Prenatal exposure to residential greenness, fetal growth, and birth outcomes: a cohort study in New York City

Abstract

Findings for greenspace’s impacts on birth outcomes are largely dependent on vegetation indexes. Examinations are needed for various greenspace indicators given varying pathways for fetal development. This prospective cohort study assessed the impacts of prenatal greenspace exposure on preterm birth (PTB), term low birthweight (TLBW), birthweight, and estimated fetal weight (EFW) for pregnant women in the New York City area, 2016–2023 (n=2765). Longitudinal greenspace exposure was measured for residential histories during pregnancy using the Enhanced Vegetation Index (EVI) for 1000m buffers and four park metrics, namely, the total number, sum of area, and the accessibility of parks within residential buffers (500 m) and the distance to the closest park. Multivariable regression models were used to estimate the associations for quartiles of exposure (with the first quartile [Q1] as reference). Greenspace exposure was not associated with TLBW, birthweight, or EFW. Odds ratios of PTB for the Q2, Q3, and Q4 EVI exposure groups compared to the Q1 group were 0.65 (95% CI: 0.43–0.98), 0.51 (0.32–0.80), and 0.56 (0.35–0.90), respectively. PTB risks decreased in higher exposure groups (Q2–Q4) of the total park number. Results indicate the benefits of prenatal greenspace exposure for fetal maturity and neonatal outcomes.

Introduction

Greenspace, referring to structured (e.g., park, recreational open space) and unstructured (e.g., undesignated open space, street canopy) vegetated areas (1), can contribute to overall health and well-being (2). Normalized Difference Vegetation Index, the most applied greenspace indicator in pregnancy studies, is calculated from visible and near-infrared light reflected by vegetation and captured by satellite sensors. Vegetation indices can reflect temporal changes in total greenness regardless of the greenspace type (e.g., garden, park). The total greenspace area indicates the scale of greenspace for a given area. The number of parks within an area, interpreted as the fragmentation of greenspace (e.g., parcels), is also a widely examined indicator in the literature (3). Other greenspace indicators are based on proximity, such as distance between home and parks. In epidemiological studies, use of various greenspace indicators can reflect different potential causal mechanisms. Vegetation may enhance health by reducing environmental stressors (e.g., air pollution, ambient heat) and providing psychological effects (e.g., stress relief) (1). Scale and proximity of neighborhood greenspace is a proxy of the likelihood of greenspace visits providing health benefits from enhanced environment, physical activities, psychological effect, and social cohesion (1,4–8). In urban planning, it is important to identify how the allocation and distribution of parks can be practically and suitably modified to improve human health (3). Given rapid global urbanization reducing greenspace, examining health effects of greenspace using multiple indicators is needed to inform health policies and urban planning.

Higher exposure to residential greenness is associated with decreased risks of adverse birth outcomes such as low birthweight and preterm birth (PTB) (7,9–13). However, gestational exposure to greenness has primarily been measured based on residential address at delivery, leading to potential exposure misclassification. Also, evidence is limited for the associations with birth outcomes for greenspace indicators other than vegetation indices (7,14). Further, few studies have examined fetal growth measures. Birthweight may not accurately reflect dynamic fetal growth patterns in utero (12). Ultrasonography-based estimated fetal weight (EFW) is a promising and popular parameter used to screen for fetal growth restriction (15). Estimation of fetal weight is important in routine obstetric care to reduce potential preterm deliveries, newborn complications during labor, puerperium of microsomia, and resultant maternal and perinatal mortality or morbidity (16,17). Study results are inconclusive for the impact of greenspace on other adverse birth outcomes including PTB, potentially because of different population characteristics, exposure measurements for greenspace, and study designs. Therefore, research using various greenspace indicators for fetal growth and PTB is needed.

This prospective cohort study investigated associations between prenatal greenspace exposure with fetal growth and adverse birth outcomes. We examined various greenspace indicators considering the quantity, size, and proximity of greenspace and leveraged routinely performed ultrasound to obtain fetal weight estimates and residential histories during pregnancy. We hypothesized that higher greenspace exposure during pregnancy is associated with lower PTB risks and increased fetal growth.

