroup analyses revealed a non-linear relationship in the eGFR ≥90 ml/min/m² subgroup (Fig. 2a): at PM_2.5 concentrations above 75 μg/m³, each 10 μg/m³ increase was associated with 10.142 higher odds for rapid eGFR decline (OR = 11.142 [95% CI 2.263–54.855]). However, the curve flattened at PM_2.5 concentrations below 75 μg/m³ (OR = 0.945 [95% CI 0.764–1.169]). In contrast, a generally positive linear relationship was suggested between PM_2.5 exposure and rapid eGFR decline risk in the eGFR <90 ml/min/m² subgroup (Fig. 2a). Furthermore, the risk was elevated at lower PM_2.5 concentrations in the eGFR <90 ml/min/m² subgroup compared with the ≥90 ml/min/m² subgroup (Fig. 2a). Additionally, for PM_2.5 levels between 67 and 73 μg/m³, the eGFR <90 ml/min/m² subgroup exhibited a significantly higher risk of rapid eGFR decline than the ≥90 ml/min/m² subgroup (P < .05) (Fig. 2b). We observed similar associations of rapid eGFR decline with the 3- and 5-year moving average PM_2.5 levels to validate the exposure measurement and lag effect of PM_2.5 (Fig. 2c and d). Subgroup analysis of hypertension status revealed a generally horizontal linear relationship between PM_2.5 exposure and rapid eGFR decline risk in individuals without hypertension, indicating that PM_2.5 had no significant effect on eGFR decline (P > .05) (Fig. 2e). However, the risk significantly increased at PM_2.5 concentrations above 55 μg/m³ in individuals with hypertension (P < .05) (Fig. 2e), which was significantly higher than in those without hypertension at concentrations above 70 μg/m³ (P < .05) (Fig. 2f).

Figure 1:

Logistic generalized additive model plots showing exposure–response curve for rapid estimated glomerular filtration rate decline with 4-year mean particulate matter 2.5 between 2011 and 2015 in CHARLS

The gray zones denote the 95% confidence intervals. The red dashed line represents the threshold PM_2.5 concentration at which the risk became significant. The model was adjusted for age, baseline eGFR, HbA1c, and SBP from waves 1 to 3. Abbreviations: PM_2.5, particulate matter 2.5; OR, odds ratio.

Figure 2:

Logistic generalized additive model illustrating the association between PM_2.5 levels and odds of rapid eGFR decline in CHARLS, stratified by baseline eGFR and hypertension. Smooth curves depicting the odds for rapid eGFR decline, stratified by baseline eGFR, along 4-year mean PM_2.5 (a), 3-year moving average PM_2.5 (c), and 5-year moving average PM_2.5 (d) during 2011–2015, respectively. (b) Estimated difference in odds of rapid eGFR decline along PM_2.5 for comparisons of eGFR <90 ml/min/m² versus eGFR ≥90 ml/min/m², stratified by eGFR. *Significant difference for PM_2.5 67–73 μg/m³ (P < .05). (e) Smooth curves depicting the odds for rapid eGFR decline along 4-year mean PM_2.5, stratified by baseline self-reported hypertension. (f) Estimated difference in odds of rapid eGFR decline along PM_2.5 for comparisons of hypertension versus non-hypertension, stratified by baseline self-reported hypertension. #Significant difference for PM_2.5 >70 μg/m³ (P < .05). The lines represent trends for each group, and shaded regions represent 95% confidence intervals. Significant differences (P < .05) exist where the 95% confidence interval does not cover 0.

In this smoothing model, all smoothing terms were significant (P < .05) (Supplementary Figure 3). The exposure-response curves for age, eGFR, HbA1c, and SBP indicated a significant increase in rapid eGFR decline risk over part or all exposure ranges. The smoothing terms for age and SBP in this model were linear (effective freedom degree for both = 1.000; Supplementary Figures 3a and d).

DISCUSSION

Our investigation provides evidence that prolonged exposure to ambient PM_2.5 is linked to an accelerated decline in eGFR among Chinese adults aged between 45 and 95. Our findings revealed a multiplicative interaction between PM_2.5 concentration and baseline eGFR or self-reported hypertension. Logistic GAM analysis revealed that individuals with eGFR ≥90 ml/min/1.73 m² are susceptible only to higher PM_2.5 concentrations, whereas those with eGFR <90 ml/min/1.73 m² are affected by lower concentrations. Furthermore, our study indicates that patients with hypertension are more vulnerable to PM_2.5 exposure than those without hypertension.

