Greenness Surrounding Schools and Visual Impairment in Chinese Children and Adolescents.

BACKGROUND: Evidence concerning the effects of greenness on childhood visual impairment is scarce.
OBJECTIVES: We aimed to assess whether greenness surrounding schools was associated with visual impairment prevalence and visual acuity levels in Chinese schoolchildren and whether the associations might be explained by reduced air pollution.
METHODS: In September 2013, we recruited 61,995 children and adolescents 6-18 years of age from 94 schools in seven provinces/municipalities in China. Greenness exposure was assessed using the normalized difference vegetation index (NDVI) and the soil-adjusted vegetation index (SAVI) from July to August 2013. Visual impairment was defined as at least one visual acuity level (dimensionless) lower than 4.9 (Snellen 5/6 equivalent). Three-year annual averages of particulate matter (PM) with an aerodynamic diameter of ≤1μm (PM1) and nitrogen dioxide (NO2) at each school were assessed using machine learning methods. We used generalized linear mixed models to estimate the associations between greenness and prevalent visual impairment and visual acuity levels and used mediation analyses to explore the potential mediating role of air pollution.
RESULTS: In the adjusted model, an interquartile range increase in NDVI500m was associated with lower odds of prevalent visual impairment [odds ratio (OR)=0.95; 95% confidence interval (CI): 0.93, 0.97]. The same increase in NDVI500m was also associated with 0.012 (95% CI: 0.008, 0.015) and 0.011 (95% CI: 0.007, 0.015) increases in visual acuity levels for left- and right-eye, respectively. Our results also suggested that PM1 and NO2 significantly mediated the association between NDVI500m and visual impairment. Similar effect estimates were observed for SAVI500m, and our estimates were generally robust in several sensitivity analyses. DISCUSSION: These findings suggest higher greenness surrounding schools might reduce the risk of visual impairment, possibly owing in part to lower PM1 and NO2 in vegetated areas. Further longitudinal studies with more precise greenness assessment are warranted to confirm these findings. https://doi.org/10.1289/EHP8429.

Introduction

Visual impairment in children and adolescents, largely caused by near-sightedness (myopia), has become a major public health concern worldwide, particularly in Asian countries (Morgan et al. 2012). Currently, one-fifth of the world’s population is affected by myopia (Holden et al. 2016), and in the urban areas of some Asian countries, 80–90% of high school children are myopic (Dong et al. 2020; Lin et al. 2004; Pan et al. 2012). There are children and adolescents with visual impairment in China alone, and the number is projected to increase to by 2030 (Sun et al. 2015).

Visual impairment is a multifactorial condition, involving both genetic and environmental factors (Hysi et al. 2020). Evidence suggests that environmental factors have a more important role in the pathogenesis of visual impairment than genetic factors (Morgan and Rose 2005). Globally, the increase in the prevalence of visual impairment, especially myopia in children and adolescents, has coincided with the rapid increase in urbanization (Morgan et al. 2012; Wolfram et al. 2014), and several studies have reported a higher prevalence of myopia in urban areas than in rural areas (Dong et al. 2020; Ip et al. 2008; Shapira et al. 2019). The exact reasons for the observed rural–urban differences are unclear, yet they may be related to urban lifestyles, urban-related environmental exposures, and differences in medical care. There is substantial evidence that extensive screen time and little time spent outdoors are risk factors for developing myopia (Morgan et al. 2012). In addition, two cohort studies, one performed in children from Barcelona, Spain, and the other in children from both rural and areas of Taiwan, have shown that exposure to nitrogen dioxide (), particulate matter (PM) with an aerodynamic diameter of (), and black carbon were associated with the increased incidence of myopia (Dadvand et al. 2017a; Wei et al. 2019). A cross-sectional study among the elderly from both rural and urban areas of six developing countries also reported positive associations of and ozone with myopia (Ruan et al. 2019). Furthermore, exposure induced myopia in an experimental hamster study (Wei et al. 2019). Mechanistically, air pollution can induce oxidative stress and inflammation in the eyes (Jung et al. 2018; Torricelli et al. 2011), causing intraocular inflammation and impairing retinal microvasculature function as well as neuro-activity and axial length growth (Adar et al. 2010; Herbort et al. 2011; Louwies et al. 2013; Njie-Mbye et al. 2013), potentially resulting in myopia.

Green spaces—such as parks, gardens, and forests—are critical components of both urban and rural areas. Evidence suggests that green space could reduce air pollution levels via filtration and adsorption (Markevych et al. 2017), encourage people to participate in outdoor physical activity (Almanza et al. 2012; Amoly et al. 2014; De la Fuente et al. 2020), and reduce recreational use of television, computers, and video games (i.e., recreational screen time) (Dadvand et al. 2014). Therefore, it is plausible to assume that greenspace could be related to a lower myopia risk. To our knowledge, there is only one previous study, which reported that green space, as measured by general levels of vegetation (greenness), was associated with lower odds of spectacles use, as a surrogate for myopia (Dadvand et al. 2017b). However, that study did not investigate potential underlying mechanisms. To help fill this literature gap, our hypothesis-driven study aimed to explore the relationship between greenness exposure and visual impairment in children and adolescents and whether lower air pollution might explain this hypothesized association.

Methods

Study Population

In September 2013, we undertook a large cross-sectional investigation in seven Chinese provinces/province-level municipalities (Figure 1). A four-stage stratified clustering sampling scheme was adopted to recruit study participants (Figure 2). First, seven Mainland China provinces or municipalities were randomly selected. Second, one or two cities were randomly selected from each province or municipality, yielding nine cities. Third, 3–17 schools were randomly selected from each city, generating 94 total schools. Fourth, all students in the selected schools and residing at their current address for were eligible to participate in the study.

Figure 1.

Map of China showing study locations. Four provinces (Liaoning, Ningxia, Hunan, and Guangdong) and three municipalities (Tianjin, Chongqing, and Shanghai) were sampled. Base map data were obtained from ArcMap (version 10.4; Esri), HERE, Garmin, and the GIS User Community.