Methods

Study population

The Institutional Review Board of the NYU Grossman School of Medicine approved the study. Written informed consent forms were obtained from all participants. We used data from the New York University (NYU) Children’s Health and Environment Study (CHES), a prospective cohort of pregnant women in the New York metropolitan area, enrolled in early pregnancy 2016–2023 (n=4439)(18). Eligible participants were age ≥18 years, <18 weeks pregnant, had a pregnancy that was not medically threatened, and planned to deliver at one of the three NYU-affiliated hospitals in New York City (NYC). The cohort population has substantial similarity to pregnant individuals with livebirths in NYC, except that the CHES participants were more likely to be Hispanic and married (18). Participants answered questionnaires during three routine prenatal visits (<18, 18–25, and >25 gestational weeks).

Our inclusion criteria were (1) being singleton livebirths, (2) having information on >1 residential addresses during gestational periods in questionnaires and/or electronic health record data, and (3) having exposure estimates for greenspace. From 3894 singleton livebirths, we excluded 444 infants whose mothers did not provide address information during pregnancy. For the remaining 3450 participants, 685 participants were excluded due to the lack of available satellite sensor vegetation data in 2022–2023, resulting in 2765 participants (Figure S1). The excluded participants were slightly older (less than a year) and had higher educational levels and income. They were more likely to have private insurance, be employed, and smoke before pregnancy (Table S1).

We extracted information on birthweight and gestational age at birth (estimated based on last mensural period if available and ultrasound dating otherwise) from electronic health records, and if missing (4% of samples), data were replaced with information from maternal reports. This brought the total missing in birthweight to 0.2% and gestational age to 0. Fetal biometry information was extracted from routinely performed ultrasound during prenatal visits in 3663 singleton livebirths (16–38 gestational weeks, total of 13075 scans) and were used to calculate EFW (see below) (19). After merging fetal growth data with exposure, 2765 mother-fetus pairs were included in the analysis.

Exposure assessment

We used five greenspace indicators: Enhanced Vegetation Index (EVI; referred to as vegetation hereafter), number of neighborhood parks, area of neighborhood parks, Accessibility Index of neighborhood parks in residential buffers, and distance to nearest park. The vegetation index was calculated for 100, 250, 500, 1000, and 2000 m circular buffers of each residence and the other four indicators were calculated for 500m circular buffers of residences.

EVI is an improved version of the Normalized Difference Vegetation Index (NDVI) with correction for measurement bias from airborne aerosols, ground below vegetation, and extreme chlorophyll density (20). EVI was calculated using the Landsat-8 Collection 1 Tier 1 database 8-Day EVI Composite data (21), as described in Figure S2. Average 8-day EVI pixel values in circular buffers were calculated for each residence where pregnant women lived during pregnancy. The combined EVI estimates per 8 days throughout pregnancy represented the time-variant exposures to EVI for each participant. We calculated the average of the 8-day EVI estimates for our exposure windows (e.g., pregnancy, trimester, 0–36 gestational weeks), using these combined 8-day EVI estimates.

We used the OpenStreetMap datasets to obtain polygons of points of interest (parks, playgrounds, and dog parks). All park polygons overlapping with one or more points of the 500m buffer of residence (e.g., intersecting with some buffer areas or touching one point of the buffer boundary) were included in calculating the number of parks, sum of park areas, and Accessibility Index of parks. EVI, number of parks, and sum of park areas are quantity-based indicators, whereas distance to the nearest park is proximity-based (1). The Accessibility Index combines the scale and proximity of neighboring parks (22). Although it does not represent other greenspace aspects such as fragmentation or quantity, it has been used to determine attractiveness of parks and examine socioeconomic disparities in park access in previous studies (23). The Accessibility Index (A_ij) is calculated as:

$$A_{ji}={\sum }_j{S}_j^{\alpha }/d_{ij}^{\beta }$$

where S_j is the size of destination park j, d_ij is distance from residence i to park j, α is a parameter (0.85) reflecting the size effects of park j on its accessibility, and β is the distance decay or friction parameter (1.91) (24). Park metrics were calculated based on residential histories and averages of each metric during exposure windows (e.g., pregnancy) were calculated for each participant, as shown in Figure S3.

Outcomes

The outcomes were preterm birth (PTB) (gestational week <37), term low birthweight (TLBW) (birthweight <2500 grams at term delivery), Z-score of birthweight, and EFW. Z-score birthweight was estimated based on the external standard population, using the Intergrowth-21st database (25). For fetal growth, EFW was calculated using Hadlock’s three-parameter formula, incorporating abdominal circumference, femur length, and head circumference (26). To minimize variability in timing of ultrasound measurements between participants, we predicted EFW at 20, 30, and 36 weeks of gestation for all singleton live births. To lessen bias from a large number of ultrasounds, we excluded scans beyond the sixth visit, the 90th percentile of anatomy ultrasound visits. EFW predictions were performed using linear mixed-effects modeling with the B-spline basis matrix serving as predictor variables.

