https://orcid.org/0000-0003-0902-0115
Abstract
Physical functional limitations (PFLs) increase the vulnerability of adults, but their pathogenesis remains unclear.
We conducted a nationwide longitudinal study on 62 749 records from 18 878 adults (aged ≥45) from 28 provinces in China. Risk of PFLs was assessed using a validated 9-item questionnaire. Exposure levels of air pollutants (PM10, PM2.5, and PM1) and greenness (normalized difference vegetation index, NDVI) were estimated using a satellite-based spatiotemporal model. We used the cumulative link mixed effects model to estimate the associations between short-term and long-term exposure to air pollutants, greenness, and risk of PFLs. We employed the interaction effect model to evaluate interactions between air pollutants and greenness.
Participants were 60.9 ± 9.6 years, with an average follow-up of 5.87 (1.65) years. Exposure to air pollution was significantly associated with a higher risk of PFLs. For instance, the odds ratio (OR) associated with each 10 μg/m3 higher in 6-month averaged PM10, PM2.5, and PM1 were 1.025 (95% CI: 1.015–1.035), 1.035 (95% CI: 1.018–1.054), and 1.029 (95% CI: 1.007–1.050), respectively. Conversely, exposure to greenness was associated with decreased risk of PFLs; the OR associated with each 1-unit higher in 1-year averaged NDVI was 0.724 (95% CI: 0.544–0.962). Furthermore, higher greenness levels were found to mitigate the adverse effects of 1-year, 6-month, 1-month averaged PM10, and 1-year averaged PM2.5 on the risk of PFLs.
Air pollution raises the risk of PFLs, whereas greenness could mitigate the adverse effects. Reducing air pollution and enhancing greenness could prevent physical functioning.
Understanding the processes that limit healthy aging is a key part of ongoing policy amendments. Physical functional limitations (PFLs), a leading contributor to disability, are prevalent worldwide but lack effective measures for prevention and treatment (1). Particularly in China, the large magnitude of the population would induce it to become the country with the highest number of individuals with PFLs, and ultimately disability. However, the pathogenesis of PFLs is complex and remains incompletely elucidated, emphasizing the critical importance of identifying modifiable risk factors of PFLs in general populations.
Emerging evidence suggests that air pollution could heighten risk of PFLs. For example, an American cohort study observed that increased NOX exposure was related to the progression of physical disability (2). Another American cohort study indicated that increased NO2 exposure was associated with poorer physical functioning (3). Notably, a Chinese cohort study reported adverse effects of PM2.5 on hand-grip strength and balance ability (4). These findings imply that exposure to unhealthy levels of air pollution could increase the risk of PFLs. However, the current evidence is still inadequate, as no prior study has investigated the effects of the key air pollutants of PM10 and PM1. It is well acknowledged that different particle size fractions have different hazards to human health (5,6). Therefore, the effect of exposure to PM10 and PM1 on PFLs should also be evaluated.
Greenness plays a critical role in improving health (7). Epidemiological studies have reported various health benefits of residential greenness, including reduced mortality (8,9), decreased incidence of cardiovascular disease, and chronic obstructive pulmonary disease (10–12). Importantly, greenness is also linked to reduced air pollution levels, suggesting a potential for alleviating its adverse effects (13,14). A study in Hong Kong reported that older adults residing in greener areas had lower susceptibility to pneumonia-related deaths caused by air pollution (15). Similarly, our recent study observed that individuals living in low-polluted regions benefited more from greenness compared to those in highly polluted areas (16). We therefore speculate that greenness may help prevent or alleviate the risk of PFLs associated with air pollution, but the potential interactions between air pollution and greenness in PFLs remain unexplored.
To address these gaps, we conducted a nationwide longitudinal study of middle-aged and older adults to investigate the associations between air pollution, greenness, and risk of PFLs, and to evaluate the interactions between air pollution and greenness.
Method
Study Population
This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS), a nationwide dynamic cohort of citizens from 150 counties within 28 provinces. The first wave of CHARLS was conducted in 2011 and follows every 2 years. For the present study, we used data from 4 waves (2011, 2013, 2015, and 2018). Exclusions were applied to participants younger than 45 years, those without data for essential questionnaire information (refer to section “Assessment of Covariates” for details), and those with data available for only 1 wave. The Ethics Review Committee of Peking University approved the CHARLS project (IRB00001052-11015).
Assessment of Air Pollution
Levels of air pollution were assigned based on the spatial concentrations of air pollutants (PM10, PM2.5, and PM1) in the vicinity of each participant’s address. The information on covered spatial areas (encompassing a total of 28 provinces) was shown in Supplementary Figure 1. GPSspgxGeocoding software was employed to geocode the home addresses of subjects. The satellite-based spatiotemporal model was used to estimate the exposure concentrations of PM10, PM2.5, and PM1. The specific methodology of this model is detailed in previous studies (17–19). The results from the 10-fold cross-validation (R2 for PM10 = 82%, root-mean-squared error for PM10 = 19.3 μg/m3; R2 for PM2.5 = 86%, root-mean-squared error for PM2.5 = 10.7 μg/m3; R2 for PM1 = 71%, root-mean-squared error for PM1 = 13.0 μg/m3) showed that the prediction ability of the model is good (17–19). Short-term and long-term exposure levels of air pollutants for each participant were estimated, using 1-month, 3-month, 6-month, and 1-year averaged concentrations of air pollutants before the date of each visit, based on their corresponding geocoded home addresses.
Assessment of Greenness
Residential greenness was estimated using normalized difference vegetation index (NDVI), which assesses vegetation cover based on near-infrared and visible red-light reflectance. NDVI spans a spectrum from −1 to +1, with higher values indicating greater presence of vegetation. We obtained the NDVI raster data set from the Geospatial Data Cloud website, which is managed by the Computer Network Information Center at the Chinese Academy of Sciences (http://www.gscloud.cn). This data set is based on the 16-day Moderate Resolution Imaging Spectroradiometer Terra NDVI data products (MOD13A1, spatial resolution of 500 m). The acquisition process entailed the extraction of sub-data sets, their subsequent amalgamation through stitching, projection of the resulting raster data, unit conversion, and final trimming. We then obtained monthly raster NDVI data using the maximum synthesis method. ArcMap 10.7 software was used to match monthly average NDVI values within a 500-m buffer of each participant’s home address based on latitude and longitude. The 1-year averaged NDVI of each participant before each visit was calculated as the greenness exposure level.
