Early-Life Exposure to Green Space and Mid-Childhood Cognition in the Project Viva Cohort, Massachusetts

Abstract

The association between early-life greenness and child cognition is not well understood. Using prospective data from Project Viva (n = 857) from 1999–2010, we examined associations of early-life greenness exposure with mid-childhood cognition. We estimated residential greenness at birth, early childhood (median age 3.1 years), and mid-childhood (7.8 years) using 30-m resolution Landsat satellite imagery (normalized difference vegetation index). In early childhood and mid-childhood, we administered standardized assessments of verbal and nonverbal intelligence, visual-motor abilities, and visual memory. We used natural splines to examine associations of early life-course greenness with mid-childhood cognition, adjusting for age, sex, race, income, neighborhood socioeconomic status, maternal intelligence, and parental education. At lower levels of greenness (greenness <0.6), greenness exposure at early childhood was associated with a 0.48% increase in nonverbal intelligence and 2.64% increase in visual memory in mid-childhood. The association between early-childhood greenness and mid-childhood visual memory was observed after further adjusting for early childhood cognition and across different methodologies, while the association with nonverbal intelligence was not. No other associations between early life-course greenness and mid-childhood cognition were found. Early childhood greenness was nonlinearly associated with higher mid-childhood visual memory. Our findings highlight the importance of nonlinear associations between greenness and cognition.

Abbreviations:

BC

black carbon

CI

confidence interval

IQ

intelligence quotient

KBIT-2

Kaufman Brief Intelligence Test

NDVI

normalized difference vegetation index

NSES

neighborhood socioeconomic status

SD

standard deviation

WRAML2

Wide Range Assessment of Memory and Learning

WRAVMA

Wide Range Assessment of Visual-Motor Abilities

Increasing empirical evidence suggests that exposure to natural vegetation—also referred to as green space or greenness—may influence health, including cognition (1). While the exact mechanisms are not well understood, literature suggests that exposure to greenness may benefit cognition by mitigating exposures to air pollution, relieving mental stress, and promoting physical activity and social interaction (2). The attention restoration theory posits that natural environments enable us to overcome the mental fatigue associated with modern life by evoking a type of attention that does not require cognitive effort, thus restoring the brain’s capacity to direct attention (3). Exposure to greenness may be especially pertinent during early-life periods as they are times of rapid brain growth and development (4).

Studies have reported associations between exposure to greenness and better cognitive function in adults (5) and a wide range of diagnosed neurobehavioral outcomes in children, including attention deficit hyperactivity disorder (6, 7) and autism spectrum disorder (8). While several studies have reported associations between greenness and cognition in children (1), there are few that are longitudinal (9–13) and that have evaluated sensitive periods of greenness exposure (7, 14, 15). To address this gap in the literature, we assessed greenness exposure during early life course (birth, early childhood, and mid-childhood) and its association with mid-childhood cognitive function. We used longitudinal data from the Project Viva cohort to examine the extent to which associations between early life-course greenness and cognition at early and mid-childhood varied by sensitive periods. We also evaluated whether associations were mediated by reduction in air pollution or increase in physical activity.

METHODS

Study population

Project Viva is a prebirth cohort that enrolled pregnant women during 1999–2002 at Atrius Harvard Vanguard Medical Associates in eastern Massachusetts (16). Demographic, health, and behavioral information were collected via in-person interviews or mailed questionnaires. We assessed neurodevelopment at early childhood (median age = 3.1 (standard deviation (SD), 0.3) years) and mid-childhood (median age = 7.8 (SD, 0.8) years). Of the 2,128 mother-child pairs enrolled in the cohort, 1,089 had complete information on cognitive outcomes, 1,047 on residential greenness exposure, and 857 on all covariates for this analysis. Study protocols were approved by the Harvard Pilgrim Health Care review board. Mothers provided written informed consent, and children provided verbal assent at the mid-childhood visit.