Covariates

Electronic health records obtained participants’ age at enrollment, parity, insurance type, weight, height (to calculate body mass index [BMI]), delivery season, and infant sex. Women self-reported race/ethnicity, education, marital status, income, employment, and alcohol/tobacco use via repeated surveys during pregnancy. We used Patient Health Questionnaire (PHQ-9) (27) to measure depressive symptoms in pregnancy (probable major depression: total PHQ-9 score >10). Level of physical activity during pregnancy was grouped into tertiles using the average score of the International Physical Activity Questionnaire-Short Form [IPAQ] (28) across all three trimesters for each participant.

Average daily ambient temperature was estimated for 500m buffers using 4×4km gridded modeling data (29). Average fine particulate matter (PM_2.5) exposure was estimated for 500m residential buffers using monthly PM_2.5 modeling data (0.1° resolution) (30).

We considered socioeconomic factors at the ZIP code Tabulation Areas (ZCTAs) of residential addresses at delivery: percent of the population that is Black, percent of population in poverty, age-standardized rate of population with high school education or less, and median household income (USD).

Statistical analysis

We imputed missing values for covariates based on the expectation-maximization with bootstrapping using R package Amelia II (31) as applied in previous birth cohort studies (32). Combining inference estimates from multiple imputed datasets was based on Rubin’s rules (33). The percentage of missing values was <12% with the largest missing rate for annual household income (11.1%).

To estimate associations of prenatal exposure to greenness with fetal growth (potential pathways shown in Figure S4), we used multivariable linear regression models for EFW. Greenness exposure corresponding to each exposure window was used (e.g., vegetation index during 1–20 weeks for EFW at 20th gestational week). Due to the non-linear associations of EFW with vegetation index or park metrics (Figure S5) and the skewed distributions of park metrics, we categorized 1000m vegetation index and park metrics as quartiles. Differences in EFW were compared by quartiles (with first quartile [Q1] as reference) of 1000m vegetation index levels. For Z-score birthweight, quartiles of 1000m vegetation index and park metrics during pregnancy were used in multivariable linear regression models.

For PTB and TLBW, multivariable logistic regression models were applied using greenness exposure levels during pregnancy and each trimester. Associations were calculated as odds ratios (ORs) and 95% confidence intervals (CIs) for each exposure quartile group, with the Q1 as reference.

Final models included maternal age at enrollment (<29, 30–34, 35–39, ≥40 years), race/ethnicity (Asian, Hispanic, non-Hispanic [NH] Black, NH White, multi-race, other), education level (≤high school, college education [received or graduated], and post-graduate degree), BMI, marital status (married/living with a partner, divorced/separated/single/widowed), employment status (employed, unemployed), alcohol and cigarette use during pregnancy (yes, no), parity (nulliparous, multiparous), insurance type (public, private), household income (<$30,000, $35,000–74,000, $50,000–74,000, $75,000–99,000, ≥$100,000/year), season of delivery, infant sex, average ambient temperature, and average PM_2.5 concentrations. In analyses with TLBW, gestational age (week) was also included as a continuous variable (34,35). We conducted additional analysis including quartile-based variables of ZCTA-level percent of the population that is Black, percent of population in poverty, age-standardized education level, and median household income.

We examined effect modifications by socioeconomic factors including maternal race/ethnicity, maternal education level, and ZCTA-level socioeconomic factors, using an interaction term. Associations in the higher exposure group (exposure ≥median) compared to the lower exposure group (exposure <median) were calculated.

In a sensitivity analysis, we used residence at delivery instead of integrating residential address history across pregnancy and included participants who only had addresses at delivery (n=7) and participants with residential history (n=2765). When residential histories during pregnancy were unavailable, some earlier studies assumed that pregnant women did not move and assigned exposure levels during pregnancy using an address at a certain time (e.g., delivery or recruitment) (9,36,37). This alternative model examined the robustness of risk estimates when using only the address at delivery and aids comparability to other research.