Assessment of PFLs
Risk of PFLs was assessed using a structured 9-item questionnaire in 2 dimensions: physical movement capability (3 items: running or jogging 1 km, walking 1 km, and walking 100 m), and muscle capacity (6 items: rising from a chair after prolonged sitting, climbing several flights of stairs without pausing, stooping, kneeling, or crouching, lifting or carrying objects weighing more than 5 kg, retrieving a coin from a table surface, and extending the arms above shoulder level). Each item was scored as 0 (no difficulty with the activity) or 1 (difficulty or inability to perform the activity), and then summed to get a PFL score ranging from 0 to 9. A higher score corresponded to a higher risk of PFLs.
Assessment of Covariates
Participants completed a standardized questionnaire through the face-to-face interview. To ascertain pertinent covariates for adjustment in our multivariate analyses, a Directed Acyclic Graph (DAG) was constructed (Supplementary Figure 2), employing the online DAGitty tool (www.dagitty.net). Finally, the selected covariates included: (i) demographic characteristics: age, gender, resided location, marital status, and educational level; (ii) lifestyle behaviors: smoking status, drinking status; (iii) residential indoor environment: cooking fuel type, heating supply, gas supply, temperature condition, clean condition (additional details in Supplementary Material Section 1).
Statistical Analyses
Continuous and categorical variables were reported as mean (SD) and frequency (%). The cumulative link mixed effects model (CLMM) was used to assess the relationships between air pollution (averaged PM2.5, PM10, and PM1 concentrations over different time periods) and the risk of PFLs, considering the ordered categorical nature of PFLs. Estimation of the CLMM is via maximum likelihood and mixed models fitted with the Laplace approximation and adaptive Gauss–Hermite quadrature. The CLMM allows to estimate the values of population-level intercepts and slopes (ie, fixed effects), at the same time, to estimate how these intercepts and slopes differ across members of distinct populations (ie, random effects) (20,21). In this study, random-effect covariate adjustment was applied for participants’ ID, ensuring each participant’s contribution to the overall effect estimate is appropriately considered, meanwhile, mitigating potential bias introduced by the repeated measures across 4 waves. Fixed-effect covariates included demographic characteristics, lifestyle behaviors, history of chronic disease, residential indoor environment, and NDVI. The CLMM was also applied to estimate the association between NDVI and the risk of PFLs, using identical covariates except NDVI. The effects were estimated as odds ratio (OR) and 95% confidence intervals (CIs) with each 10 μg/m3 increment in air pollutants concentrations or 1-unit increment in NDVI.
In particular, the interaction effect model was performed to evaluate the interactions between air pollution and greenness. Greenness was categorized into high and low grades based on the median value. The interaction term was calculated for each greenness grade and 10 μg/m³ PM10, PM2.5, or PM1 increment.
The multicollinearity among covariates was assessed. Given the unavailability of variance inflation factors (VIF) calculation within the framework of CLMM, we treated the ordered variables as continuous variables and fitted the linear regression model. This model was fitted using the same data sets and variables, and VIF values were subsequently calculated. The resulting VIF values for control variables were all below 2, with a mean VIF for the model of 1.102. This outcome indicates that multicollinearity among the control variables in this study is negligible and does not significantly affect the analysis.
To assess the robustness of the main findings, we executed several sensitivity analyses. Initially, we implemented additional adjustments for NO2 concentration alongside the comprehensive model adjustments, thereby establishing a dual-pollutant model. Subsequently, only participants who participated in all 4 follow-up visits were analyzed. Lastly, we further adjusted the histories of chronic disease based on the main model. All the statistical analyses were performed in the R software (4.2.0), with statistical significance determined at a 2-side p value of less than .05.
Results
In total, there were 62 749 observations from 18 878 participants in the final analysis. Of the 18 878 participants, 3 382 participants underwent 2 visits, 5 999 participants underwent 3 visits, and 9 497 participants underwent 4 visits. The mean (standard deviation, SD) follow-up duration was 5.87 (1.65) years. Distributions of characteristics of the study participants and PFLs score are shown in Table 1. The average age was 60.9 years (SD = 9.6) at the baseline, and approximately half of the participants were female (51.4%). Most participants were married (86.3%), had low education levels (88.0%), and resided in rural areas (61.3%). Among the participants, 43.3% were self-reported smokers, and 34.4% were self-reported drinkers. Table 2 displays the distribution of air pollutant concentrations across the 4 waves. The overall mean concentrations of PM10, PM2.5, and PM1 were 84.07 µg/m3 (SD = 25.01), 48.59 µg/m3 (SD = 12.86), and 37.26 µg/m3 (SD = 10.76), respectively.