Currently, the literature on ambient PM_2.5 exposure and renal function decline remains limited and inconsistent. Wu et al. [36] reported a 2.60-fold increased odds of ≥30% eGFR decline in a nested case-control cohort of 871 295 participants; Lee et al. [37] found a 17% increased risk of acute kidney injury hospitalization per 5-μg/m³ PM_2.5 increase in 61.3 million U.S. Medicare beneficiaries; and Kwon et al. [38] demonstrated that PM_2.5 independently increased end-stage kidney disease (ESKD) progression risk in diabetic kidney disease patients (aHR 1.28). Nevertheless, other studies have found no association between long-term PM_2.5 exposure and impaired renal function [24, 39]. This inconsistency may be partially explained by varying susceptibilities in the target population, including age and racial differences, as well as substantial differences in exposure levels across study settings: the average ambient PM_2.5 concentrations in these prior studies (∼10 μg/m³ in the U.S. and 25 μg/m³ in Taipei City) were much lower than those in our study (51.2 μg/m³). For context, annual mean PM_2.5 in most European and North American urban areas typically ranges from 8 to 15 μg/m³, while levels in many rapidly developing regions of Asia frequently exceed 50 μg/m³ [40]. This marked exposure gradient likely contributes to the divergent findings, as the biological effects of PM_2.5 may be more readily detectable at higher concentrations. Our findings contribute to the growing evidence that high PM_2.5 exposure increases the risk of rapid renal function decline in middle-aged and elderly populations.

Our study has several strengths. First, it employed a nationwide design using a standard, multistage, stratified sampling method and strict quality-control measures for data reliability [26]. Second, to our knowledge, this study is the first in China to employ a nonlinear model to examine the moderating effect of baseline eGFR and comorbid hypertension on the association between PM_2.5 exposure and rapid eGFR decline, confirming the robustness of the results. Third, we obtained PM_2.5 data from the global surface PM_2.5 dataset generated using GEOS-Chem chemical transport models combined with AOD data from multiple orbiting satellites (NASA MODIS, MISR, and SeaWIFS) to assess PM_2.5 exposure and adjusted for multiple covariates and individual-level risk factors, strengthening the credibility of our results. Identifying vulnerable populations is crucial to mitigate the adverse air pollution effects. Low baseline eGFR and hypertension are widely recognized as risk factors for a rapid decline in renal function [41, 42]. Our research has discovered a synergistic effect between these factors and PM_2.5 exposure on rapid eGFR decline. We observed a significant association between PM_2.5 exposure and rapid eGFR decline in all eGFR groups within the self-reported hypertension subgroup. Participants without hypertension did not show a significant association between PM_2.5 exposure and rapid eGFR decline. However, this interactive relationship between hypertension or low baseline eGFR and air pollution in kidney injury has not been observed in other studies [43–45]. Plausible assumptions state that patients with hypertension or low eGFR, who generally have poor health, may be more susceptible to high air pollution levels. However, our study is the first to substantiate this hypothesis. Our analysis did not find evidence that factors such as residence, sex, age, BMI, smoking, drinking, and diabetes comorbidity modified the association between PM_2.5 exposure and rapid eGFR decline. The apparently higher risk of PM_2.5–associated rapid eGFR decline observed in rural participants (Table 2) may reflect a greater burden of uncontrolled hypertension rather than a direct effect of rural residence. Evidence indicates that hypertension treatment and control rates are substantially lower in rural China than in urban areas [46], which would amplify susceptibility to PM_2.5–related kidney injury, consistent with the significant effect modification by hypertension we observed.

Our investigation into the dose–response relationship between PM_2.5 and the odds of rapid eGFR decline reveals an absence of a safe threshold for PM_2.5 exposure. Some studies have identified a threshold for PM_2.5 exposure and adverse health outcomes [47, 48], such as a J-shaped association between PM_2.5 exposure with a threshold concentration of 33 μg/m³ and functional disability in older adults. However, others have not observed a safe threshold for PM_2.5 exposure in relation to hypertension incidence, asthma mortality, and overall mortality risk [49–51]. Evidence regarding safe PM_2.5 exposure levels in relation to rapid eGFR decline is sparse. Based on our findings, an elevated risk was observed at relatively low PM_2.5 concentrations (below China’s Grade-2 standard [35 μg/m³]), particularly among individuals with health conditions associated with higher renal function decline risk. These findings hold potential implications for public policy development.

Ambient PM_2.5 association with health outcomes may appear relatively minor compared with established risk factors such as diabetes; however, its pervasive presence suggests a potential influence on the risk of rapid eGFR decline in the population. Our analysis indicates that adherence to the current PM_2.5 standards in China (35 μg/m³) was correlated with fewer instances of rapid eGFR decline by ∼18.8% within the studied cohorts. Similarly, Bowe et al. [52] attributed 17%–20% of the global CKD burden to PM_2.5, estimating an annual 6.95 million cases resulting from PM_2.5 exposure. Furthermore, a study analyzing renal biopsy series across 282 Chinese cities discovered that 15.2% of membranous nephropathy cases could be attributed to PM_2.5 air pollution exposure [53]. Compared with ORs, the population attributable fraction offers a more practical and meaningful metric for public health, particularly in estimating the potential health benefits of achieving air pollution control targets.