We re-ran main models (n=2765 participants) additionally adjusted for depression and physical activity in pregnancy, which might be associated with exposure and outcomes (36,38,39). These sensitivity analyses included a dichotomous variable of probable major depression measured during the first trimester, using the PHQ-9 questionnaire. The physical activities were adjusted using the tertile group variables of the average IPAQ score of participants.

We analyzed associations between outcomes and vegetation index by different buffer sizes (100, 250, 500, and 2000m). The 500m radius for residential buffers and parks was chosen based on prior finding that the use of the park decreased after the 400m (14). We also applied park metrics at 1000m buffers instead of the 500m buffers used in our main analyses.

In another sensitivity analysis, we used network distance through the OpenStreetMap road connectivity between residences and parks, instead of Euclidean distance. For a destination park, there can be multiple travel pathways from the residence, and we used the shortest network distance. The park-related metrics (e.g., number) were estimated for parks reachable within 500-m network distances.

Sensitivity analyses for varying residential buffer sizes were designed a priori. Other sensitivity analyses were conducted post hoc.

Results

Analysis included 2,765 mother-child pairs. Table 1 shows the characteristics of the pregnant women with singleton livebirths. Participants were largely Hispanic (49.2%), followed by NH White (33.0%). The percentage of non-Hispanic Black, Asian, and multi-racial groups was 6.0%, 9.2%, and 1.7%, respectively. About 38.6% of participants had annual income ≥$100,000, and 86.3% were married or lived with a partner.

Table 1.

Descriptive statistics of individual-level and community-level (ZCTA) variables (2016– 2023) of study participants (n = 2,765).

	N (%)
Individual-level factors
Maternal characteristics
Age at enrolment (years), mean ± SD	32.0 ± 5.6
Race and ethnicity
Hispanic	1333 (49.0)
Non-Hispanic White	891 (32.8)
Non-Hispanic Black	164 (6.0)
Asian	249 (9.2)
Other	36 (1.3)
Multi-race	46 (1.7)
Educational levels
High school or less	828 (33.1)
Some college but no degree	281 (11.2)
Bachelor’s or associate degree	691 (27.8)
Post-graduate degree	698 (27.9)
Relationship status
Married/living with a partner	2310 (86.1)
Single/widowed/divorced/separated	372 (13.9)
Annual household income
<$30,000	415 (16.9)
$30,000–$49,999	210 (8.5)
$50,000–$74,999	159 (6.5)
$75,000–$99,999	116 (4.7)
≥$100,000	947 (38.5)
Do not know/prefer not to say	611 (24.6)
Insurance type
Public	1393 (50.8)
Private	1348 (49.2)
Employed during pregnancy
Yes	1669 (63.1)
No	1096 (36.9)
BMI, mean ± SD	26.5 ± 5.8
Alcohol use during pregnancy
Yes	384 (15.2)
No	2149 (84.8)
Smoking before pregnancy
Yes	90 (3.5)
No	2450 (96.5)
Parity (nulliparous)	1284 (46.6)
Infant characteristics
Birthweight (g), mean ± SD	3256.0 ± 543.9
Estimated fetal weight at the 20th gestational age (g), mean ± SD	335.3 ± 50.0
Estimated fetal weight at the 30th gestational age (g), mean ± SD	1574.0 ± 140.4
Estimated fetal weight at the 36th gestational age (g), mean ± SD	2826.0 ± 258.2
Preterm birth, n (%)	235 (8.5)
Term low birthweight, n (%)	61 (2.5)
Exposure levels of PM2.5 during pregnancy (µg/m3)	7.9 ± 0.6
Exposure levels of ambient temperature during pregnancy (ºC)	13.3 ± 2.7
ZCTA-level factors
Percent of population in poverty, mean SD	14.1 ± 8.2
Percent of Black population, mean ± SD	14.5 ± 19.6
Age-standardized percent of population with high school education or less	41.6 ± 17.5
Annual mean household income ($), mean ± SD	108826 ± 51039.0
Annual median household income ($), mean ± SD	92808 ± 65634.4

Abbreviations: BMI: body mass index, PM_2.5, particulate matter with aerodynamic diameter ≤2.5µm.

Rate of missing information: maternal race–ethnicity (1.7%), maternal education level (9.7%), relationship status (3.0%), annual household income (11.1%), insurance type (0.9%), employment status during pregnancy (4.4%), BMI (2.1%), alcohol use during pregnancy (8.4%), smoking before pregnancy (8.1%), parity (0.4%), estimated fetal weight (13.1% for the 20th gestational week, 13.6% for the 30th gestational age, and 17.8% for the 36th gestational age), and ZCTA-level factors (0.3%).