Figure 2.

Flowchart of the national cross-sectional study of Chinese school children and adolescents (6–18 years of age) selected from seven provinces or municipalities in 2013 ().

After obtaining permission from each school’s principal, we provided classroom teachers with information packages (including child and parent questionnaires, study descriptions, and consent forms) to distribute to students or their parents (or legal guardians). We carefully emphasized the voluntary nature of participation to students and their parents/guardians. For children 6–8 years of age, their parents/guardians completed and delivered both the child and parent questionnaires to the classroom teachers. Children and adolescents 9–18 years of age completed the child questionnaires themselves, and their parents/guardians completed only the parent questionnaires. The questionnaire included items on sociodemographic and lifestyle factors.

All participants or their parents/guardians gave written informed consent before participation. The study protocol was approved by the ethnical committee of the Peking University Health Science Center (reference no. IRB000010523034).

Visual Acuity Measurement and Visual Impairment

We measured the visual acuity of each eye according to the Standard for Logarithmic Visual Acuity Charts set by the Standardization Administration of China (GB11533-2011). This standard recommends a five-mark record for Chinese children and adolescents, which equals to five minus the logarithm of the minimum angle of resolution (LogMAR) (Standardization Administration of the People’s Republic of China 2014). The visual acuity presented as dimensionless values from 4.0 to 5.3, with higher values indicating better visual acuity. In brief, in a well-lit room, experienced eye care professionals measured visual acuity for each eye using a retroilluminated logMAR chart with trumling-E optotypes at a distance of 5 m. If a child correctly identified four or more of five top-line optotypes (4.0 in a 5-mark record, Snellen 5/50 equivalent), they were reexamined at 4.3 (Snellen 6/30 equivalent), 4.6 (Snellen 5/13 equivalent), and then line by line to 5.3 (Snellen 5/2.5 equivalent). We recorded visual acuity for each eye as the lowest line on which four of five optotypes were identified correctly. If a child was unable to read the top line at 5 m, the assessment was repeated at a distance of 1 m and the visual measurements were divided by 4 (Sun et al. 2015). Before measuring visual acuity, we asked the children and adolescents whether they wore glasses, contact lenses, or orthokeratology contact lenses. If a child wore glasses or contact lenses, we first tested visual acuity 30 min after they removed their glasses/contacts (i.e., without correction) and then repeated the measurements with the child wearing glasses/contacts (i.e., with best correction). The measurements without correction were used in the present analysis.

Following the International Council of Ophthalmology population survey recommendation set forth in collaboration with the World Health Organization and the International Agency for the Prevention of Blindness (Colenbrander 2002), we defined visual impairment as at least one eye with visual acuity lower than 4.9 (Snellen 5/6). Both continuous visual acuity levels and dichotomous visual impairment were used as outcome variables.

Greenness Assessment

We estimated greenness surrounding schools using two satellite-based vegetation indices: the normalized difference vegetation index (NDVI) (Tucker 1979) and the soil-adjusted vegetation index (SAVI) (Huete 1988). Both indices were derived from Landsat 8 Operational Land Imager satellite images at a resolution. NDVI and SAVI were calculated based on the land surface reflectance of the visible (red) and near-infrared parts of the electromagnetic spectrum, and SAVI further incorporated a correction factor to minimize influences of soil background. Both indices range from to 1, with higher values indicating higher vegetation levels, values close to 0 representing barren areas, and values close to corresponding to bodies of water. We downloaded cloud-free satellite images taken from July to August 2013, the period of maximum vegetation cover for the study areas and the time closest to the health data collection. For each school, we used one image, and a total of nine images were used. For Liaoning province and Chongqing municipality, we used two images each, and for the remaining five provinces or municipalities, one image each (detailed information about the images is shown in Table S1). Negative pixel values were excluded to avoid the potential confounding effects of water. Greenness surrounding schools was estimated as the mean NDVI and SAVI values in buffers of 500 and around the centroid of each school. We used ArcGIS (version 10.4; Esri) to perform these calculations.

Confounders and Potential Mediators

For a variable to be considered as a confounder, the following criteria had to be satisfied: a) the variable had to be a risk factor for visual impairment; b) it had to be related to greenness exposure; and c) it could not be a mediator on the pathway between greenness and visual impairment (Jager et al. 2008). We constructed a directed acyclic graph (DAG; Figure S1) to retain a minimally sufficient set of confounders in regression models. The following confounders were selected: children’s age (in years), children’s sex (boy vs. girl), children’s ethnicity (Han vs. others), urbanicity (urban vs. rural), parental education level (defined as the highest degree of either parent: below senior high school vs. completed senior high school vs. completed junior college vs. completed college or above), and district/county-level gross domestic product (GDP) ( vs. Yuan/capita; 1 Yuan in 2013), and district/county-level population density ( vs. persons/). All of the above confounders were self-reported except for the district/county-level GDP and population density. More specifically, information on ethnicity was collected from the parents of children of age and from children and adolescents of age by the question: “What is your child's/your ethnicity?” The potential responses were: Han, Hui, Tibet, Mongol, and “other” ethnicities. If “other” ethnicities was chosen, then the parent/guardian or child was asked for more detailed information on the exact ethnicity. For participants who were unsure about their ethnicity or belonged to multiple ethnic groups, we advised them to provide the ethnicity information listed on their national identification card. Because of the limited numbers of Hui, Tibet, Mongol, and other ethnicities, we dichotomized ethnicity into Han vs. others and then modeled the new classification. The reasons for incorporating ethnicity as a potential confounder included the following: a) people with different ethnicities may have different genetic backgrounds, which are closely correlated with myopia risk (Hysi et al. 2020); and b) people with different ethnicities may live in different areas and have different lifestyles, which may affect greenness exposure. Urbanicity was divided into urban and rural areas at the school level according to 2013 administrative divisions, which were provided by the National Bureau of Statistics (http://www.stats.gov.cn/tjsj/ndsj/2013/indexch.htm). Information on GDP (Yuan per capita) and population density (persons per kilometer squared) in the districts (or counties) containing the study schools was also obtained from the National Bureau of Statistics in 2013 and then presented and modeled as a dichotomous variable. Districts (urban area) and counties (rural area) are sub–city-level administrative divisions, and the size of the districts and counties ranged from 29 to and from 2,912 to , respectively, in our study.

