Assessing the Impact of School-Based Greenness on Mental Health Among Adolescent Students in Ontario, Canada

Neighbourhood greenness has been frequently associated with improved mental health in adulthood, yet its impact among youth is less clear. Additionally, though youth spend large portions of time at school, no study has investigated associations between school-based measures of greenness and students’ mental health in Canada. We addressed this gap by linking participant responses from the 2016–2017 Ontario Student Drug Use and Health Survey to school-based features of the built environment. Our analyses included 6313 students, ages 11–20. Measures of greenness were the mean and max of the annual mean Normalized Difference Vegetation Index within 500 m and 1000 m from the centroid of the school postal code. Measures of mental health included: serious psychological distress (Kessler 6-item Psychological Distress Scale), self-rated mental health (using a five-point Likert scale), suicide ideation, and suicide attempt. In our study population, the prevalence of serious psychological distress and low self-rated mental health was 16.7% and 20.3%, respectively. Suicide ideation was reported by 13.5% of participants, while 3.7% reported a suicide attempt. Quantity of greenness was similar between schools in the lower and upper quartiles. In logistic regressions, we found no association between objective school-based greenness and mental health, as assessed by multiple measures, both before and after adjustment. Null findings held true after stratification by season, as well. Whether other characteristics of school greenness (such as type, quality, or access and use) are more impactful to students’ mental health should be a focus of future analyses.


Introduction
The transition between childhood and adulthood is a period of life characterized by rapid physical, psychological, and cognitive development, and lays a foundation for future mental health and resilience [1,2]. An estimated 70% of mental health problems first appear during this developmental period [3]. According to a recent provincial survey, approximately 1 in 3 Ontario middle-and high-school students (grades 7-12) indicated moderate to serious psychological distress, with prevalence increasing greatly by grade and over time [4]. Mental illness is responsible for the most years of life lost due to disability and premature mortality of any other condition among Canadian youth aged 10-24 [5].

Data Sources
The OSDUHS was first administered in 1977 and is the longest-running school survey of adolescents in Canada. Detailed description of the methods used in the survey can be found elsewhere [4]. Briefly, the OSDUHS is a biennial cross-sectional survey of Ontario students enrolled in grades 7-12 (11-20 years old) and is completed anonymously. The survey employs a stratified (region and school-level), two-stage (school and class) cluster sample design, devised to produce a representative sample of the approximately 1 million students in grades 7-12 in publicly-funded schools in Ontario. For our study, we used student data from the 2016-2017 cycle of the OSDUHS to take advantage of recently-constructed built environment exposure surfaces. In this cycle, 11,435 students drawn from 764 classes in 214 schools participated in the self-administered, anonymous survey between November 2016 and June 2017. This represented a response rate of 61%, 94%, and 61% for students, classes, and schools, respectively. To include as many topics as possible within the confines of a fixed-class period and while reducing time-burden, two questionnaire forms differing in half of the items were randomized among students; only one of the two included information on mental health, therefore our analysis was restricted to this half-sample of adolescents (N = 6313). Built environment data from Ontario in 2016 (such as greenness and neighbourhood-level socio-economic status) were obtained from the Canadian Urban Environmental Health Research Consortium (CANUE).
The Research Ethics Boards at the Centre for Addiction and Mental Health and York University, as well as Research Review Committees at 31 school boards, approved the 2016-2017 OSDUHS protocol. This specific study was approved by Carleton University's Research Ethics Board (CUREB).

Psychological Distress
The Kessler 6-item Psychological Distress Scale (K6) was employed as a validated measure of non-specific psychological distress in the past month, including symptoms of anxiety and depression [4,15]. The K6 has previously been shown to be a valid assessment tool of psychological wellbeing among adolescents [16,17], and showed high internal reliability in our study sample (Cronbach's α = 0.88). Each item was assessed on a five-point Likert scale, where responses were scored from 0-4 and summed together to a range of 0-24 for students who completed all six items; serious distress was classified as a total score of 13 or more, as suggested by Kessler and colleagues [15].