Characteristics of the Study Participants
| Characteristics | Total Observations* | 2011 | 2013 | 2015 | 2018 |
|---|---|---|---|---|---|
| N | 62 749 | 14 186 | 16 089 | 16 390 | 16 084 |
| Age, Mean ± SD | 60.9 ± 9.6 | 58.2 ± 9.1 | 59.8 ± 9.6 | 61.4 ± 9.6 | 63.8 ± 9.3 |
| Marital status, n (%) | |||||
| Married | 54 161 (86.3) | 12 562 (88.6) | 14 082 (87.5) | 14 078 (85.9) | 13 439 (83.6) |
| Others | 8 588 (13.7) | 1 624 (11.4) | 2 007 (12.5) | 2 312 (14.1) | 2 645 (16.4) |
| Education, n (%) | |||||
| Low | 55 204 (88.0) | 12 428 (87.6) | 14 142 (87.9) | 14 460 (88.2) | 14 174 (88.1) |
| Middle | 6 413 (10.2) | 1 490 (10.5) | 1 654 (10.3) | 1 625 (9.9) | 1 644 (10.2) |
| High | 1 132 (1.8) | 268 (1.9) | 293 (1.8) | 305 (1.9) | 266 (1.7) |
| Resided location | |||||
| Rural | 38 472 (61.3) | 8 675 (61.2) | 9 777 (60.8) | 10 089 (61.6) | 9 931 (61.7) |
| Urban | 24 277 (38.7) | 5 511 (38.8) | 6 312 (39.2) | 6 301 (38.4) | 6 153 (38.3) |
| Smoking status, n (%) | |||||
| No | 35 570 (56.7) | 8 387 (59.1) | 9 062 (56.3) | 8 977 (54.8) | 9 144 (56.9) |
| Yes | 27 175 (43.3) | 5 797 (40.9) | 7 026 (43.7) | 7 412 (45.2) | 6 940 (43.1) |
| Missing | None | 2 (<0.1) | 1 (<0.1) | 1 (<0.1) | None |
| Drinking status, n (%) | |||||
| No | 41 144 (65.6) | 9 235 (65.1) | 10 400 (64.6) | 10 681 (65.2) | 10 828 (67.3) |
| Yes | 21 571 (34.4) | 4 949 (34.9) | 5 669 (35.2) | 5 697 (34.8) | 5 256 (32.7) |
| Missing | None | 2 (<0.1) | 20 (0.1) | 12 (<0.1) | None |
| PFLs score, n (%) | |||||
| 0 | 19 868 (31.7) | 5 217 (36.8) | 5 204 (32.3) | 5 098 (31.1) | 4 349 (27.0) |
| 1 | 12 324 (19.6) | 3 136 (22.1) | 3 112 (19.3) | 3 151 (19.2) | 2 925 (18.2) |
| 2 | 8 846 (14.1) | 2 044 (14.4) | 2 325 (14.5) | 2 223 (13.6) | 2 254 (14.0) |
| 3 | 6 580 (10.5) | 1 457 (10.3) | 1 684 (10.5) | 1 707 (10.4) | 1 732 (10.8) |
| 4 | 5 231 (8.3) | 1 060 (7.5) | 1 365 (8.5) | 1 348 (8.2) | 1 458 (9.1) |
| 5 | 3 784 (6.0) | 672 (4.7) | 935 (5.8) | 1 033 (6.3) | 1 144 (7.1) |
| 6 | 2 637 (4.2) | 347 (2.4) | 666 (4.1) | 766 (4.7) | 858 (5.3) |
| 7 | 1 775 (2.8) | 164 (1.2) | 414 (2.6) | 549 (3.3) | 648 (4.0) |
| 8 | 1 070 (1.7) | 67 (0.5) | 246 (1.5) | 309 (1.9) | 448 (2.8) |
| 9 | 634 (1.0) | 22 (0.2) | 138 (0.9) | 206 (1.3) | 268 (1.7) |
| Characteristics | Total Observations* | 2011 | 2013 | 2015 | 2018 |
|---|---|---|---|---|---|
| N | 62 749 | 14 186 | 16 089 | 16 390 | 16 084 |
| Age, Mean ± SD | 60.9 ± 9.6 | 58.2 ± 9.1 | 59.8 ± 9.6 | 61.4 ± 9.6 | 63.8 ± 9.3 |
| Marital status, n (%) | |||||
| Married | 54 161 (86.3) | 12 562 (88.6) | 14 082 (87.5) | 14 078 (85.9) | 13 439 (83.6) |
| Others | 8 588 (13.7) | 1 624 (11.4) | 2 007 (12.5) | 2 312 (14.1) | 2 645 (16.4) |
| Education, n (%) | |||||
| Low | 55 204 (88.0) | 12 428 (87.6) | 14 142 (87.9) | 14 460 (88.2) | 14 174 (88.1) |
| Middle | 6 413 (10.2) | 1 490 (10.5) | 1 654 (10.3) | 1 625 (9.9) | 1 644 (10.2) |
| High | 1 132 (1.8) | 268 (1.9) | 293 (1.8) | 305 (1.9) | 266 (1.7) |
| Resided location | |||||
| Rural | 38 472 (61.3) | 8 675 (61.2) | 9 777 (60.8) | 10 089 (61.6) | 9 931 (61.7) |
| Urban | 24 277 (38.7) | 5 511 (38.8) | 6 312 (39.2) | 6 301 (38.4) | 6 153 (38.3) |
| Smoking status, n (%) | |||||
| No | 35 570 (56.7) | 8 387 (59.1) | 9 062 (56.3) | 8 977 (54.8) | 9 144 (56.9) |
| Yes | 27 175 (43.3) | 5 797 (40.9) | 7 026 (43.7) | 7 412 (45.2) | 6 940 (43.1) |
| Missing | None | 2 (<0.1) | 1 (<0.1) | 1 (<0.1) | None |
| Drinking status, n (%) | |||||
| No | 41 144 (65.6) | 9 235 (65.1) | 10 400 (64.6) | 10 681 (65.2) | 10 828 (67.3) |
| Yes | 21 571 (34.4) | 4 949 (34.9) | 5 669 (35.2) | 5 697 (34.8) | 5 256 (32.7) |
| Missing | None | 2 (<0.1) | 20 (0.1) | 12 (<0.1) | None |
| PFLs score, n (%) | |||||
| 0 | 19 868 (31.7) | 5 217 (36.8) | 5 204 (32.3) | 5 098 (31.1) | 4 349 (27.0) |
| 1 | 12 324 (19.6) | 3 136 (22.1) | 3 112 (19.3) | 3 151 (19.2) | 2 925 (18.2) |
| 2 | 8 846 (14.1) | 2 044 (14.4) | 2 325 (14.5) | 2 223 (13.6) | 2 254 (14.0) |
| 3 | 6 580 (10.5) | 1 457 (10.3) | 1 684 (10.5) | 1 707 (10.4) | 1 732 (10.8) |
| 4 | 5 231 (8.3) | 1 060 (7.5) | 1 365 (8.5) | 1 348 (8.2) | 1 458 (9.1) |
| 5 | 3 784 (6.0) | 672 (4.7) | 935 (5.8) | 1 033 (6.3) | 1 144 (7.1) |
| 6 | 2 637 (4.2) | 347 (2.4) | 666 (4.1) | 766 (4.7) | 858 (5.3) |
| 7 | 1 775 (2.8) | 164 (1.2) | 414 (2.6) | 549 (3.3) | 648 (4.0) |
| 8 | 1 070 (1.7) | 67 (0.5) | 246 (1.5) | 309 (1.9) | 448 (2.8) |
| 9 | 634 (1.0) | 22 (0.2) | 138 (0.9) | 206 (1.3) | 268 (1.7) |
Notes: PFLs = physical functional limitations.
*The cumulative participants throughout the 4 waves.