Further, based on the DAG (Figure S1), air pollution at school address [i.e., PM with an aerodynamic diameter of () and ] was selected as a candidate mediator of greenness–visual impairment associations. Three-year (2010–2012) average concentrations of and for each school were estimated at a resolution using machine learning methods, which combined satellite-based aerosol optical depth or tropospheric data, ground-monitored air pollutants data, land cover, meteorology, and other spatial predictors (Chen et al. 2018; Zhan et al. 2018). All children and adolescents at a given school shared the same and concentrations.

Statistical Analysis

We employed generalized linear mixed models to evaluate the associations between greenness, prevalent visual impairment, and visual acuity levels, in which provinces (or province-level municipalities) were incorporated as a random effect and greenness metrics and confounders were incorporated as fixed effects. Associations of greenness indices with visual impairment [odds ratios (ORs) and 95% confidence intervals (CIs)] and visual acuity levels [regression coefficients () and 95% CIs] were presented both corresponding to an interquartile range (IQR) difference in greenness indices and by quartiles because natural cubic regression splines [gam-function in R (version 4.1.0; R Development Core Team)] indicated slightly nonlinear relationships (Figures S2–S4). We used NDVI and SAVI within a buffer of schools in the main regression models and adjusted the main models for the potential confounders listed previously.

To assess the robustness of our results, we performed a set of sensitivity analyses. First, we repeated the analysis excluding participants whose parents had myopia (defined as at least one parent reporting myopia; modeled as a dichotomous variable: yes vs. no) (Morgan et al. 2012; Low et al. 2010), those born with low birth weight (defined as parental report of birth weight [WHO 1977]; modeled as a dichotomous variable: yes vs. no) (O’Connor et al. 2002), or studying in the first grade of school (exposure time to school environments was short), as well as by excluding each province or municipality one at a time. Second, we additionally individually adjusted the main models for household income (Lim et al. 2012), parental smoking (Iyer et al. 2012), sleep time (Liu et al. 2021), and lifestyle interventions for weight loss (Lim et al. 2010; Morgan et al. 2012) (which are factors that are potentially associated with myopia risk). Household income was collected from parents by the question: “What is your household income per month?” The potential responses were: , 2,000–4,999, 5,000–7,999, 8,000–11,999, 12,000–14,999, and Yuan. We presented and dichotomized household income as a dichotomous variable: vs. Yuan. Parental smoking was collected from parents by the question: “On how many days of the last month did you smoke one or more cigarettes?” Smokers were defined as those who smoked at least 1 d during the last month. Parental smoking was defined as at least one parent being a smoker. This variable was presented and modeled as a dichotomous variable: yes vs. no. Sleep time was collected from the parents of children of age and from children and adolescents of age by the question: “How many hours do your child/you sleep per day?” The potential responses were: , 7–9, 10–11, and . We dichotomized and modeled this variable as: vs. . Lifestyle interventions for weight loss were reported by the parents of children of age and by children and adolescents of age by answering the question: “During the last 30 days, has/have your child/you ever tried to lose weight via changing dietary habits, increasing physical activity, taking weight-loss drugs, or going on a diet?” The potential responses were yes or no and were modeled as a dichotomous variable. Third, we estimated NDVI and SAVI in larger () buffers to assess the impact of a more distant green space. Finally, we estimated average levels of in the 10 months when students were at school (NDVI products were downloaded from the National Aeronautics and Space Administration’s MODerate-resolution Imaging Spectroadiometer at https://modis.ornl.gov/data/modis_webservice.html) to assess the potential impact of holiday break (i.e., February and July). SAVI products were not available and thus were not estimated.

We used a two-way decomposition method to assess the potential mediating effects of air pollution (3-y annual average concentrations of and at each school and modeled as continuous variables) on the associations of and with visual impairment. The mediation analyses were performed using the CAUSALMED procedure (counterfactual approach) in SAS, and the total effects of greenness on visual impairment was divided into two components that were attributable to a) the greenness directly (direct effect) and b) mediation (indirect effect) (Valeri and VanderWeele 2013). Standard errors (SEs) were calculated using the delta method (VanderWeele and Vansteelandt 2009). The CAUSALMED procedure does not allow for incorporating random effects, so we did not account for the multilevel structure of the data in the mediation analysis. Alternatively, we adjusted for province/municipality as a fixed effect to partially accommodate the issue of bias in the SEs of the estimates due to the clustered outcomes.

In addition, we performed moderated mediation analysis using the CAUSALMED procedure to explore whether the mediating effects of air pollution were modified by children’s sex (boy vs. girl), children’s age (6–11 vs. 12–18 y), and parents’ highest education level (completed senior high school or below vs. completed junior college or above). We tested the significance of the moderation effects using a two-sample -test based on the point estimates and SEs (Altman and Bland 2003).

We further explored potential modification of the associations between and with prevalent visual impairment and visual acuity levels by outdoor exercise time and screen time. Children and adolescents’ outdoor exercise time was evaluated by asking parents of children of age and children and adolescents of age the following question: “How long do you spend in doing outdoor exercise per day?” The potential responses were: , 1–2, 2–4, and . We further categorized the participants into two subgroups: low outdoor exercise group () and high outdoor exercise group (). Children and adolescents’ screen time was collected by asking parents of children of age and children and adolescents of age the following questions: a) “How many minutes do you spend in watching TV per day?”; and b) “How many minutes do you spend in using a computer or play video games per day?” We calculated screen time for each participant by summing the time watching TV and using computer or playing video games and further categorized the data into two subgroups: and (mean value of screen time). We fitted separate regression models to the data for each subgroup and obtained subgroup-specific effect estimates. We assessed differences in the associations between subgroups using a two-sample -test, based on the point estimates and SEs (Altman and Bland 2003).