Exposure assessment

To ascertain participant’s exposure to greenness at each geocoded residential address at birth and in early and mid-childhood, we used the normalized difference vegetation index (NDVI). NDVI is a satellite-based objective indicator of the quantity of ground green vegetation (range, −1 to 1). Negative values represent water, values around zero correspond to barren areas of sand, and values approaching 1 indicate tropical rainforests. NDVI has been widely used in previous studies of greenness and health (17–19). Since NDVI reaches its maximum and highest level of geographic variation during the height of the summer, we used Landsat satellite data at 30-m resolution from July 1999 to 2010. Average greenness at 90-m and 270-m buffers were estimated using the focal function in the R raster package (https://cran.r-project.org/web/packages/raster/). We used continuous NDVI at the 90-m buffer for the main results, and NDVI interquartile range and the continuous 270-m buffer for sensitivity analyses.

Outcome assessment

At early childhood and mid-childhood in-person visits, trained Project Viva staff administered standardized and well-validated assessments of cognitive development (16, 20–22). At the early childhood visit, staff administered the Peabody Picture Vocabulary Test (PPVT-III) to assess vocabulary comprehension (20) and the Visual Motor subtest of the Wide Range Assessment of Visual Motor Abilities (WRAVMA) (21) to assess visual-motor, fine-motor, and visuospatial skills. At the mid-childhood visit, staff administered the Kaufman Brief Intelligence Test (KBIT-2) (22), the Visual Motor subtest of the WRAVMA (21), and the Visual Memory Index of the Wide Range Assessment of Memory and Learning (WRAML2) (23). The verbal intelligence quotient (IQ) subscore of the KBIT-2 test measures crystalized intelligence (i.e., vocabulary), while the nonverbal IQ subscore measures fluid intelligence (i.e., ability to perceive relationships). The Visual Motor subtest is a specific component of the WRAVMA that involves drawing/copying increasingly complex figures (21). Visual Memory was derived as the sum of standardized picture and design memory subtests (23). KBIT-2 IQ and WRAVMA Visual Motor scores are standardized to a mean of 100 (SD, 15); WRAML2 design memory and picture memory subtests are each standardized to a mean of 10 (SD, 3) using published reference data (21). Neurobehavioral functional domains measured by each of the cognitive assessments are outlined in Web Table 1 (available at https://doi.org/10.1093/aje/kwab209).

Covariates

Covariates included characteristics of the child (sex, race (White vs. non-White), and age at cognitive testing), mother’s intelligence, parents’ education (high-school degree, some college, or at least a college graduate), annual household income at enrollment (in $, <40,000, 40,000–70,000, >70,000), and neighborhood median annual income (at Census-tract level at birth). Neighborhood population density, used as a proxy for urbanicity, was assessed from residential address at birth using nationwide land-use data derived from satellite images with approximately 30-m resolution (24). Mothers’ intelligence was assessed using the KBIT-2 (22) (at the mid-childhood visit), which is a valid and reliable measure for ages 4–90, standardized using a representative US sample.

We analyzed air pollution and physical activity measured at early childhood as mediators. For traffic-related air pollution, we averaged daily black carbon (BC) residential exposure over 365 days prior to the child’s cognitive assessment with a land-use regression model (25). Physical activity was reported by the mother as a response to the question: “In the past month, on average, how many hours a day is your child involved in active play (e.g. running)?” (16).

Statistical analysis

We assessed correlations among exposures, outcomes, and covariates with Pearson correlation. Statistical significance was assessed at α = 0.05. We examined prospective associations of greenness at birth with early childhood cognition (Web Figure 1, line L1) and greenness at birth and early childhood with mid-childhood cognition assessments (Web Figure 1, lines L2–L3). We also assessed cross-sectional associations of greenness at early childhood with early childhood cognition (Web Figure 1, line C1) and greenness at mid-childhood with mid-childhood cognitive assessments (Web Figure 1, line C2). To assess the shape of exposure-outcome associations, we fitted generalized additive models for continuous exposures. Penalized splines suggested deviations from linearity (degrees of freedom > 1) for associations with all cognition measurements. Thus, we used natural splines with 3–4 knots based on Akaike’s information criterion. Unless otherwise indicated, all variables were evaluated as continuous. To ease interpretation of the splines, we evaluated change in cognition scores for a percentile increase in exposure.

All models adjusted for potential confounders based on prior evidence (9, 15, 26), directed acyclic graphs (27), and greenness exposure in all preceding age periods but not in subsequent age periods (28). Thus, models assessing exposure at birth adjusted for confounders, models assessing exposure at early childhood adjusted for confounders and birth exposure, and so on. Plotted splines were assessed visually. Sensitivity analyses were conducted excluding outliers for cognitive scores (>3 SD) to explore their influence. Further, in order to contribute to meta-analytical work, we estimated change in outcome scores associated with 1 interquartile-range increase in average NDVI, adjusting for a minimal set of confounders, matching prior work (9, 15, 26, 29), as well as additional confounders available in Project Viva.