Characteristics of the Study Participants
| Characteristics | Total Observations* | 2011 | 2013 | 2015 | 2018 |
|---|---|---|---|---|---|
| N | 62 749 | 14 186 | 16 089 | 16 390 | 16 084 |
| Age, Mean ± SD | 60.9 ± 9.6 | 58.2 ± 9.1 | 59.8 ± 9.6 | 61.4 ± 9.6 | 63.8 ± 9.3 |
| Marital status, n (%) | |||||
| Married | 54 161 (86.3) | 12 562 (88.6) | 14 082 (87.5) | 14 078 (85.9) | 13 439 (83.6) |
| Others | 8 588 (13.7) | 1 624 (11.4) | 2 007 (12.5) | 2 312 (14.1) | 2 645 (16.4) |
| Education, n (%) | |||||
| Low | 55 204 (88.0) | 12 428 (87.6) | 14 142 (87.9) | 14 460 (88.2) | 14 174 (88.1) |
| Middle | 6 413 (10.2) | 1 490 (10.5) | 1 654 (10.3) | 1 625 (9.9) | 1 644 (10.2) |
| High | 1 132 (1.8) | 268 (1.9) | 293 (1.8) | 305 (1.9) | 266 (1.7) |
| Resided location | |||||
| Rural | 38 472 (61.3) | 8 675 (61.2) | 9 777 (60.8) | 10 089 (61.6) | 9 931 (61.7) |
| Urban | 24 277 (38.7) | 5 511 (38.8) | 6 312 (39.2) | 6 301 (38.4) | 6 153 (38.3) |
| Smoking status, n (%) | |||||
| No | 35 570 (56.7) | 8 387 (59.1) | 9 062 (56.3) | 8 977 (54.8) | 9 144 (56.9) |
| Yes | 27 175 (43.3) | 5 797 (40.9) | 7 026 (43.7) | 7 412 (45.2) | 6 940 (43.1) |
| Missing | None | 2 (<0.1) | 1 (<0.1) | 1 (<0.1) | None |
| Drinking status, n (%) | |||||
| No | 41 144 (65.6) | 9 235 (65.1) | 10 400 (64.6) | 10 681 (65.2) | 10 828 (67.3) |
| Yes | 21 571 (34.4) | 4 949 (34.9) | 5 669 (35.2) | 5 697 (34.8) | 5 256 (32.7) |
| Missing | None | 2 (<0.1) | 20 (0.1) | 12 (<0.1) | None |
| PFLs score, n (%) | |||||
| 0 | 19 868 (31.7) | 5 217 (36.8) | 5 204 (32.3) | 5 098 (31.1) | 4 349 (27.0) |
| 1 | 12 324 (19.6) | 3 136 (22.1) | 3 112 (19.3) | 3 151 (19.2) | 2 925 (18.2) |
| 2 | 8 846 (14.1) | 2 044 (14.4) | 2 325 (14.5) | 2 223 (13.6) | 2 254 (14.0) |
| 3 | 6 580 (10.5) | 1 457 (10.3) | 1 684 (10.5) | 1 707 (10.4) | 1 732 (10.8) |
| 4 | 5 231 (8.3) | 1 060 (7.5) | 1 365 (8.5) | 1 348 (8.2) | 1 458 (9.1) |
| 5 | 3 784 (6.0) | 672 (4.7) | 935 (5.8) | 1 033 (6.3) | 1 144 (7.1) |
| 6 | 2 637 (4.2) | 347 (2.4) | 666 (4.1) | 766 (4.7) | 858 (5.3) |
| 7 | 1 775 (2.8) | 164 (1.2) | 414 (2.6) | 549 (3.3) | 648 (4.0) |
| 8 | 1 070 (1.7) | 67 (0.5) | 246 (1.5) | 309 (1.9) | 448 (2.8) |
| 9 | 634 (1.0) | 22 (0.2) | 138 (0.9) | 206 (1.3) | 268 (1.7) |
| Characteristics | Total Observations* | 2011 | 2013 | 2015 | 2018 |
|---|---|---|---|---|---|
| N | 62 749 | 14 186 | 16 089 | 16 390 | 16 084 |
| Age, Mean ± SD | 60.9 ± 9.6 | 58.2 ± 9.1 | 59.8 ± 9.6 | 61.4 ± 9.6 | 63.8 ± 9.3 |
| Marital status, n (%) | |||||
| Married | 54 161 (86.3) | 12 562 (88.6) | 14 082 (87.5) | 14 078 (85.9) | 13 439 (83.6) |
| Others | 8 588 (13.7) | 1 624 (11.4) | 2 007 (12.5) | 2 312 (14.1) | 2 645 (16.4) |
| Education, n (%) | |||||
| Low | 55 204 (88.0) | 12 428 (87.6) | 14 142 (87.9) | 14 460 (88.2) | 14 174 (88.1) |
| Middle | 6 413 (10.2) | 1 490 (10.5) | 1 654 (10.3) | 1 625 (9.9) | 1 644 (10.2) |
| High | 1 132 (1.8) | 268 (1.9) | 293 (1.8) | 305 (1.9) | 266 (1.7) |
| Resided location | |||||
| Rural | 38 472 (61.3) | 8 675 (61.2) | 9 777 (60.8) | 10 089 (61.6) | 9 931 (61.7) |
| Urban | 24 277 (38.7) | 5 511 (38.8) | 6 312 (39.2) | 6 301 (38.4) | 6 153 (38.3) |
| Smoking status, n (%) | |||||
| No | 35 570 (56.7) | 8 387 (59.1) | 9 062 (56.3) | 8 977 (54.8) | 9 144 (56.9) |
| Yes | 27 175 (43.3) | 5 797 (40.9) | 7 026 (43.7) | 7 412 (45.2) | 6 940 (43.1) |
| Missing | None | 2 (<0.1) | 1 (<0.1) | 1 (<0.1) | None |
| Drinking status, n (%) | |||||
| No | 41 144 (65.6) | 9 235 (65.1) | 10 400 (64.6) | 10 681 (65.2) | 10 828 (67.3) |
| Yes | 21 571 (34.4) | 4 949 (34.9) | 5 669 (35.2) | 5 697 (34.8) | 5 256 (32.7) |
| Missing | None | 2 (<0.1) | 20 (0.1) | 12 (<0.1) | None |
| PFLs score, n (%) | |||||
| 0 | 19 868 (31.7) | 5 217 (36.8) | 5 204 (32.3) | 5 098 (31.1) | 4 349 (27.0) |
| 1 | 12 324 (19.6) | 3 136 (22.1) | 3 112 (19.3) | 3 151 (19.2) | 2 925 (18.2) |
| 2 | 8 846 (14.1) | 2 044 (14.4) | 2 325 (14.5) | 2 223 (13.6) | 2 254 (14.0) |
| 3 | 6 580 (10.5) | 1 457 (10.3) | 1 684 (10.5) | 1 707 (10.4) | 1 732 (10.8) |
| 4 | 5 231 (8.3) | 1 060 (7.5) | 1 365 (8.5) | 1 348 (8.2) | 1 458 (9.1) |
| 5 | 3 784 (6.0) | 672 (4.7) | 935 (5.8) | 1 033 (6.3) | 1 144 (7.1) |
| 6 | 2 637 (4.2) | 347 (2.4) | 666 (4.1) | 766 (4.7) | 858 (5.3) |
| 7 | 1 775 (2.8) | 164 (1.2) | 414 (2.6) | 549 (3.3) | 648 (4.0) |
| 8 | 1 070 (1.7) | 67 (0.5) | 246 (1.5) | 309 (1.9) | 448 (2.8) |
| 9 | 634 (1.0) | 22 (0.2) | 138 (0.9) | 206 (1.3) | 268 (1.7) |
Notes: PFLs = physical functional limitations.