Self-Rated Mental Health
On a five-point Likert scale, participants were asked to rate their mental or emotional health. Responses of fair and poor were categorized as low self-rated mental health, while those of good, very good, and excellent were categorized as high. This measure of general mental health has previously been validated in Canada [18].

Suicide Ideation or Attempt
Suicide ideation or attempt were based on 'yes' or 'no' responses to: "In the last 12 months, did you seriously consider attempting suicide?" and "In the last 12 months, did you actually attempt suicide?", respectively. Similar wording has been used elsewhere [19,20].

School-Based Greenness (Exposure)
School-based greenness was assessed based on the mean or maximum of the annual mean Normalized Difference Vegetation Index (NDVI) within 500 m or 1000 m from the centroid of the six-character school postal code (representing green space within a 10-min or 20-min walk, respectively). Many urban and suburban schools in Ontario have their own postal codes, whereas rural schools are likely to be included in the postal code of the surrounding area. Nevertheless, due to being weighted to the spatial distribution of populations within the postal code boundary [21], the centroid is an accurate representation of the location of residents within the area and should thus approximate the school location, both in urban and rural areas. As NDVI is a widely-used measure of greenness, we employed this measure to be able to compare our results with similar studies in different settings and age-ranges; the index employs the difference in radiation between red and near-infrared wavelengths-used by photosynthetic plants to absorb and emit radiation, respectively-to identify green vegetation using satellite data [22]. The measure of greenness ranges from −1 (no greenness) to 1 (all green) and was categorized into quartiles. NDVI data from 2016 were taken from the United States Geological Survey Landsat 8 satellite with a spatial resolution of 30 m and accessed via Google Earth Engine [23][24][25][26]. NDVI data from the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the TERRA satellites and accessed via Google Earth Engine [23,27] were also used in a sensitivity analysis. More information about the greenness measures can be found elsewhere [28]. Season, determined by date of questionnaire completion, was classified as fall/winter (November-February) or spring/summer (March-June). No questionnaires were completed between July and October.

Socio-Economic Status
The Material and Social Deprivation Index (MDSI), a proxy for neighbourhood-level socio-economic status (SES) which included information on six socio-economic indicators (such as neighbourhood income, education, proportion of single-parent families, etc.), was linked to the dataset [24,29,30]. The material and social deprivation dimensions were each categorized into tertiles (from low (T1) to high (T3) deprivation) and combined into one measure with three categories: low, medium, and high material and social deprivation, similar to methods previously published [31]. The OSDUHS questionnaire also included self-reported socio-economic status (10-point scale based on how 'well off' the participant believed their family to be). We grouped the scale by twos to create a five-point scale, since the five-point scale was significantly associated with both our main outcome (serious psychological distress) and some measures of greenness, while the 10-point version was not.

Walkability
The walkability of school neighbourhoods in 2016 was compiled by researchers through the Canadian Active Living Environments (Can-ALE) dataset and made available on the CANUE data repository [32,33]. The continuous measure of degree of walkability was calculated based on the summed z-scores of points of interest, intersection density, dwelling density, and transit around the neighbourhood centroid [34]. A z-score greater than zero would represent a neighbourhood that is more walkable than the Canadian average.