In addition, we assessed effect measure modification of associations of greenness with cognition by child’s sex, parental education (both parents with college education vs. none or only 1 parent with college education), neighborhood socioeconomic status (NSES; tertiles), and neighborhood population density (tertiles) using stratified analyses.

Greenness has been associated with decreased levels of air pollution (19, 30) and increased levels of physical activity in children (31); we used a causal mediation framework (32–34) to determine whether the effect of greenness (at early childhood) on mid-childhood cognitive measures was mediated through pathways represented by air pollution and physical activity (both measured at early childhood). We used BC as a marker of traffic pollution following previous studies on the association between air pollution and cognition in children (35, 36). We used the “mediation” package in R (R Core Team, R Foundation for Statistical Computing, Vienna, Austria), which combines estimates from the mediator and outcome models to estimate natural direct and indirect effects (37). Mediation analysis requires the assumptions that there are no unmeasured exposure-outcome, mediator-outcome, or exposure-mediator confounders and no mediator-outcome confounders affected by exposure. While these assumptions are not verifiable, we believe ours are reasonable since we have included major confounders in our mediation analyses. We used 1,000 Monte Carlo draws for nonparametric bootstrap confidence intervals (CIs). Natural direct and indirect effects are reported in our results. Statistical analyses were performed in R, version 3.4.0 (R Core Team) (38). Code to run all analyses and reproduce our results is available on GitHub (GitHub Inc., San Francisco, California) (39).

RESULTS

A total of 857 participants provided data for this analysis. The mean age at mid-childhood cognitive assessment was 7.85 (SD, 0.77) years (Table 1). Overall, 71% of participants were White, but this percentage increased among the higher quartiles of greenness, as did the percentage of parents who had at least a college education. Less than half (44.72%) of the households in the lowest greenness quartile reported having a household income over $70,000/year at enrollment, compared with 85.96% in the highest quartile. Median neighborhood household income at birth varied across greenness quartiles from $46,557 to $69,697 (Table 1). In the lowest greenness quartile, the maternal intelligence score was 104.8 (SD, 16.4), slightly above the standardized mean of 100, while in the highest quartile, it was 113.0 (SD, 11.6), about 1 standard deviation higher than the reference population average. In mid-childhood, the average score for verbal IQ was 114 (SD, 14), nonverbal IQ 107 (SD, 17), visual-motor 93 (SD, 17), and visual memory 17 (SD, 4) (Table 1). All mid-childhood cognition scores were on average higher than the reference population mean, except for visual-motor abilities, which were below the reference population average.

Table 1.

Characteristics of 857 Children With Information on Greenness Exposures and Mid-Childhood Cognitive Outcomes, Project Viva, Massachusetts, 1999–2010