All statistical analyses were performed in SAS statistical software (version 9.4; SAS Institute Inc.) unless otherwise stated. A for a two-tailed test denoted statistical significance.

Results

Study Participants and Greenness Exposures

We initially invited 65,437 children and adolescents to participate in the study, of whom 62,517 completed the questionnaire (response rate 95.5%). After excluding 522 participants with missing visual acuity measurements, 61,995 children and adolescents were included in the present analysis. The distribution of basic participant characteristics were comparable before and after excluding those missing visual acuity measurements (Table S2).

The mean (SD) age of the 61,995 participants was 11.0 (3.3) y and nearly half of them (48.5%) were girls (Table 1). Most participants were of Han ethnicity (92.3%) and born to parents who graduated from senior high school or below (71.4%). Approximately half of the participants (48.3%) exercised for . In total, 34,216 (55.2%) children and adolescents were diagnosed with visual impairment, and both mean (SD) left- and right-eye visual acuity levels were 4.8 (0.4). Compared with participants without visual impairment, those with visual impairment were more likely to be older (12.3 vs. 9.7 y), girls (52.3% vs. 43.7%), live in urban areas (67.3% vs. 64.6%), live in areas with higher GDP (GDP Yuan/capita: 49.7% vs. 48.4%) or population density (population density persons/: 58.2% vs. 49.0%), have parents with higher education levels (parental highest education level of junior college or above: 29.6% vs. 27.4%), have less outdoor exercise time (: 50.9% vs. 52.8%), and have less screen time (: 40.5% vs. 47.2%).

Table 1

Distribution of sociodemographic, lifestyle, environmental, and visual characteristics of the study population collected from seven Chinese provinces/municipalities in 2013.

Characteristics	All participants	Participants with visual impairment	Participants without visual impairment	p-Valuea
	(n=61,995)	(n=34,216)	(n=27,779)
Age {y [mean (SD)]}	11.0 (3.3)	12.2 (3.3)	9.7 (2.9)	<0.0001
Sex [n (%)]				<0.0001
Boy	31,934 (51.5)	16,315 (47.7)	15,619 (56.2)
Girl	30,061 (48.5)	17,901 (52.3)	12,160 (43.7)
Ethnicity [n (%)]b				<0.0001
Han	57,245 (92.3)	31,570 (92.3)	25,675 (92.4)
Hui	2,771 (4.5)	1,666 (4.9)	1,105 (4.0)
Tibet	110 (0.2)	64 (0.2)	46 (0.2)
Mongol	653 (1.1)	236 (0.7)	417 (1.5)
Otherc	1,216 (2.0)	680 (2.0)	536 (2.0)
Urbanicity [n (%)]				<0.0001
Urban	40,975 (66.1)	23,021 (67.3)	17,954 (64.6)
Rural	21,020 (33.9)	11,195 (32.7)	9,825 (35.4)
Sleep time {h/d [n (%)]}				<0.0001
<9	49,042 (79.1)	28,647 (83.7)	20,395 (73.4)
≥9	12,953 (20.9)	5,569 (16.3)	7,384 (26.6)
Low birth weight [n (%)]				0.95
No	60,054 (96.9)	33,146 (96.9)	26,908 (96.9)
Yes	1,941 (3.1)	1,070 (3.1)	871 (3.1)
Weight control [n (%)]				<0.0001
No	29,534 (47.6)	15,764 (46.1)	13,770 (49.6)
Yes	32,461 (52.4)	18,452 (53.9)	14,009 (50.4)
Parental highest education level [n (%)]				<0.0001
Below senior high school	29,611 (47.8)	16,058 (46.9)	13,553 (48.8)
Completed senior high school	14,598 (23.6)	7,986 (23.3)	6,612 (23.8)
Completed junior college	8,631 (13.9)	4,826 (14.1)	3,805 (13.7)
Completed college or above	9,155 (14.8)	5,346 (15.6)	3,809 (13.7)
Parental tobacco smoking [n (%)]				<0.0001
No	36,227 (58.4)	20,424 (59.7)	15,803 (56.9)
Yes	25,768 (41.6)	13,792 (40.3)	11,976 (43.1)
Parental myopia [n (%)]				<0.0001
No	42,960 (69.3)	22,498 (65.8)	20,462 (73.7)
Yes	19,035 (30.7)	11,718 (34.2)	7,317 (26.3)
Household income {Yuan/month [n (%)]}				0.68
<5,000	53,180 (85.8)	29,333 (85.7)	23,847 (85.9)
≥5,000	8,815 (14.2)	4,883 (14.3)	3,932 (14.2)
District/county-level GDP {Yuan/capita [n (%)]}				0.001
<65,311	31,557 (50.9)	17,214 (50.3)	14,343 (51.6)
≥65,311	30,438 (49.1)	17,002 (49.7)	13,436 (48.4)
District/county-level population density {persons/km^2 [n (%)]}				<0.0001
<2,011	28,441 (45.9)	14,286 (41.8)	14,155 (51.0)
≥2,011	33,554 (54.1)	19,930 (58.2)	13,624 (49.0)
PM_1 {μg/m^3 [median (IQR)]}d	47.4 (10.2)	47.2 (13.0)	47.4 (10.2)	0.06
NO_2 {μg/m^3 [median (IQR)]}d	43.7 (9.8)	43.9 (7.9)	41.6 (20.9)	<0.0001
Children’s outdoor exercise time {h/d [n (%)]}				<0.0001
<2	29,934 (48.3)	16,813 (49.1)	13,121 (47.2)
≥2	32,061 (51.7)	17,406 (50.9)	14,658 (52.8)
Children’s screen time {h/d [n (%)]}				<0.0001
<1	35,031 (56.5)	20,358 (59.5)	14,673 (52.8)
≥1	26,964 (43.5)	13,858 (40.5)	13,106 (47.2)
Children’s grade level [n (%)]				<0.0001
First grade	20,760 (33.5)	12,858 (37.6)	7,902 (28.5)
Other grades	41,235 (66.5)	21,358 (62.4)	19,877 (71.5)
Province/municipality [n (%)]				<0.0001
Hunan	7,563 (12.2)	3,206 (9.4)	4,357 (15.7)
Ningxia	7,755 (12.5)	4,526 (13.2)	3,229 (11.6)
Tianjin	9,592 (15.5)	5,930 (17.3)	3,662 (13.2)
Chongqing	10,352 (16.7)	5,510 (16.1)	4,842 (17.4)
Liaoning	9,121 (14.7)	4,476 (13.1)	4,645 (16.7)
Shanghai	9,009 (14.5)	4,666 (13.6)	4,343 (15.6)
Guangdong	8,603 (13.9)	5,902 (17.3)	2,701 (9.7)
Left eye visual acuity [mean (SD)]	4.8 (0.4)	4.6 (0.4)	5.1 (0.1)	<0.0001
Right eye visual acuity [mean (SD)]	4.8 (0.4)	4.5 (0.4)	5.1 (0.1)	<0.0001

