Residential greenness, air pollution and visual impairment: a prospective cohort study | BMC Public Health

Study population

The data utilized in this study were derived from the China Health and Retirement Longitudinal Study (CHARLS), a comprehensive, nationally representative survey focusing on the aging population in China. CHARLS was initiated in 2011 by the National School of Development at Peking University, aiming to evaluate the health, economic status, and social well-being of individuals aged 45 and older [12]. The baseline survey recruited over 17,000 participants from 150 counties across China, and the follow-up surveys were conducted in 2013, 2015, 2018, and 2020, respectively. CHARLS received ethical approval from the Biomedical Ethics Committee of Peking University (IRB00001052–11015), and informed consent was obtained from all participants.

The present study included participants who were 45 years or older and had no DVI/NVI at baseline (2011), and participated at least one of the subsequent follow-up surveys (2013, 2015 and 2018). Participants with missing data on exposure measurements were excluded from the study. Finally, a total of 9,591 participants were included in the data analysis (Figure S1).

Assessment of exposure

Residential greenness was assessed using the Normalized Difference Vegetation Index (NDVI), a satellite-derived measure of vegetation density. NDVI values range from − 1 to 1, with higher values indicating greater vegetation density. For each participant, we calculated the average NDVI within a 250-meter buffer around their residential address. The NDVI data were obtained from global MODIS product MOD13Q1 version 5, a 16-day composite with a high spatial resolution (250 m × 250 m), using annual average NDVI estimates during the study period to minimize seasonal variations. We calculated the annual average NDVI for the year over 1 year prior to the occurrence of visual impairment, loss to follow-up, or the end of the study period, whichever came first. The exposure to air pollution including particulate matter (PM_2.5 and PM₁₀) and nitrogen dioxide (NO₂) was obtained from the ChinaHighAirPollutants (CHAP) dataset [13,14,15,16,17]. The CHAP dataset provides long-term, full-coverage, high-resolution estimates of ground-level air pollutants across China, leveraging artificial intelligence to account for the spatiotemporal heterogeneity of air pollution. The cross-validation coefficient of determination (CV-R²) for PM_2.5, PM₁₀ and NO₂ were 0.92, 0.90, and 0.80, respectively, demonstrating the high accuracy of these estimates. Similarly, we calculated participants’ long-term air pollution exposure using annual average values, applying the same method used for NDVI.

Assessment of visual impairment

Visual impairment was categorized into DVI and NVI. Distance vision was assessed by the question- “How good is your eyesight for seeing things at a distance, like recognizing a friend from across the street (with glasses or corrective lenses if you wear them)?” Near vision was evaluated by the question- “How good is your eyesight for seeing things up close, like reading ordinary newspaper print (with glasses or corrective lenses if you wear them)?”. The answers were “excellent, very good, good, fair, or poor”. A response of “poor” was defined as having either DVI or NVI.

Covariates

Data on covariates included sex, age (< 65 years and ≥ 65 years), body mass index (BMI, calculated as weight in kilograms divided by height in meters squared), education level (illiterate, primary school and middle school or above), marital status (married and single or other), residence (rural and urban), smoking status (smoking and non- smoking), alcohol consumption (> 1 times per month, < 1 times per month and non-consumer), and household income (low and high). Additionally, a history of hypertension, dyslipidemia, diabetes, heart disease, and stroke were collected through face-to-face interviews. Environmental factors such as cooking fuel (solid fuel or clean fuel), average annual temperature, and relative humidity were also included, as they have been identified in previous studies having potential influences on visual impairment [18,19,20]. Temperature and humidity data were obtained from the National Earth System Science Data Center, National Science & Technology Infrastructure of China ( using the same method employed for calculating the NDVI to determine average annual temperature and relative humidity.

Statistical analysis

Descriptive statistics for baseline characteristics were presented as mean ± standard deviation or median with interquartile range for continuous variables, and as number with proportion for categorical variables, respectively. Group differences were assessed using t-test or Mann-Whitney U test for continuous variables, and chi-squared test or Fisher’s exact test for categorical variables.

Cox proportional hazards models were used to estimate hazard ratio (HR) and 95% confidence intervals (CI) for the association between residential greenness (as measured by the NDVI) and air pollution (PM_2.5, PM₁₀, and NO₂) with the risk of DVI and NVI. We performed single exposure model by including residential greenness and air pollution separately and two-exposure models by including residential greenness and air pollution simultaneously. When considered as categorical variables, the results were presented as HR (95%CI) for the quintiles of exposure variables (Quintile 1: lowest, Quintile 5: highest), with Quintile 1 as the reference. For continuous variables, the HR (95%CI) was reported for each increase of 0.1 unit for NDVI and 10 µg/m³ for air pollution. Nonlinear relationship between single exposure with risk of DVI and NVI was assessed by restricted cubic spline (RCS) with three degrees of freedom [21]. All models were adjusted for sex, age, BMI, education level, marital status, residence, smoking status, alcohol consumption, household income, history of hypertension, dyslipidemia, diabetes, heart disease, stroke, cooking fuel type, and environmental factors, such as average annual temperature and humidity.

To assess whether the potential association between air pollution and the risk of DVI and NVI was modified by residential greenness, we divided the cohort into quintiles based on residential greenness and conducted separate models for air pollution within each quintile. Mediation analysis was performed to quantify the proportion of the effect of residential greenness on visual impairment that could be explained by air pollution, following the method of Huang et al. [22]. First, we fitted a linear regression model to calculate the influence of residential greenness on air pollution. A Cox proportional hazards model was then applied to examine the relationship between the mediator variable (air pollution), residential greenness, and visual impairment after adjusting for various covariates. The direct and indirect effects were calculated, and the proportion mediated was determined for each pollutant (PM_2.5, PM₁₀, and NO₂). To ensure robust statistical inference, a resampling method (Monte Carlo simulation) was used to estimate the CIs and P values for these effects. The mediation proportion was calculated as follows: Direct Effect (DE) = ln(HR_DE), Indirect Effect (IE) = ln(HR_IE), and Total Effect (TE) = ln(HR_TE). Here, DE + IE = TE, and the mediation proportion was computed as IE/TE.

Subgroup analyses were conducted to explore potential effect by age (< 65 years, ≥ 65 years), sex (male, female), education level (illiterate, primary school, middle school or above), residence (rural, urban), smoking status (no, yes), hypertension status (no, yes), and household income (low, high). Interaction P values between exposures and covariates were obtained using likelihood ratio tests. To assess the robustness of the findings, sensitivity analyses were performed by recalculating the average values of NDVI and air pollution from baseline until the occurrence of visual impairment, loss to follow-up, or the end of the study period.

All statistical analyses were performed using R version 4.3.2, with a two-sided P value of < 0.05 considered statistically significant.