	NDVI Early Childhood
Variable	Quartile 1		Quartile 2		Quartile 3		Quartile 4		Overall
	%	Mean (SD)	%	Mean (SD)	%	Mean (SD)	%	Mean (SD)	%	Mean (SD)
Female sex	43.22		58.42		49.32		54.04		51.34
White race	56.78		61.88		78.73		84.68		71.30
Mother’s education level
Some college	29.15		23.76		19.46		11.06		20.42
College degree or more	60.80		71.78		77.38		86.38		74.68
Father’s education level
Some college	26.63		26.24		19.46		11.91		20.65
College degree or more	54.77		61.39		73.30		82.55		68.73
Household income at enrollment, $
40,000–70,000	29.15		25.25		22.62		10.64		21.47
>70,000	44.72		59.9		71.95		85.96		66.63
Age, years		7.87 (0.84)		7.83 (0.72)		7.87 (0.77)		7.83 (0.73)		7.85 (0.77)
Neighborhood median household income, $		46,557.59 (17,005.33)		54,033.61 (17,757.75)		64,709.90 (18,974.18)		69,697.94 (21,489.68)		59,355.23 (21,004.38)
Maternal intelligence score		104.79 (16.44)		106.08 (14.63)		109.5 (13.52)		112.97 (11.56)		108.55 (14.37)
Exposure to greenness at birth		0.4 (0.12)		0.52 (0.09)		0.59 (0.12)		0.62 (0.16)		0.54 (0.15)
Exposure to greenness at early childhood		0.38 (0.08)		0.53 (0.03)		0.63 (0.02)		0.72 (0.04)		0.57 (0.14)
Exposure to greenness at mid-childhood		0.46 (0.14)		0.57 (0.08)		0.64 (0.07)		0.72 (0.07)		0.6 (0.13)
Mediators
Air pollution (black carbon) exposure		0.72 (0.16)		0.57 (0.14)		0.49 (0.11)		0.43 (0.12)		0.55 (0.17)
Physical activity, hours per day		2.8 (1.64)		2.72 (1.55)		3.05 (1.62)		2.95 (1.61)		2.89 (1.61)
Cognition measures at early childhood
Vocabulary		101.81 (14.93)		104.26 (13.87)		105.16 (13.2)		107.4 (13.04)		104.88 (13.81)
Motor abilities		101.8 (11.33)		103.03 (11.08)		102.57 (10.15)		103.43 (11.04)		102.77 (10.88)
Cognition measures at mid-childhood
Verbal intelligence score		109.87 (16.4)		113.66 (13.28)		114.36 (12.98)		116.97 (12.16)		113.87 (13.92)
Nonverbal intelligence score		104.85 (17.59)		108.23 (15.67)		106.85 (16.25)		109.59 (16.43)		107.46 (16.55)
Visual-motor abilities		93.62 (17.86)		90.2 (15.04)		92.86 (17.52)		95.28 (16.79)		93.07 (16.92)
Visual memory		16.09 (4.64)		17.34 (4.22)		17.06 (4.41)		17.78 (4.21)		17.1 (4.4)

Abbreviations: NDVI, normalized difference vegetation index; SD, standard deviation.

The correlation between greenness exposures within individuals across time ranged between 0.50 and 0.74 (Web Figure 2). We did not find significant associations between greenness exposure at any time point with early childhood cognition (Web Figure 3), but we did with mid-childhood cognitive measures. We observed positive nonlinear relationships between greenness at early childhood and mid-childhood nonverbal IQ as well as mid-childhood visual memory, accounting for confounders (Figure 1). In early childhood, at greenness levels below the median of 0.6, higher greenness was associated with higher levels of mid-childhood nonverbal IQ; specifically, an increase from the 25th to the 50th NDVI percentile was associated with a 0.48% increase in mid-childhood nonverbal IQ. Above an NDVI of 0.6, there was no clear evidence of additional improvement in mid-childhood nonverbal IQ (Figure 1E). Similarly, increasing greenness at early childhood was associated with higher mid-childhood visual memory, where an increase from the 25th to the 50th NDVI percentile was associated with a 2.64% increase in mid-childhood visual memory (Figure 1H). No evidence of consistent improvement was observed at levels of greenness above 0.8. The association between early childhood greenness and mid-childhood visual memory was also observed in the fully adjusted linear regression analysis (Web Table 2), with a minimal set of confounders (Web Table 3), and further adjusting for early childhood cognition (Web Figure 4), while the association with mid-childhood nonverbal IQ was not. We observed a U-shaped association between greenness at early childhood and mid-childhood visual-motor abilities (Figure 1K). However, this association was not observed in the linear regression analysis (Web Table 2). Further, we did not find significant associations of greenness at any time point with mid-childhood verbal IQ, and there was no clear evidence of a beneficial association of NDVI at birth or mid-childhood with any of the mid-childhood cognitive measures.

Figure 1.

Nonlinear relationship between greenness and cognitive assessments at mid-childhood, Project Viva, Massachusetts, 1999–2010. A–C) Verbal intelligence quotient; D–F) nonverbal intelligence quotient; G–I) visual memory; J–L) visual-motor abilities. First column, greenness at birth; second column, greenness in early childhood; third column, greenness in mid-childhood. Adjusted for sex, age, maternal characteristics (intelligence quotient, education, race), paternal education, household annual income at enrollment in early pregnancy, neighborhood median annual income at birth, and greenness exposure in all preceding age periods but not in subsequent age periods. P values correspond to spline terms for greenness exposure at each time point: A) P = 0.09; B) P = 0.82; C) P = 0.27; D) P = 0.61; E) P = 0.04; F) P = 0.80; G) P = 0.74; H) P < 0.01; I) P = 0.36; J) P = 0.23; K) P = 0.01; L) P = 0.48.