Note: Data are complete for all variables shown. GDP, gross domestic product; IQR, interquartile range; , nitrogen dioxide; , particles with an aerodynamic diameter of ; SD, standard deviation.

-Values were calculated from chi-square tests (categorical variables) or rank-sum tests (continuous variables without normal distribution) or -tests (continuous variables with normal distribution).

Ethnicity was operationalized as Han vs. others during regression modeling.

Detailed information on “other” ethnicities is provided in Table S11.

and were 3-y annual average concentrations at each school.

The median (IQR) values of and were 0.32 (0.12) and 0.19 (0.07), respectively (Table 2). Greenness levels varied greatly across the schools: and levels had a range of 0.14–0.66 and 0.08–0.44, respectively. In addition, NDVI and SAVI were highly correlated ( across the same buffer size, i.e., 0.98) (Table S3).

Table 2

Distribution of normalized difference vegetation index and soil-adjusted vegetation index levels ().

Greenness	Mean (SD)	Median (IQR)	P_25	P_75	Minimum	Maximum
NDVI_500m	0.35 (0.11)	0.32 (0.12)	0.27	0.39	0.14	0.66
NDVI_1000m	0.35 (0.12)	0.33 (0.13)	0.27	0.40	0.12	0.70
SAVI_500m	0.21 (0.08)	0.19 (0.07)	0.16	0.23	0.08	0.44
SAVI_1000m	0.21 (0.08)	0.19 (0.07)	0.16	0.23	0.07	0.47

Note: The vegetation indices (NDVI and SAVI) were calculated using satellite images taken from July to August 2013. IQR, interquartile range; NDVI, normalized difference vegetation index; , the 25th percentile; , the 75th percentile; SAVI, soil-adjusted vegetation index.

Associations between Greenness and Visual Acuity

We estimated an inverse association between greenness levels and visual impairment (Table 3). In the adjusted model, an IQR increase in both and was significantly associated with lower odds of visual impairment (; 95% CI: 0.93, 0.97). When participants with the highest quartiles of () and () values were compared with those with the lowest quartiles of () and (), the ORs for visual impairment were 0.61 (95% CI: 0.57, 0.66) and 0.60 (95% CI: 0.55, 0.64), respectively. We also estimated that increased greenness was associated with higher visual acuity levels (Table 3). An IQR increase in was associated with (95% CI: 0.008, 0.015) and (95% CI: 0.008, 0.015) greater visual acuity for the left and the right eye, respectively. Compared with the participants with the lowest quartiles of and values, those with the highest quartiles showed (95% CI: 0.021, 0.058) and (95% CI: 0.024, 0.048), respectively, increased left-eye visual acuity levels and (95% CI: 0.011, 0.048) and (95% CI: 0.020, 0.043), respectively, increased right-eye visual acuity levels. In the crude models, continuous IQR results were closer to the null, yet the quartile results were the same or further from the null compared with those from the adjusted models (Table S4).

Table 3

Adjusted associations of normalized difference vegetation index and soil-adjusted vegetation index with visual impairment and visual acuity levels among children and adolescents ().

Categories	Visual impairment		Left eye visual acuity		Right eye visual acuity
	OR (95% CI)a,b	p-Value	β (95% CI)a,c	p-Value	β (95% CI)a,c	p-Value
NDVI_500m
Q1 (<0.27)	1 (Ref)		0 (Ref)		0 (Ref)
Q2 (0.27–0.32)	0.61 (0.58, 0.65)	<0.0001	0.033 (0.019, 0.048)	<0.0001	0.027 (0.013, 0.041)	0.0002
Q3 (0.33–0.39)	0.92 (0.87, 0.98)	0.001	0.026 (0.010, 0.041)	0.001	0.020 (0.005, 0.036)	0.012
Q4 (≥0.40)	0.61 (0.57, 0.66)	<0.0001	0.039 (0.021, 0.058)	<0.0001	0.029 (0.011, 0.048)	0.002
Per IQR increased	0.95 (0.93, 0.97)	<0.0001	0.012 (0.008, 0.015)	<0.0001	0.011 (0.008, 0.015)	<0.0001
SAVI_500m
Q1 (<0.16)	1 (Ref)		0 (Ref)		0 (Ref)
Q2 (0.16–0.19)	0.67 (0.63, 0.71)	<0.0001	0.023 (0.014, 0.032)	<0.0001	0.019 (0.010, 0.028)	<0.0001
Q3 (0.20–0.23)	0.69 (0.65, 0.73)	<0.0001	0.015 (0.006, 0.024)	0.001	0.013 (0.004, 0.022)	0.004
Q4 (≥0.24)	0.60 (0.55, 0.64)	<0.0001	0.036 (0.024, 0.048)	<0.0001	0.032 (0.020, 0.043)	<0.0001
Per IQR increased	0.95 (0.93, 0.97)	<0.0001	0.011 (0.007, 0.016)	<0.0001	0.011 (0.008, 0.015)	<0.0001

Note: Effect estimates were calculated using generalized linear mixed models. , regression coefficient; CI, confidence interval; IQR, interquartile range; NDVI, normalized difference vegetation index; OR, odds ratio; Ref, reference; SAVI, soil-adjusted vegetation index.