Average (SD) of EFW at 20, 30, and 36th gestational weeks was 335.3 (50.0), 1574.0 (140.4), and 2826.0 g (258.2 g), respectively. Birthweight averaged 3256 g (543.9). Among the included infants, 231 (8.5%) were PTB. Among term births (n=2415), 61 (2.5%) were TLBW.

Figure 1 shows regional differences in vegetation levels (i.e., EVI), participant locations, and parks in the study areas. Average (SD) and range of prenatal exposure to 1000m vegetation index were 0.155 (0.067) and 0.030–0.517, respectively (Table 2). The maximum total number of parks (500m buffers) was 38. The maximum sum of park areas was 7.333km² (average=0.217km²). The Access Index showed a large difference between average and median values. Average distance to the nearest park was 0.281km.

Figure 1.

Average EVI (A) and locations of parks (B) in study regions. Purple circles indicate participants’ residential locations. Red stars indicate recruiting hospitals.

Table 2.

Exposure levels for residential greenspace during pregnancy.

Variable	Mean ± SD	Min	Q1	Median	Q3	Max
EVI (1000m buffer)	0.155 ± 0.067	0.030	0.115	0.144	0.175	0.517
Total number of parks (500m buffer)	4.7 ± 4.5	0.0	1.6	3.0	6.2	38.0
Sum of park areas (km2) (500m buffer)	0.217 ± 0.757	0.000	0.007	0.025	0.094	7.333
Access Index of park (km) (500m buffer)	4.375 ± 64.457	0.000	0.0002	0.001	0.033	2870.444
Distance to nearest park (km)	0.281 ± 0.310	0.000	0.128	0.222	0.356	7.262

Abbreviations: EVI, enhanced vegetation index; SD, standard deviation.

We found lower odds of PTB associated with higher vegetation (i.e., EVI) exposure quartile groups than the reference group (i.e., Q1)(Table 3). ORs of PTB in the model adjusted for individual-level covariates and community-level socioeconomic status were 0.65 (95% CI: 0.43– 0.98), 0.51 (0.32–0.80), and 0.56 (0.35–0.90) in the Q2, Q3, and Q4 exposure groups of 1000m vegetation (Table 3). Significantly lower PTB risks were also found in higher exposure groups compared to Q1 for the total number of parks within 500m of residence: OR=0.57 (95% CI: 0.39–0.85), 0.59 (95% CI: 0.39–0.87), and 0.59 (95% CI: 0.38–0.90) for the Q2, Q3, and Q4 groups, respectively (Table 3). Associations for TLBW were not observed in term-born infants except for Q4 for number of parks, Q2 for sum of park areas, and Q3–Q4 for distance to parks (Table 3). We did not observe associations of greenspace with fetal measures or birthweight, except for distance to the nearest park, EFW (20-week gestation), and birth weight (Table 4).

Table 3.

Odds ratios (ORs) of preterm birth and term low birthweight for quartiles of exposure levels of residential greenspace.