Demographics and Other Covariates
Additional covariates selected from the OSDUHS survey were age (categorized as: 11-12, 13-14, 15-16, and ≥17 years); sex; region (Greater Toronto Area, including Toronto, Durham Region, York Region, Peel Region, and Halton Region; Northern Ontario, including Parry Sound District, Nipissing District, and farther north; Western Ontario, including Dufferin County and farther west; and Eastern Ontario, including Simcoe County and farther east); home language (English and/or French only, as well as Other); years since immigration to Canada (Canadian-born, ≤2 years, 3-5 years, 6-10 years, and ≥11 years); physical activity; outdoor play; frequency of being bullied, including traditional and cyberbullying; and ethno-racial background (White, Asian, Black, Indigenous, Latin American, and multiple). Physical activity (PA) was assessed with the question: "On how many of the last 7 days were you physically active for a total of at least 60 min each day?". Responses equal to seven days were coded as 'meeting guidelines' based on Canadian PA guidelines [35]; all other responses were coded as 'not meeting guidelines'. Outdoor play was defined as being physically active outside after school at least one day over the last five school days, excluding travel time from school. For ethno-racial background, participants could report more than one category to which they identified; if more than one was reported (even if the second category was 'not sure'), the participant was placed in the 'multiple' ethno-racial background category. However, if the respondent identified as Indigenous, they were placed in this category regardless of other ethno-racial backgrounds reported due to the unique history and culture of this group in Canada.

Statistical Analyses
All analyses were conducted using SAS Enterprise Guide 5.1 (SAS Institute, Cary, NC, USA). Descriptive statistics were calculated and presented with the distribution of mean greenness at 500 m across these categories. Logistic regressions, which applied the generalized estimating equations (GEE) method with an independent covariance structure to account for the clustering of students in schools, were employed to examine the adjusted odds of having serious psychological distress, a proxy for mental health, among students by objective school-based greenness (defined by four NDVI measures in separate models). Models were adjusted for age, sex, ethno-racial background, self-reported SES, and season. We adjusted for these variables based on their associations with both the exposure and outcome (i.e., due to confounding), except in the case of sex, which adjusted for a large imbalance in the outcome between males and females. Multiple sensitivity analyses were conducted; models were re-run with different measures of mental health (self-rated mental health, suicide ideation, or suicide attempt), different greenness data (MODIS satellite), different combinations of adjustment factors (including home language (as a proxy for minority status), years in Canada, bullying (including traditional and cyber), and walkability) and interaction terms (between greenness and: SES (self-reported SES or MDSI tertiles), physical activity, outdoor play, region, or season), and stratified by season to determine whether these changes affected overall conclusions.

Results
Sample characteristics are presented in Table 1. The majority of participants were 15 years old or older, White, or female. Most students did not meet the Canadian physical activity guidelines but played outside at least once in the previous five school days. Most spoke English and/or French exclusively at home, were Canadian-born, and self-reported high SES. Approximately 16.7% and 20.3% of students reported serious psychological distress and low self-rated mental health, respectively. Suicide ideation was indicated by 13.5% of participants, while 3.7% reported a suicide attempt. School-based greenness was similar between the lower and upper quartiles of each measure (average interquartile range of 0.08 in a scale ranging from −1 to 1; Figure 1). low (T1) to high (T3) deprivation) and combined into one measure with three categories: low, medium, and high material and social deprivation. g Greater Toronto Area includes: Toronto, Durham Region, York Region, Peel Region, and Halton Region; Northern Ontario includes: Parry Sound District, Nipissing District, and farther north; Western Ontario includes: Dufferin County and farther west; and Eastern Ontario includes: Simcoe County and farther east. h Serious psychological distress is based on a score ≥13 on the Kessler 6-item Psychological Distress Scale (K6). i Low self-rated mental health based on responses of fair and poor on a five-point scale of mental or emotional health.  Data from logistic regressions are shown in Table 2. Odds of serious psychological stress increased with age and were significantly higher for females compared to males (data not shown). When looking at ethno-racial background, only Indigenous students reported more serious psychological stress compared to White students (data not shown). A very strong relationship was observed according to self-reported socio-economic status (SES), with those in the lowest category almost 10 times more likely to report serious psychological stress compared to those in the highest category (data not shown). None of the models showed an association between any greenness measure and serious psychological distress. Sensitivity analyses replicated our null findings when the models were re-run with different measures of mental health (Table 2), different greenness data, and different combinations of adjustment factors and interaction terms (data not shown). Furthermore, stratification by season produced similar null findings. In a post hoc sensitivity analysis, employing the 10-point SES scale-instead of the five-point version-did not change the results of the logistic regressions. In a separate post hoc sensitivity analysis, we refit the models using the workable exchange covariance matrix and found that the standard errors (and associated confidence intervals) were essentially unchanged. Due to the nature of the data, and students being provided the surveys at the same time, we deemed the autoregressive covariance matrix to be unsuitable. We further considered multi-level modelling, but again, this analytic approach did not fundamentally change our findings.