*The cumulative participants throughout the 4 waves.
Distribution of Air Pollution Levels for Participants
| Pollutants (μg/m3) | Mean ± SD | Min | Median | Max | IQR |
|---|---|---|---|---|---|
| Total | |||||
| PM10 | 84.07 ± 25.01 | 20.27 | 84.18 | 148.55 | 31.61 |
| PM2.5 | 48.59 ± 12.86 | 19.37 | 47.76 | 87.88 | 19.43 |
| PM1 | 37.26 ± 10.76 | 14.50 | 36.34 | 72.79 | 17.79 |
| 2010–2011 | |||||
| PM10 | 83.56 ± 25.98 | 20.27 | 83.23 | 137.23 | 34.51 |
| PM2.5 | 48.38 ± 13.51 | 19.37 | 47.11 | 80.06 | 20.65 |
| PM1 | 36.83 ± 11.12 | 14.50 | 35.42 | 64.62 | 18.82 |
| 2012–2013 | |||||
| PM10 | 85.13 ± 25.34 | 24.42 | 85.60 | 148.55 | 32.32 |
| PM2.5 | 48.89 ± 12.88 | 19.51 | 48.30 | 87.88 | 18.68 |
| PM1 | 37.90 ± 10.93 | 16.25 | 37.08 | 72.79 | 16.66 |
| 2014–2015 | |||||
| PM10 | 84.35 ± 24.02 | 30.52 | 84.18 | 137.26 | 29.06 |
| PM2.5 | 48.74 ± 12.44 | 24.95 | 48.11 | 79.50 | 16.26 |
| PM1 | 37.57 ± 10.55 | 15.99 | 36.62 | 64.86 | 17.05 |
| 2017–2018 | |||||
| PM10 | 83.18 ± 24.74 | 31.09 | 84.37 | 136.69 | 34.77 |
| PM2.5 | 48.33 ± 12.65 | 25.08 | 47.41 | 78.09 | 20.38 |
| PM1 | 36.70 ± 10.43 | 15.98 | 36.01 | 63.72 | 18.13 |
| Pollutants (μg/m3) | Mean ± SD | Min | Median | Max | IQR |
|---|---|---|---|---|---|
| Total | |||||
| PM10 | 84.07 ± 25.01 | 20.27 | 84.18 | 148.55 | 31.61 |
| PM2.5 | 48.59 ± 12.86 | 19.37 | 47.76 | 87.88 | 19.43 |
| PM1 | 37.26 ± 10.76 | 14.50 | 36.34 | 72.79 | 17.79 |
| 2010–2011 | |||||
| PM10 | 83.56 ± 25.98 | 20.27 | 83.23 | 137.23 | 34.51 |
| PM2.5 | 48.38 ± 13.51 | 19.37 | 47.11 | 80.06 | 20.65 |
| PM1 | 36.83 ± 11.12 | 14.50 | 35.42 | 64.62 | 18.82 |
| 2012–2013 | |||||
| PM10 | 85.13 ± 25.34 | 24.42 | 85.60 | 148.55 | 32.32 |
| PM2.5 | 48.89 ± 12.88 | 19.51 | 48.30 | 87.88 | 18.68 |
| PM1 | 37.90 ± 10.93 | 16.25 | 37.08 | 72.79 | 16.66 |
| 2014–2015 | |||||
| PM10 | 84.35 ± 24.02 | 30.52 | 84.18 | 137.26 | 29.06 |
| PM2.5 | 48.74 ± 12.44 | 24.95 | 48.11 | 79.50 | 16.26 |
| PM1 | 37.57 ± 10.55 | 15.99 | 36.62 | 64.86 | 17.05 |
| 2017–2018 | |||||
| PM10 | 83.18 ± 24.74 | 31.09 | 84.37 | 136.69 | 34.77 |
| PM2.5 | 48.33 ± 12.65 | 25.08 | 47.41 | 78.09 | 20.38 |
| PM1 | 36.70 ± 10.43 | 15.98 | 36.01 | 63.72 | 18.13 |
Notes: IQR = interquartile range; PM10 = particulate matter with aerodynamic diameters ≤10 μm; PM2.5 = particulate matter with aerodynamic diameters ≤2.5 μm; PM1 = particulate matter with aerodynamic diameters ≤1.0 μm.