Models were adjusted for age, sex, ethnicity, parental highest education level, district/county-level gross domestic product, district/county-level population density, and urbanicity.

, , and suggests protective, no, and deleterious effects of greenness on visual impairment, respectively.

, , and suggests protective, no, and deleterious effects of greenness on visual acuity levels, respectively.

The IQR was 0.12 and 0.07 for and , respectively.

The above associations were generally robust in several sensitivity analyses. We found similar results when we excluded children whose parents had myopia (), children born with low birthweight (), and children in the first grade () (Table S5). We also observed similar results when excluding each study province one at a time, although the associations were closer to the null when Hunan or Liaoning provinces was excluded (Table S5). After we additionally and individually adjusted the main regression model for other predictors (i.e., household income, parental smoking, sleeping time, and weight control) of visual impairment, we found that the results were consistent with those of the main analysis (Table S6). When we repeated the analyses using and , the associations were consistent with the main analysis, although the associations with visual acuity levels became stronger for (Table S7). The results were also similar to the main analysis when we repeated the analyses using 10-month average levels of (Table S8).

Screen time significantly modified the associations of and with visual acuity levels (Table 4). For example, in children and adolescents with screen time of , an IQR increase in was associated with (95% CI: 0.000, 0.011)- and (95% CI: 0.001, 0.012)-unit increase in left- and right-eye visual acuity, respectively. The corresponding estimates were (95% CI: 0.012, 0.022) and (95% CI: 0.010, 0.020), respectively, in the left eye and right eye visual acuity respectively, in children and adolescents with screen time of (-values for effect modification were 0.003 and 0.027 for the left and right eye, respectively). The estimated ORs for and with visual impairment prevalence were similar in children and adolescents with a screen time of compared with those with a screen time of . There was no significant effect modification for outdoor exercise time (Table 4).

Table 4

Associations of normalized difference vegetation index and soil-adjusted vegetation index in a buffer with prevalent visual impairment and visual acuity levels among children and adolescents, stratified by outdoor exercise time and screen time ().

Strata/exposure	Visual impairment		Left eye visual acuity		Right eye visual acuity
	OR (95% CI)a,b	p-Value for effect modification	β (95% CI)a,c	p-Value for effect modification	β (95% CI)a,c	p-Value for effect modification
Screen time/NDVI_500m per IQR increase (h/d)d		0.683		0.003		0.027
<1	0.95 (0.91, 0.98)		0.006 (0.000, 0.011)		0.006 (0.001, 0.012)
≥1	0.96 (0.92, 0.99)		0.017 (0.012, 0.022)		0.015 (0.010, 0.020)
Outdoor exercise time/NDVI_500m per IQR increase (h/d)d		0.101		0.992		0.538
<2	0.93 (0.89, 0.97)		0.012 (0.006, 0.017)		0.013 (0.007, 0.018)
≥2	0.97 (0.94, 1.00)		0.012 (0.007, 0.016)		0.010 (0.005, 0.015)
Screen time/SAVI_500m per IQR increase (h/d)d		0.711		0.002		0.025
<1	0.95 (0.92, 0.98)		0.005 (0.000, 0.010)		0.007 (0.001, 0.012)
≥1	0.96 (0.92, 0.99)		0.016 (0.011, 0.021)		0.015 (0.010, 0.020)
Outdoor exercise time/SAVI_500m per IQR increase (h/d)d		0.076		0.790		0.369
<2	0.93 (0.89, 0.96)		0.011 (0.006, 0.017)		0.013 (0.008, 0.018)
≥2	0.97 (0.94, 1.00)		0.011 (0.006, 0.015)		0.010 (0.005, 0.014)

Note: Effect estimates were calculated using generalized linear mixed models. , regression coefficient; CI, confidence interval; IQR, interquartile range; NDVI, normalized difference vegetation index; OR, odds ratio; SAVI, soil-adjusted vegetation index.

Models were adjusted for age, sex, ethnicity, parental highest education level, district/county-level gross domestic product, district/county-level population density, and urbanicity.

, , and suggests protective, no, and deleterious effects of greenness on visual impairment, respectively.

, , and suggests protective, no, and deleterious effects of greenness on visual acuity levels, respectively.

The IQR was 0.12 and 0.07 for and , respectively.

Exploring Potential Pathways Linking Greenness to Visual Acuity

Estimates from mediation analyses suggest that mediated a substantial part of the associations between greenness and visual impairment (indirect effect estimates of ; 95% CI: 51.2, 123.9% and ; 95% CI: 55.4, 96.2% for and , respectively) (Table 5). mediated only a small part of the association (indirect effect estimates of ; 95% CI: 6.2, 25.6% and ; 95% CI: 12.4, 64.0% for and , respectively). The estimated mediation of the associations were not significantly different by age, sex, or parent’s education, and there was no consistent pattern for differences in mediation of effect estimates between groups (Table 6; Table S9).

Table 5

Mediation of associations between normalized difference vegetation index, soil-adjusted vegetation index, and visual impairment among children and adolescents, by air pollutants ().