Several biological pathways may explain the association between long–term PM_2.5 exposure and accelerated eGFR decline. Inhaled particles can enter the circulation and deposit in the kidneys, where they induce oxidative stress [54]. PM_2.5 also triggers systemic and renal inflammation, with elevated pro–inflammatory cytokines such as TNF–α and IL–6, and promotes NLRP3–mediated macrophage activation [55]. Chronic inflammation contributes to mesangial expansion and tubulointerstitial fibrosis. PM_2.5 further impairs renal microvascular integrity by reducing endothelial nitric oxide bioavailability and depleting glomerular endothelial cells [56]. Activation of the intrarenal renin–angiotensin system leads to glomerular hypertension and hyperfiltration [57]. PM_2.5 may also cause DNA damage, mitochondrial dysfunction, and autophagy dysregulation in tubular cells [54]. The convergence of these pathways likely underlies the accelerated renal decline in highly exposed populations. The observed effect modification by hypertension is biologically plausible, as pre–existing vascular and RAS abnormalities may amplify the susceptibility of the kidney to PM_2.5–induced oxidative and inflammatory injury. Although our observational design cannot directly measure these mediators, future studies incorporating biomarkers such as 8–hydroxy–2′–deoxyguanosine for oxidative damage and kidney injury molecule–1 for tubular injury could directly test these mechanisms.

Our study had some limitations. First, county–level PM_2.5 assignment may not capture intra–county variation or specific chemical constituents, as our satellite–derived data provide only total mass concentration. With only two eGFR measurements (2011 and 2015), we could not apply slope–based definitions of rapid progression (e.g. >3 or >5 ml/min/1.73 m²/year). Additionally, the 4–year average exposure does not distinguish short–term pollution episodes from sustained exposure. Future studies with finer exposure data (including chemical speciation), repeated eGFR assessments, and higher temporal resolution are needed. Second, hypertension was based on self–reported physician diagnosis, although such data have been shown to be accurate for common chronic diseases [58, 59]. Third, unmeasured confounders, including indoor air pollution, secondhand smoke, dietary intake, physical activity, and occupational exposures, may persist despite our extensive adjustments. However, an E–value analysis suggests that an unmeasured confounder would need implausibly strong associations with both PM_2.5 and rapid eGFR decline to explain our findings. Fourth, history of cardiovascular or kidney disease was not included to avoid collinearity with adjusted determinants such as hypertension and HbA1c. Fifth, we were unable to examine the impact of PM_2.5 on discrete kidney disease subtypes (e.g. diabetic nephropathy, hypertensive nephrosclerosis, or primary glomerulonephritis), as CHARLS does not collect etiological data such as biopsy findings or specific diagnostic codes for kidney disease subtypes. Future studies with detailed clinical phenotyping are needed to explore whether the association varies across specific subtypes. Finally, our investigation focused on long–term PM_2.5 exposure in Chinese adults aged ≥45 years, so the findings should not be generalized to short–term exposure–outcome relationships, younger individuals, or other ethnic groups.

In conclusion, this study examined the association between long-term ambient PM_2.5 exposure and rapid renal function decline among the Chinese middle-aged and elderly population. A significant association between ambient PM_2.5 and rapid eGFR decline risk was observed. This association was more pronounced in participants with self-reported hypertension and those with a baseline eGFR <90 ml/min/1.73 m² than their counterparts. The results underscore the importance of stricter air quality standards and targeted kidney function screening to support healthy aging in China. Beyond this, the nephrology community can also promote environmental sustainability by adopting “green” healthcare practices, as recently highlighted in the growing field of green nephrology [60].

AUTHORS' CONTRIBUTIONS

L.Z., Y.N., and Y.F. contributed to the study design. L.Z. and B.Z. were responsible for data analysis. L.Z., Y.L., and Y.F. drafted the initial manuscript. L.Z., B.Z., X.D., and Y.F. provided critical feedback during data analyzing and manuscript drafting. Final manuscript was approved by all authors.

CONFLICT OF INTEREST STATEMENT

The authors declare no competing interests.

FUNDING

This research was funded by National Key Research and Development Program of China (2024YFC3607400, 2024YFC3607403), National Key Research and Development Program—Intergovernmental International Scientific and Technological Innovation Cooperation (2024YFE0199600), Natural Science Foundation of Shanghai General Program (24ZR1411600), Research and Development Fund of Zhongshan Hospital (2024XKPT07-RC1), the National Natural Science Foundation of China (81900699), Shanghai Municipal Key Clinical Specialty (shslczdzk02501), Shanghai Federation of Nephrology Project supported by Shanghai ShenKang Hospital Development Center (SHDC2202230), Shanghai Clinical Research Center for Kidney Diseases (22MC1940100), and Shanghai “Science and Technology Innovation Plan” Yangtze River Delta Scientific and Technological Innovation Community Project (21002411500).

DATA AVAILABILITY STATEMENT

The data underlying this article are available in the China Health and Retirement Longitudinal Study at http://charls.pku.edu.cn/index/en.html and the global surface PM_2.5 dataset established by the Atmospheric Composition Analysis Group at https://sites.wustl.edu/acag/datasets/surface-pm2-5/ and can be accessed after application.