Sensitivity analyses showed that the relationships between greenness and mid-childhood cognition excluding outliers (Web Figure 5) and at a 270-m buffer (Web Figure 6) were similar but attenuated. A comparison between the final (n = 857) and the full sample (n = 2,128) showed that participants in the final sample were more likely to be White, have mothers with higher than college education, and live in a household with an income greater than $70,000/year. Greenness exposure at all time points and cognition measures at mid-childhood were similar between the full and final sample (Web Table 4).

Stratified analysis

In general, the highest cognitive scores were observed among girls, children of parents with high education, and those with high NSES (Web Table 5). We did not observe effect measure modification by sex, consistent with previous findings (10). The relationship between greenness at birth and mid-childhood verbal intelligence differed by parental education, NSES, and neighborhood population density (Figure 2). The analysis by parental education suggested that the positive association between greenness at birth and mid-childhood verbal intelligence was mostly driven by parental high education. Stratified analysis by NSES suggested that at greenness levels of >0.5, there was a positive association with mid-childhood verbal IQ for low NSES. For high NSES, the positive association with mid-childhood verbal IQ seemed to decrease after 0.6 NDVI. These results are in contrast to a recent study reporting no significant effect modification by maternal education or NSES (9). The analysis by population density suggested a positive association between greenness at birth and verbal IQ at mid-childhood for areas with high population density. This was similar to a study that reported stronger associations between greenness and psychiatric disorder in the highest urbanized areas (14). There was no effect measure modification by any strata for early or mid-childhood greenness.

Figure 2.

Stratified analysis of verbal intelligence quotient (IQ), Project Viva, Massachusetts, 1999–2010. A) Parental education; B) neighborhood socioeconomic status (NSES); C) population density. Bold line indicates estimated spline, and dotted lines indicate 95% confidence intervals.

Mediation

The estimate of the indirect effect of the association between early childhood greenness and mid-childhood nonverbal IQ mediated by BC (assuming that underlying assumptions of the mediation analyses hold) was statistically significant (P < 0.001, Table 2). The indirect effect estimate was negative since the estimated association between early childhood greenness and early childhood BC was −0.13 (95% CI: –0.15, –0.12), while the estimated association between early childhood BC and mid-childhood nonverbal IQ was 14.72 (95% CI: 5.59, 23.83), and thus the estimated indirect effect of early childhood greenness on mid-childhood cognition via early childhood BC was −1.94 (i.e., −0.132 × 14.72, 95% CI: –3.13, −0.78). The total effect was 1.66 (95% CI: −0.59, 3.79), resulting in a negative ratio of the natural indirect effect to the total effect (“proportion” mediated, i.e., −1.94/1.66); however, the confidence interval for the ratio included the null (−1.17, 95% CI: −9.54, 6.67) (Table 2). No other estimates of mediation by BC or physical activity were significant. Further, interaction terms between exposure and mediator were not significant in any of the models.

Table 2.

Mediation Analysis^a Estimating the Proportion of the Association Between Greenness at Early Childhood and Mid-Childhood Cognition Explained by Air Pollution Exposure, and Physical Activity, Both Measured at Early Childhood (n = 733^b), Project Viva, Massachusetts, 1999–2010

Mediators at Early Childhood	Verbal IQ		Nonverbal IQ		Visual-Motor Abilities		Visual Memory
	Proportion	95% CI	Proportion	95% CI	Proportion	95% CI	Proportion	95% CI
Air pollution
Indirect effect	−0.70	−1.58, 0.18	−1.94	−3.13, −0.78	0.2	−1.05, 1.61	−0.07	−0.42, 0.24
Direct effect	1.07	−0.92, 3.07	3.61	1.06, 6.02	−0.17	−3.28, 2.33	0.84	0.17, 1.46
Total effect	0.37	−1.28, 2.10	1.66	−0.59, 3.80	0.03	−2.43, 2.18	0.76	0.21, 1.32
Proportion of association explained by mediator	−1.87	−12.7, 11.91	−1.17	−9.54, 6.67	6.06	−7.52, 6.72	−0.10	−0.8, 0.47
Physical activity
Indirect effect	−0.03	−0.17, 0.06	−0.04	−0.2, 0.08	0.01	−0.08, 0.15	0	−0.03, 0.04
Direct effect	0.40	−1.27, 2.11	1.71	−0.62, 3.83	0.02	−2.18, 2.32	0.76	0.22, 1.30
Total effect	0.37	−1.28, 2.10	1.66	−0.51, 3.90	0.03	−2.15, 2.33	0.76	0.22, 1.30
Proportion of association explained by mediator	−0.09	−0.62, 0.81	−0.03	−0.44, 0.20	0.25	−0.68, 0.67	0	−0.05, 0.06