Exposures/potential mediators	Decomposition of total effect [% (95% CI)]a
	Mediation effect	p-Value	Direct effect	p-Value
NDVI_500m
PM_1	87.5 (51.2, 123.9)	<0.0001	12.5 (−23.9, 48.9)	0.50
NO_2	15.9 (6.2, 25.6)	0.001	84.1 (74.4, 93.8)	<0.0001
SAVI_500m
PM_1	75.8 (55.4, 96.2)	<0.0001	24.2 (3.8, 44.6)	0.02
NO_2	38.2 (12.4, 64.0)	0.004	61.8 (36.0, 87.6)	<0.0001

Note: Effect estimates were calculated by a two-way decomposition method employing the CAUSALMED procedure in SAS software. Mediators and were 3-y annual average concentrations at each school and were modeled as continuous variables. CI, confidence interval; NDVI, normalized difference vegetation index; , nitrogen dioxide; , particles with an aerodynamic diameter of ; SAVI, soil-adjusted vegetation index.

Adjusted for age, sex, ethnicity, parental highest education level, district/county-level gross domestic product, district/county-level population density, and urbanicity.

Table 6

Moderated mediation effect of on associations between normalized difference vegetation index, soil-adjusted vegetation index, and visual impairment among children and adolescents, by children’s age and sex and parental highest education level ().

Exposures/potential moderators	n	Decomposition of total effect [% (95% CI)]a
		Mediation effect	p-Value for moderated mediation	Direct effect	p-Value for moderated direct effect
NDVI_500m
Age (y)			0.64		0.64
6–11	34,128	73.3 (15.4, 131.1)		26.7 (−31.1, 84.6)
12–18	27,867	94.1 (30.7, 157.5)		5.9 (−57.5, 69.3)
Sex			0.25		0.25
Boys	31,934	70.9 (36.9, 104.9)		29.1 (−4.9, 63.1)
Girls	30,061	94.2 (73.4, 115.0)		5.8 (−15.0, 26.6)
Parents’ highest education level			0.56		0.56
Completed senior high school or below	44,209	89.6 (58.6, 120.6)		10.4 (−20.6, 41.4)
Completed junior college or above	17,786	75.2 (37.6, 112.9)		24.8 (−12.9, 62.4)
SAVI_500m
Age (y)			0.69		0.69
6–11	34,128	84.6 (51.8, 117.5)		15.4 (−17.5, 48.2)
12–18	27,867	92.3 (73.4, 111.2)		7.7 (−11.2, 26.6)
Sex			0.37		0.37
Boys	31,934	79.8 (58.1, 101.4)		20.2 (−1.4, 41.9)
Girls	30,061	93.0 (74.0, 111.9)		7.1 (−11.9, 26.0)
Parental highest education level			0.47		0.47
Completed senior high school or below	44,209	83.3 (73.4, 93.1)		16.7 (6.9, 26.6)
Completed junior college or above	17,786	73.0 (46.8, 99.3)		27.0 (1.0, 53.2)

Note: Effect estimates were calculated by a two-way decomposition method employing the CAUSALMED procedure in SAS software. Mediator was the 3-y annual average concentration at each school and was modeled as a continuous variable. CI, confidence interval; NDVI, normalized difference vegetation index; , particles with an aerodynamic diameter of ; SAVI, soil-adjusted vegetation index.

Adjusted for age, sex, ethnicity, parental highest education level, district/county-level gross domestic product, district/county-level population density, and urbanicity, except where the variable was stratified.

Discussion

In this large cross-sectional study, we investigated associations between school-based greenness and prevalent visual impairment and visual acuity levels, as well as potential mechanisms underlying the associations. Greater greenness surrounding schools, as indicated by NDVI and SAVI values, was associated with lower odds of visual impairment and higher mean values of visual acuity, and screen time might be a potential modifier of the associations. Mediation analyses suggested that these associations may be partly explained by lower ambient and levels around the schools.

To the best of our knowledge, this is the first study exploring the association of greenness with visual acuity and the potential mechanisms. We are aware of only one similar study investigating the association between greenness and myopia, as measured by spectacles use among 2,727 children in Barcelona, Spain (Dadvand et al. 2017b). Consistent with our findings, that study reported significant associations between higher levels of greenness surrounding homes and schools and a lower risk of using spectacles (Dadvand et al. 2017b). The strengths of that study were its longitudinal nature and comprehensive exposure assessment, which considered greenness at homes, at schools, and during commuting; however, the study was limited to children from only one city. The present study includes a nationwide sample representative of Chinese children and adolescents with a much larger sample size. In addition, we objectively measured visual acuity levels using a clinically validated and standardized protocol, rather than relying on self-reported use of spectacles as a surrogate of myopia, and hence minimized outcome misclassification.

Human eyes are directly exposed to air pollutants, and evidence has indicated that air pollution exposure may increase the odds of myopia (Dadvand et al. 2017a; Ruan et al. 2019; Wei et al. 2019) by inducing systematic and local inflammation and oxidative stress on eyes (Jung et al. 2018; Torricelli et al. 2011), and by impairing retinal microvasculature function, neuro-activity, and axial length (Adar et al. 2010; Herbort et al. 2011; Louwies et al. 2013; Njie-Mbye et al. 2013). Because vegetation could reduce levels of some air pollutants, it is often hypothesized that greenness exposure may benefit human health through reducing air pollution levels (Markevych et al. 2017). We performed mediation analysis to test this hypothesis and estimated that and significantly explained the association between greenness and visual impairment.

We cannot compare this finding with other studies given that no other study has explored the mechanisms of the association between greenness and myopia. The only other study investigating greenness and myopia mentioned above did not adjust for air pollution (Dadvand et al. 2017b), although a strong signal for an association between air pollution and spectacles usage was later reported in the sample (Dadvand et al. 2017a). However, it is noteworthy that the validity of our mediation estimates depends on the validity of the assumptions of the mediation analysis, which include; a) causal and unconfounded effects of greenness exposure on air pollution; b) causal and unconfounded effects of air pollution on visual impairment; and c) no interaction between greenness and air pollution. Because our study was a cross-sectional design, uncontrolled confounding and misclassification (for both greenness and mediators) could not be avoided, thus the accuracy of the mediation estimates could not be verified in our study.