Distribution of Air Pollution Levels for Participants
| Pollutants (μg/m3) | Mean ± SD | Min | Median | Max | IQR |
|---|---|---|---|---|---|
| Total | |||||
| PM10 | 84.07 ± 25.01 | 20.27 | 84.18 | 148.55 | 31.61 |
| PM2.5 | 48.59 ± 12.86 | 19.37 | 47.76 | 87.88 | 19.43 |
| PM1 | 37.26 ± 10.76 | 14.50 | 36.34 | 72.79 | 17.79 |
| 2010–2011 | |||||
| PM10 | 83.56 ± 25.98 | 20.27 | 83.23 | 137.23 | 34.51 |
| PM2.5 | 48.38 ± 13.51 | 19.37 | 47.11 | 80.06 | 20.65 |
| PM1 | 36.83 ± 11.12 | 14.50 | 35.42 | 64.62 | 18.82 |
| 2012–2013 | |||||
| PM10 | 85.13 ± 25.34 | 24.42 | 85.60 | 148.55 | 32.32 |
| PM2.5 | 48.89 ± 12.88 | 19.51 | 48.30 | 87.88 | 18.68 |
| PM1 | 37.90 ± 10.93 | 16.25 | 37.08 | 72.79 | 16.66 |
| 2014–2015 | |||||
| PM10 | 84.35 ± 24.02 | 30.52 | 84.18 | 137.26 | 29.06 |
| PM2.5 | 48.74 ± 12.44 | 24.95 | 48.11 | 79.50 | 16.26 |
| PM1 | 37.57 ± 10.55 | 15.99 | 36.62 | 64.86 | 17.05 |
| 2017–2018 | |||||
| PM10 | 83.18 ± 24.74 | 31.09 | 84.37 | 136.69 | 34.77 |
| PM2.5 | 48.33 ± 12.65 | 25.08 | 47.41 | 78.09 | 20.38 |
| PM1 | 36.70 ± 10.43 | 15.98 | 36.01 | 63.72 | 18.13 |
| Pollutants (μg/m3) | Mean ± SD | Min | Median | Max | IQR |
|---|---|---|---|---|---|
| Total | |||||
| PM10 | 84.07 ± 25.01 | 20.27 | 84.18 | 148.55 | 31.61 |
| PM2.5 | 48.59 ± 12.86 | 19.37 | 47.76 | 87.88 | 19.43 |
| PM1 | 37.26 ± 10.76 | 14.50 | 36.34 | 72.79 | 17.79 |
| 2010–2011 | |||||
| PM10 | 83.56 ± 25.98 | 20.27 | 83.23 | 137.23 | 34.51 |
| PM2.5 | 48.38 ± 13.51 | 19.37 | 47.11 | 80.06 | 20.65 |
| PM1 | 36.83 ± 11.12 | 14.50 | 35.42 | 64.62 | 18.82 |
| 2012–2013 | |||||
| PM10 | 85.13 ± 25.34 | 24.42 | 85.60 | 148.55 | 32.32 |
| PM2.5 | 48.89 ± 12.88 | 19.51 | 48.30 | 87.88 | 18.68 |
| PM1 | 37.90 ± 10.93 | 16.25 | 37.08 | 72.79 | 16.66 |
| 2014–2015 | |||||
| PM10 | 84.35 ± 24.02 | 30.52 | 84.18 | 137.26 | 29.06 |
| PM2.5 | 48.74 ± 12.44 | 24.95 | 48.11 | 79.50 | 16.26 |
| PM1 | 37.57 ± 10.55 | 15.99 | 36.62 | 64.86 | 17.05 |
| 2017–2018 | |||||
| PM10 | 83.18 ± 24.74 | 31.09 | 84.37 | 136.69 | 34.77 |
| PM2.5 | 48.33 ± 12.65 | 25.08 | 47.41 | 78.09 | 20.38 |
| PM1 | 36.70 ± 10.43 | 15.98 | 36.01 | 63.72 | 18.13 |
Notes: IQR = interquartile range; PM10 = particulate matter with aerodynamic diameters ≤10 μm; PM2.5 = particulate matter with aerodynamic diameters ≤2.5 μm; PM1 = particulate matter with aerodynamic diameters ≤1.0 μm.
Table 3 presents the associations of air pollution and greenness with the PFLs. Every 10 μg/m3 higher in 1-year averaged PM10 was associated with higher odds of PFLs (OR = 1.013, 95% CI: 1.002–1.024). For 6-month averaged PM10, PM2.5, and PM1, there was a higher PFLs risk by 2.5% (95% CI: 1.015–1.035), 3.5% (95% CI: 1.018–1.054), and 2.9% (95% CI: 1.007–1.050), respectively. For 3-month averaged PM10, PM2.5, and PM1, there was a higher PFLs risk by 2.4% (95% CI: 1.015–1.033), 3.1% (95% CI: 1.015–1.048), and 2.0% (95% CI: 1.001–1.040), respectively. Additionally, for 1-month averaged PM10, PM2.5, and PM1, there was a higher PFLs risk by 1.9% (95% CI: 1.010–1.027), 2.5% (95% CI: 1.010, 1.041), and 2.0% (95% CI: 1.002–1.040), respectively. In contrast, greenness was associated with lower odds of PFLs (OR = 0.724, 95% CI: 0.544–0.962).
Associations of NDVI (Every 1 Unit Higher) and Air Pollutants (Every 10 μg/m3 Higher) With Risk of PFLs
| Exposures | Time Scale | OR (95% CIs) |
|---|---|---|
| NDVI | 1 y | 0.724 (0.544, 0.962) |
| PM10 | 1 y | 1.013 (1.002, 1.024) |
| 6 mo | 1.025 (1.015, 1.035) | |
| 3 mo | 1.024 (1.015, 1.033) | |
| 1 mo | 1.019 (1.010, 1.027) | |
| PM2.5 | 1 y | 1.011 (0.992, 1.031) |
| 6 mo | 1.035 (1.018, 1.054) | |
| 3 mo | 1.031 (1.015, 1.048) | |
| 1 mo | 1.025 (1.010, 1.041) | |
| PM1 | 1 y | 0.997 (0.974, 1.021) |
| 6 mo | 1.029 (1.007, 1.050) | |
| 3 mo | 1.020 (1.001, 1.040) | |
| 1 mo | 1.020 (1.002, 1.039) |
| Exposures | Time Scale | OR (95% CIs) |
|---|---|---|
| NDVI | 1 y | 0.724 (0.544, 0.962) |
| PM10 | 1 y | 1.013 (1.002, 1.024) |
| 6 mo | 1.025 (1.015, 1.035) | |
| 3 mo | 1.024 (1.015, 1.033) | |
| 1 mo | 1.019 (1.010, 1.027) | |
| PM2.5 | 1 y | 1.011 (0.992, 1.031) |
| 6 mo | 1.035 (1.018, 1.054) | |
| 3 mo | 1.031 (1.015, 1.048) | |
| 1 mo | 1.025 (1.010, 1.041) | |
| PM1 | 1 y | 0.997 (0.974, 1.021) |
| 6 mo | 1.029 (1.007, 1.050) | |
| 3 mo | 1.020 (1.001, 1.040) | |
| 1 mo | 1.020 (1.002, 1.039) |
Notes: NDVI = normalized difference vegetation index; PFLs = physical functional limitations; PM10: particulate matter with aerodynamic diameters ≤ 10 μm; PM2.5 = particulate matter with aerodynamic diameters ≤ 2.5 μm; PM1 = particulate matter with aerodynamic diameters ≤ 1.0 μm.