Abbreviations: CI, confidence interval; IQ, intelligence quotient.

^a Mediation analyses assume that there is no unmeasured exposure-outcome confounding, no unmeasured mediator-outcome confounding, no unmeasured exposure-mediator confounding, and no mediator-outcome confounder affected by exposure.

^b Analytical sample for mediation analysis excluded participants with missing data on any of the mediators of interest.

DISCUSSION

We observed nonlinear relationships of exposure to greenness at early childhood with cognition at mid-childhood among Project Viva participants, with findings varying across NDVI values. Our results suggested that greenness exposure at early childhood was associated with better visual memory in mid-childhood. The observed association with visual memory was consistent after adjustment for child age, sex, and race and with mother’s IQ, mother’s education, father’s education, household annual income, and NSES. The results with visual memory were also observed after further adjusting for early childhood cognitive assessment, strengthening the evidence of an association. Sensitivity analyses excluding outliers had little influence on the relationship between exposure to greenness across childhood and mid-childhood cognition.

There was some evidence that the association between greenness exposure at birth and mid-childhood verbal intelligence differed by parental education, NSES, and neighborhood population density. We observed that greenness did not benefit participants with low parental education. Participants with low parental education had lower cognitive scores at baseline and were not benefited by greenness exposure (perhaps due to limited access), while participants with higher parental education started off with higher cognitive scores and seemed to benefit from greenness. It is possible that parents with lower levels of education had occupations that kept them and their children from enjoying outdoor space. On the other hand, for participants with low NSES, we observed cognitive benefits of greenness at levels above 0.5, reaching verbal IQ scores comparable to participants with high NSES. This finding is in line with the theory of equigenic environments, which suggests that greenness disrupts the usual conversion of socioeconomic adversity to greater risk of poor health and might be important to reduce socioeconomic health inequalities (40). The observed positive association between greenness at birth and mid-childhood verbal intelligence in areas with high population density, that may be more urbanized, suggests that policies to increase urban greenness may have sustainable public health benefits. In areas of lower population density, we observed a plateau/drop-off of benefit at higher NDVI values which may suggest that more greenness in suburban areas may have different associations with cognition.

Previous experimental studies have found that, compared with urban experience, nature experience leads to cognitive benefits such as improved attention among school-aged children (41–47). However, prior cohort studies assessing greenness surrounding schools and residences have reported inconsistent associations with brain development in children (1, 9, 12, 13), and few studies have examined sensitive periods of greenness exposure (7, 14, 15). The life-course associations found in our study suggest that early childhood is potentially a sensitive period for greenness exposure and visual memory development. Early childhood (approximately 3 years of age) has been considered a major window of developmental vulnerability (48), particularly for brain maturation, during which environmental exposures may be particularly relevant (4, 49). Research has shown that high levels of green space during childhood are associated with lower risk of psychiatric disorders later in life (14). Our findings were not consistent with those from a birth cohort study that found that higher residential greenness at birth was associated with better neurodevelopmental assessments at 24 months (10) and a longitudinal study that found that lifelong exposure to residential surrounding greenness since birth was positively associated with working memory and reduced inattentiveness in 7-year-old children (50). However, our finding on early childhood as a sensitive period for visual memory is in line with a meta-analysis that found an association between exposure to toxic substances throughout early childhood and childhood cognition (51). Contrary to our hypotheses, our results suggested that higher greenness exposure at early childhood might also be related to lower visual-motor abilities in mid-childhood. A recent analysis of the same population also found unexpected associations between polyfluoroalkyl substances and visual-motor abilities, where visual-motor abilities were higher among children born to mothers in the highest quartile of perfluorooctanoate plasma concentrations (vs. lowest) (52). Another study, assessing the association between greenness and visual-motor delays in children living in Berlin, reported no significant findings (13).