Green spaces are associated with increased physical activity in children (Markevych et al. 2017), and a study of Barcelona schoolchildren reported that greater outdoor greenness was associated with more time spent playing in green spaces (Amoly et al. 2014). It has been well documented that outdoor physical activity is a strong protective factor for developing myopia (Morgan et al. 2012). Further, high greenness levels have been linked to less recreational screen time (Dadvand et al. 2014), which is a strong risk factor for myopia.

Given that the crude data on outdoor exercise and screen time that could not differentiate whether the activities took place at schools or homes, we did not explore the potential mediation effects on the greenness–visual impairment association. However, students generally have little control over their mobility and activities while at school, thus the mediating effects of outdoor exercise and screen time on the association between school greenness and visual impairment would be minimal. Alternatively, we explored potential effect modification of outdoor exercise time and screen time and found stronger associations of and with visual acuity levels in children and adolescents who had more screen time. The exact reason for this difference is unclear, yet it may be due in part to the apparent association of screen time with visual impairment in our study. A possible explanation for this phenomenon might be that children and adolescents with visual impairment reduced their screen time as part of a myopia treatment plan. However, because of the cross-sectional nature of our study, we were unable investigate the possibility.

Several epidemiological studies have also documented that higher greenness exposure was associated with reduced blood pressure and psychological stress in children and young adults (Dzhambov et al. 2019; Markevych et al. 2014; Xiao et al. 2020). A study in Italian children reported that elevated blood pressure was a significant determinant of ocular hypertension, which is involved in the initiation and development of myopia (Pileggi et al. 2021). Evidence also indicates that psychological stress could lead to myopia by causing tension of the tissues and muscles around the eyes, altering the shape of the eyeball (Sabel et al. 2018). Therefore, greenness might also benefit visual acuity levels by positively affecting blood pressure and mental health. Still, we could not explore these potential mechanisms owing to limited data on blood pressure and mental health. Thus, what is driving the observed school greenness–visual impairment associations is still not entirely clear and needs to be uncovered by future well-designed studies.

A major strength of our study is the large nationally representative sample of children and adolescents with a high response rate. We objectively measured visual acuity levels using a clinically validated and standardized protocol, which greatly reduced the probability of outcome misclassification error. In addition, we carefully selected confounders from many candidates by employing a DAG, which allowed us to adjust for a parsimonious but comprehensive set of covariates to minimize confounding in our results. Further, we explored potential mechanisms underlying the association between greenness and visual acuity by estimating the mediating effects of air pollution. Finally, the effect estimates were generally robust to the sensitivity analyses we performed. An exception is that when Hunan and Liaoning provinces were excluded one at a time, the effect estimates were consistently closer to the null, which might be related to the lower prevalence of visual impairment in Hunan and Liaoning provinces than in other study provinces and municipalities or to differences in population density, genetic background of the residents, or lifestyle differences in the different study areas (Table S10). The exact reasons for the discrepancy were unclear and thus need to be explored in a future study.

Our study is not without limitations. First, owing to its cross-sectional nature, we cannot definitively indicate a temporal relationship between greenness exposure and visual impairment. However, the probability of reverse causality—that children and adolescents with lower visual acuity moved to less green areas—seems unlikely. Second, we did not have the boundaries of school grounds needed to measure greenness specifically for schools. In addition, although we used two satellite-based indices within different buffers to assess greenness levels, the indices are sensitive to season and cannot differentiate between quality, type, and structure of vegetation. Further, we assessed greenness and air pollutants at the school level, but we were unable to include individual-level greenness and air pollutants at the residence and those between home and school owing to unavailable information on residential address. This might have caused nondifferential measurement bias and thus biased the effect estimates toward null (Hutcheon et al. 2010). In addition, the school-level greenness exposure precluded us from incorporating school as a random effect (we only incorporated province/municipality as a random effect), despite our data being clustered primarily into schools. Moreover, we used only summer images to calculate greenness with the aim of grasping maximum vegetation contrasts, which might have misclassified the greenness exposure because students generally have a summer holiday break from school. However, we performed an additional sensitivity analysis by calculating average greenness levels in the 10 months when children were at school and found that the results were consistent with those using summer images. Third, we did not perform cycloplegic refraction (i.e., the gold standard) to measure visual acuity, but rather a simple vision test, which might have caused some outcome misclassification. However, simple vision tests have high sensitivity and specificity in screening for myopia (Tong et al. 2002). A recent national study in Chinese children and adolescents showed that the prevalence of visual impairment was highly correlated with myopia prevalence () (Jan et al. 2019), justifying our approach to outcome assessment. Fourth, crude data on outdoor exercise time and screen time did not allow us to distinguish the locations of these activities, for example, whether the children were at home, at school, or in green spaces. We were thus unable to explore their potential mediating roles in the pathways between greenness and visual acuity although we tested for their potential modifying effects on and as mediators. Fifth, data on covariates and some mediators were collected using a self-completed questionnaire, thus biases due to misclassification remain possible. Finally, our effect estimates for greenness indexes and visual acuity levels were small (0.011–0.012 per IQR increase in and ), and the potential clinical significance remains uncertain.

Conclusions

In summary, higher greenness surrounding schools was associated with lower odds of visual impairment and higher visual acuity levels in a large sample of Chinese children from seven different provinces and municipalities. Results of mediation analyses suggest that the association may be partially explained by lower and levels around greener schools. However, given the limitations of our study, these results need to be validated by longitudinal epidemiological studies with more sophisticated measures of green space, air pollution, screen time, and outdoor exercise.

Click here for additional data file.

Click here for additional data file.