Associations of NDVI (Every 1 Unit Higher) and Air Pollutants (Every 10 μg/m3 Higher) With Risk of PFLs
| Exposures | Time Scale | OR (95% CIs) |
|---|---|---|
| NDVI | 1 y | 0.724 (0.544, 0.962) |
| PM10 | 1 y | 1.013 (1.002, 1.024) |
| 6 mo | 1.025 (1.015, 1.035) | |
| 3 mo | 1.024 (1.015, 1.033) | |
| 1 mo | 1.019 (1.010, 1.027) | |
| PM2.5 | 1 y | 1.011 (0.992, 1.031) |
| 6 mo | 1.035 (1.018, 1.054) | |
| 3 mo | 1.031 (1.015, 1.048) | |
| 1 mo | 1.025 (1.010, 1.041) | |
| PM1 | 1 y | 0.997 (0.974, 1.021) |
| 6 mo | 1.029 (1.007, 1.050) | |
| 3 mo | 1.020 (1.001, 1.040) | |
| 1 mo | 1.020 (1.002, 1.039) |
| Exposures | Time Scale | OR (95% CIs) |
|---|---|---|
| NDVI | 1 y | 0.724 (0.544, 0.962) |
| PM10 | 1 y | 1.013 (1.002, 1.024) |
| 6 mo | 1.025 (1.015, 1.035) | |
| 3 mo | 1.024 (1.015, 1.033) | |
| 1 mo | 1.019 (1.010, 1.027) | |
| PM2.5 | 1 y | 1.011 (0.992, 1.031) |
| 6 mo | 1.035 (1.018, 1.054) | |
| 3 mo | 1.031 (1.015, 1.048) | |
| 1 mo | 1.025 (1.010, 1.041) | |
| PM1 | 1 y | 0.997 (0.974, 1.021) |
| 6 mo | 1.029 (1.007, 1.050) | |
| 3 mo | 1.020 (1.001, 1.040) | |
| 1 mo | 1.020 (1.002, 1.039) |
Notes: NDVI = normalized difference vegetation index; PFLs = physical functional limitations; PM10: particulate matter with aerodynamic diameters ≤ 10 μm; PM2.5 = particulate matter with aerodynamic diameters ≤ 2.5 μm; PM1 = particulate matter with aerodynamic diameters ≤ 1.0 μm.
Figure 1 presents the modification effects of greenness on the association between air pollution and the risk of PFLs. Greenness could alleviate the negative effects of air pollution. For example, the association between 1-year averaged PM10 exposure and PFLs was modified by greenness, as greater changes in PFLs were found in participants exposed to lower levels of greenness (lower-level greenness: OR = 1.014 [95% CI: 1.006, 1.022], higher-level greenness: OR = 0.994 [95% CI: 0.986, 1.002], p interaction < .001); the association between 1-year averaged PM2.5 exposure and PFLs was modified by greenness as well, that greater changes in PFLs were found in participants exposed to lower levels of greenness (lower-level greenness: OR = 1.011 [95% CI: 0.995, 1.027], higher-level greenness: OR = 0.987 [95% CI: 0.971, 1.003], p interaction = .039). Additionally, modification effects of greenness level were also found in 6-month and 3-month averaged PM10 exposure.
The modification effects of greenness level on associations between air pollution and risk of PFLs. The blue line represents higher-level greenness, and the red line represents lower-level greenness. PFLs = physical functional limitations; PM10 = particulate matter with aerodynamic diameters ≤10 μm; PM2.5 = particulate matter with aerodynamic diameters ≤2.5 μm; PM1 = particulate matter with aerodynamic diameters ≤1.0 μm. *Interaction p value < .05.
The results of sensitivity analyses are shown in Supplementary Table 1. Air pollution exposure remained associated with increased risk of PFLs using dual-pollutant modeling of particulate matter and NO2, or when participants who participated in all 4 follow-up visits were analyzed, or further adjusted their histories of chronic diseases.
Discussion
This is the first nationwide longitudinal study to investigate the effects of air pollution and greenness on the risk of PFLs among Chinese middle-aged and older adults. The key findings are: (i) exposure to PM2.5, PM10, and PM1 were associated with a higher risk of PFLs, whereas greenness was related to a lower risk; and (ii) high levels of greenness exhibited a mitigating effect on the detrimental impacts of air pollution on PFLs. To maintain physical functioning, interventions aimed at reducing air pollution and increasing residential greenness are needed.
Insufficient evidence exists regarding the relationship between air pollution and physical functioning. Up to now, there are only 2 relevant studies have been conducted in developed countries. The first study, conducted by de Zwart et al., which is a cohort designed of Dutch older adults, indicated that annual average PM10 and PM coarse exposure were associated with a decrease in physical functioning (3). The second study was conducted by Weuve et al., a cohort designed study of older adults in Chicago, and found that 5-year period NOX exposure was significantly associated with the progression of physical disability (2). In China, only 2 studies focused on PM2.5 and its effects on physical functioning in the older population, revealing significant positive relationships between annual average PM2.5 exposure and physical functioning (4,22). Notably, our findings indicate that PM2.5 has the strongest association with the risk of PFLs no matter in long-term exposure or short-term exposure. This may be because of the different sources and properties of different particles (23,24). For example, in Beijing, secondary aerosols, coal combustion, vehicles, industry, biomass burning, and dust are important sources, each contributing differently to PM1 and PM2.5 (24), which underscore the intricate nature of air pollution and the varied impacts it can have on physical functioning (25).