Assuming that the assumptions of the mediation analysis hold, our estimates suggest that air pollution did not explain the association between green space exposure in early life and mid-childhood cognition. This is inconsistent with a study by Dadvand et al. (9) and another by Liao et al. (10), which found that indoor levels of elemental carbon explained 20% to 65% and traffic-related air pollution explained 13.6% to 28.0%, respectively, of the association between green space exposure and early childhood neurodevelopment. The difference might be explained by the use of BC as traffic-related air pollution as opposed to indoor levels of elemental carbon or fine particular matter (having an aerodynamic diameter ≤2.5 μm). Moreover, the positive association found between early childhood BC and mid-childhood nonverbal IQ was unexpected but also found in a previous study of Project Viva. The authors found that BC exposure in the first 6 years of life was associated with higher nonverbal IQ in mid-childhood (1.7 points per interquartile-range increase, 95% CI: 0.1, 3.4) (53). The difference in our estimate on the association between BC and nonverbal IQ compared with Harris et al. (53) might be due to using BC as a continuous variable (vs. interquartile range) or due to different covariates in the model. This unexpected association could be due to participants living in more urbanized areas with higher access to education but also greater traffic-related air pollution exposure. Further, our estimates suggested that the observed association between greenness and mid-childhood cognitive function was not explained through physical activity, similar to 2 previous studies (54, 55). However, another study reported evidence that physical activity mediates the association between residential surrounding green space and mental health among young adults (56). These differences may be due to distinct outcome and mediator measurements, participants’ ages (children vs. older adults), and covariate adjustment.

This study had a few limitations. First, residential greenness may reflect socioeconomic disparities. Although we adjusted for parental education and NSES, we cannot rule out residual confounding by unmeasured socioeconomic factors. Residual confounding of the mediator-outcome association may also explain the unexpected association between early childhood BC and mid-childhood cognition. We used sensitivity analysis further adjusting for potential mediator-outcome confounders; however, our results remained similar (results not shown). Second, use of NDVI did not allow us to identify specific types of vegetation (e.g., grass vs. trees) that may drive the association with cognition. Third, our study population had on average high socioeconomic status, so generalizability to less-advantaged populations may be limited. However, based on 2010 Census data (57), regional Boston characteristics such as median household income, percentage of households with income over $70,000, and percentage of White individuals were comparable to those of Project Viva, suggesting that the Project Viva cohort is generalizable to the Boston population. Further, Project Viva participants did very well on the outcome measures, with mean scores well above the standardization population, which is consistent with their generally being a sociodemographically/economically advantaged cohort. However, this also suggests that their lower (relative to the standardization population) performance on visual-motor testing is notable and could play a role in the paradoxical findings for that outcome. Fourth, missing data may result in selection bias. Although it is reassuring that early-life exposure to greenness and mid-childhood cognition measures of included participants were similar to those of the excluded participants, there were small differences in some factors correlated with both exposures and outcomes (mother’s education and household income), suggesting potential selection bias (Web Table 4). Selection bias could also be possible due to parental socioeconomic status and their ability to decide where to live, which would also be related to child cognition. Fifth, we tested multiple time periods for several outcomes and evaluated effect modifiers and mediators, which could potentially lead to multiple testing problems and finding statistically significant results purely by chance. Finally, statistical power may be limited for mediation or stratified analysis.

This study has several notable strengths. The use of the longitudinal data from the Project Viva cohort allowed us to examine prospective associations between residential exposure to greenness and validated cognitive measures. To our knowledge, this study is the first to examine multiple windows of greenness exposure (birth, early childhood, and mid-childhood) in relation to cognitive development (early childhood and mid-childhood). The study was conducted in a relatively large, prospective cohort with rich data on maternal and child health, which enabled adjustment for important potential confounders, including maternal IQ and sociodemographic factors. The temporal ordering of the exposure, mediator, and outcome measurements facilitated interpretation of the mediation results.

Among children in Project Viva, higher exposure to residential greenness at early childhood was associated with higher visual memory scores in mid-childhood. In a neurotypical population, subtle alterations in continuous measures of cognition have important implications for clinical impact at the population level (58). These findings suggest that exposure to greenness in early childhood might have a greater influence on cognitive development than later in childhood.