Discussion
In this study, we examined the relationship between school-based greenness and mental health in a large sample of middle-and high-school students in publicly-funded schools in Ontario. To our knowledge, this is the first study to assess the impact of objective greenness around the school neighbourhood on mental health in Canada. We found no association between objective quantity of greenness around the school neighbourhood and students' mental health, as assessed by multiple measures, both before and after adjustment.
Several potential theories may explain why the school greenness-mental health relationship was null in our study. First, not all greenness is equal. A study of approximately 5200 Washington State residents found no relationship between overall residential-neighbourhood greenness (combining urban greenspaces, forests, rangelands, agricultural lands, and wetlands) and measures of mental or general health; however, when stratified by greenspace type, significant positive associations emerged between forests, urban greenspaces, and mental health [36]. Similarly, a study employing data from 3103 counties in the United States found that forests and shrubs were negatively associated with health-care spending, though grass-land cover had no association to the outcome [37]. Further, in an Australian study, quality, not quantity, of greenspaces was associated with low psychological distress, regardless of use [38]. These data suggest that the type and quality of greenness are important factors that modulate the impact of greenness on mental health. Our study was limited by employing a measure of overall greenness alone, and one that was objective, rather than subjective, which could not address subjective quality [6].
Second, we found relatively little variation in greenness between the schools in the lower and upper quartiles. Since most schools are similarly 'green', our ability to identify a significant association may have been limited.
Third, greenspace access and use may be important mediators of the association between greenness and health. While adolescents spend a large portion of time in school, most school hours are spent indoors and not out in the school neighbourhood. Although the presence of greenness alone can improve mental health through perceived restorative quality and reduction of air and noise pollution, greenspace use may also increase social cohesion and physical activity [39]. However, our study could not assess access nor use of greenspace among students, potentially masking any association. Still, it is presently unclear whether the presence of greenspace or its access and use is more impactful on mental health outcomes [40].
Additionally, there are other limitations to our study. We employed the centroid of the school postal code instead of the schoolyard boundary due to not having access to the shape of every schoolyard in the study. Therefore, smaller buffers may include part of the schoolyard, whereas the larger buffers are more likely to include the entire schoolyard. We found little difference in the results between the larger and smaller buffers, suggesting this is not a large source of bias. Further, we did not delve into how green design may be applied to school environments to increase mental health among students. These limitations, as well as the others outlined above, should be taken into consideration in future studies.
Lastly, notwithstanding the limitations of our study described above, school greenness may simply not be an important influence on students' mental health. Exposure to school-based greenness is limited in Ontario by the seasons during the school year; students are exposed to little greenness during the late fall and winter, and they spend summers away from school, during which time greenness and greenspace use are presumably at their peak. Furthermore, the majority of a student's week is spent at home, where residential greenness may be a more salient risk factor for mental health. Finally, other factors may be more impactful to mental health during adolescence, such as biological factors, social disadvantage, family conflicts, and bullying [41], leaving a smaller role for greenness in this relationship.

Conclusions
Our study is the first to assess the relationship between objective school-based measures of greenness and students' mental health in Canada. Though we found no association between these factors, it is still far from clear whether school greenness can impact students' mental health due to the lack of research in this area. Future studies should elucidate this relationship, while focussing on overcoming the key limitations of our study in order to improve our understanding of this issue.