This study is the first to explore the short-term effects of air pollutants on physical functioning, revealing a heightened impact of short-term air pollution exposure on physical functioning compared to long-term exposure. Several factors may underlie this finding. Firstly, the acute and immediate physiological responses triggered by short-term exposure, including inflammation, oxidative stress, and respiratory irritation (26–28), may exert a more pronounced and immediate impact on the cardiovascular and respiratory systems, leading to discernible effects on physical functioning. Secondly, the variability in air pollution during specific environmental events or seasons, such as wildfires or winter conditions (29–31), introduces additional complexity. These specific environmental events often result in elevated levels of pollutants over a brief timeframe, causing intensified health effects and more noticeable impacts on individuals’ daily activities. For example, during wildfire events, increased PM can lead to acute respiratory and cardiovascular issues (32,33). Notably, PM2.5 associated with wildfires may pose greater harm to human health than PM2.5 from other sources (32), thereby impinging on the ability to engage in physical functioning. Thus, the combination of acute physiological responses and the presence of potent pollutants during short-term exposure events collectively contribute to the heightened impact on physical functioning during short-term exposures. Future research endeavors should continue to explore these relationships to enhance our understanding of the nuanced effects of air pollution exposure on physical functioning.
Previous studies have found the beneficial impacts of greenness (34–38), but the effects of greenness on physical functioning have not been clearly illustrated. Our study indicates that exposure to higher levels of greenness is significantly associated with better physical functioning. We are only aware of a single epidemiological study that explored the association between greenness and physical functioning, which is consistent with our findings, also found a beneficial effect in the Chinese adults (39). Existing evidence reported that greening can effectively reduce the regional temperature, thus affecting the comfort of life, and then affecting the human health outcome from many aspects (40).
We conducted the first epidemiological study reporting higher levels of greenness can mitigate the adverse effects of air pollutants on physical functioning. Several studies have explored the interaction effect of air pollution and greenness on other health outcomes. A study in China among adults aged 45+ demonstrated that greenness mitigates the negative effects of air pollution on glycolipid metabolism biomarkers (16). Another study conducted in United States including approximately 0.18 million adults aged 40+, reported that high neighborhood greenness attenuates the relationship between PM2.5 exposure and cardiovascular mortality (41). Comparable results were observed in our study for physical functioning, a measure of physical health status, providing further substantiation for the protective effects of greenness against the adverse health impacts of air pollution. However, considering the limited evidence regarding the modifying impact of residential greenness on the relationship between air pollution and adverse health outcomes, more population-based and experimental research is needed to confirm the effects and provide the underlying mechanisms.
Although the precise biological mechanisms underlying the interplay between air pollution and greenness on human health remain incompletely elucidated, numerous potential pathways have been suggested. Firstly, higher residential greenness may encourage a greater likelihood of regular physical activity among individuals, yielding significant psychological and physiological advantages that can enhance physical fitness (13,42). Secondly, greenness could mitigate the ambient particulate matter concentrations by adsorbing them in bark and leaves (43). This effect is particularly noticeable in areas with higher levels of greenness, where it can significantly decrease the levels of air pollution exposure for residents. Additionally, greenness, including vegetation and tree, have the potential to modify chemical compositions and concentrations by effectively removing substances like heavy metals and polycyclic aromatic hydrocarbons that may be adsorbed in particulate matter, thereby helping to mitigate associated health risks (44). Further evidence verifying the potential mechanisms of modification effect is needed to inform strategies for mitigating health risks through the integration of greenness.
Our study represents the inaugural nationwide longitudinal study that concentrates on assessing the independent and modification effects of ambient air pollution, and residential greenness on physical functioning in China, considering comprehensive confounding factors. However, several limitations should be acknowledged. First, exposure misclassification may have occurred, as the exposure assessments of air pollution and greenness were contingent upon residential addresses, without considering individual activity patterns such as working addresses and indoor activities. However, we considered the confounding effects of indoor pollution such as using heating supply or not, types of cooking fuels, and smoking status, which makes up for this deficiency to a certain extent. Second, PM2.5 is a heterogeneous mixture, the effects of major PM2.5 components on physical functioning need to be explored in future studies. Third, the assessment of PFLs relied on questionnaires, introducing potential recall bias. Environmental and psychological factors may contribute to subjective variations. Future research remains needed to conduct a detailed detection of physical function in representative areas or to develop a new, low-cost, and accurate approach for physical function measurement that can be applied nationally and globally. Fourth, it is essential to recognize that the modification effects of greenness on air pollution may be influenced by various factors, including the sources and compositions of air pollutants, as well as variations in the structure of greenness and types of vegetation. Therefore, more advanced and comprehensive research is warranted to attain a deeper insight into the intricate and interactive association between greenness and diverse air pollution. Finally, although our study yielded statistically significant results, the observed small effect sizes raise concerns about clinical relevance. It is pertinent to note that the significance is, in part, attributed to the large sample size utilized in our analysis. The clinical implications of these findings may be tempered by the modest effect sizes observed, necessitating a cautious interpretation of the practical significance of the results.
Conclusion
Exposure to PM10, PM2.5, and PM1 could negatively affect physical functioning, whereas exposure to higher greenness could alleviate the negative effects. Our findings provide support for the efficacy of interventions aimed at addressing air pollution and promoting greenness in a synergistic manner, with the goal of enhancing overall health outcomes.
Funding
This work was supported by Natural Science Foundation of China [42375181, 42307540]; Wuhan Center for Disease Control and Prevention [1602-250000196]; Wuhan Municipal Health Commission [WY19A01]; China Postdoctoral Science Foundation [2022M722450]; Fundamental Research Funds for the Central Universities [2042022kf1029].
Conflict of Interest
None.
Acknowledgments
The authors are grateful to the survey staff and all participants of the China Health and Retirement Longitudinal Study (CHARLS).
Author Contributions
K.Z.: Investigation, methodology, validation, formal analysis, data curation, and writing—original draft. J.H.: Investigation, methodology, validation, formal analysis, data curation, and writing—original draft. Z.C.: Investigation, formal analysis, data curation, and software. M.P.: Investigation, methodology, and software. J.T.: Investigation and formal analysis. D.K.: Investigation and formal analysis. F.L.: Investigation, supervision, project administration, and funding acquisition. H.X.: Investigation, supervision, project administration, and funding acquisition.
Data Sharing Statement
The data used above were all available on the CHARLS public website.