Exposure [min, max]	PTB		TLBW
	Adjusted OR (95% CI) from base modela	Adjusted OR (95% CI) from base model+further adjusted for ZCTA-level SESb	Adjusted OR (95% CI) from base modela	Adjusted OR (95% CI) from base model+further adjusted for ZCTA-level SESb
EVI (1000m buffer)
Q1 [0.030, 0.114]	1.00	1.00	1.00	1.00
Q2 [0.115, 0.143]	0.69 (0.47, 1.02)	0.65 (0.43, 0.98)c	1.48 (0.68, 3.25)	1.84 (0.78, 4.34)
Q3 [0.144, 0.174]	0.58 (0.38, 0.89)c	0.51 (0.32, 0.80)c	1.09 (0.44, 2.65)	1.52 (0.57, 4.03)
Q4 [0.175, 0.517]	0.71 (0.46, 1.07)	0.56 (0.35, 0.90)c	1.79 (0.75, 4.27)	2.88 (1.06, 7.81)
Number of parks (500m buffer)
Q1 [0, 1.5]	1.00	1.00	1.00	1.00
Q2 [1.6, 2.9]	0.55 (0.38, 0.81)c	0.57 (0.39, 0.85)c	0.61 (0.29, 1.32)	0.53 (0.24, 1.18)
Q3 [3.0, 6.1]	0.59 (0.40, 0.86)c	0.59 (0.39, 0.87)c	0.65 (0.30, 1.43)	0.48 (0.20, 1.11)
Q4 [6.2, 38.0]	0.60 (0.41, 0.88)c	0.59 (0.38, 0.90)c	0.52 (0.23, 1.18)	0.29 (0.11, 0.74)c
Sum of park areas (km2) (500m buffer)
Q1 [0, 0.006]	1.00	1.00	1.00	1.00
Q2 [0.007, 0.024]	0.90 (0.61, 1.32)	0.91 (0.62, 1.34)	0.39 (0.16, 0.95)	0.35 (0.14, 0.88)c
Q3 [0.025, 0.093]	0.93 (0.63, 1.36)	0.95 (0.64, 1.41)	0.60 (0.26, 1.35)	0.43 (0.18, 1.03)
Q4 [0.094, 7.333]	0.76 (0.51, 1.13)	0.77 (0.52, 1.16)	1.15 (0.56, 2.35)	1.00 (0.47, 2.14)
Access Index of park (km) (500m buffer)
Q1 [0, 1×10−4]	1.00	1.00	1.00	1.00
Q2 [2×10−4, 9×10−4]	0.89 (0.61, 1.31)	0.91 (0.61, 1.34)	0.85 (0.37, 1.93)	0.87 (0.37, 2.06)
Q3 [0.001, 0.032]	1.05 (0.72, 1.53)	1.05 (0.71, 1.55)	0.72 (0.30, 1.72)	0.53 (0.21, 1.34)
Q4 [0.033, 2.870×103]	0.70 (0.47, 1.04)	0.71 (0.47, 1.08)	1.55 (0.72, 3.33)	1.47 (0.66, 3.27)
Distance to nearest park (km)
Q1 [0, 0.127]	1.00	1.00	1.00	1.00
Q2 [0.128, 0.221]	0.75 (0.51, 1.10)	0.74 (0.50, 1.08)	0.76 (0.38, 1.49)	0.74 (0.37, 1.50)
Q3 [0.222, 0.355]	0.74 (0.50, 1.09)	0.72 (0.49, 1.08)	0.38 (0.17, 0.87)c	0.41 (0.17, 0.96)c
Q4 [0.356, 7.262]	0.91 (0.62, 1.32)	0.86 (0.58, 1.28)	0.28 (0.11, 0.68)c	0.30 (0.12, 0.76)c

Abbreviations: PTB, preterm birth; TLBW, term low birthweight; EVI, enhanced vegetation index.

^aBase model adjusted for participants’ individual-level covariates, PM_2.5, and temperature.

^bFurther adjusted for ZCTA’s socioeconomic variables and participants’ individual-level covariates and exposures to PM_2.5 and temperature in the base model.

^cSignificant results at significance level of 0.05.

Table 4.

Differences (95% CI) in Z-score estimated fetal weights and birthweight in quartile exposure levels for residential greenspace.

Exposure group	Beta (95% CI)a
	Z-score EFW at 20-week gestation	Z-score EFW at 30-week gestation	Z-score EFW at 36-week gestation	Z-score Birthweight
EVI (1000m buffer)
Q1	Reference	Reference	Reference	Reference
Q2	−0.044 (−0.159, 0.072)	−0.089 (−0.209, 0.031)	0.002 (−0.119, 0.123)	0.064 (−0.049, 0.178)
Q3	−0.160 (−0.280, −0.040)b	−0.066 (−0.201, 0.069)	−0.014 (−0.147, 0.119)	0.145 (0.023, 0.267)b
Q4	-0.067 (−0.194, 0.059)	−0.045 (−0.190, 0.099)	−0.019 (−0.162, 0.124)	0.088 (−0.043, 0.219)
Number of parks (500m buffer)
Q1	Reference	Reference	Reference	Reference
Q2	0.004 (−0.112, 0.120)	0.006 (−0.109, 0.120)	0.082 (−0.036, 0.199)	0.137 (0.027, 0.246)b
Q3	0.014 (−0.103, 0.132)	−0.007 (−0.123, 0.109)	0.017 (−0.103, 0.136)	0.078 (−0.036, 0.190)
Q4	−0.025 (−0.153, 0.102)	−0.009 (−0.136, 0.118)	0.012 (−0.119, 0.142)	0.072 (−0.052, 0.196)
Sum of park areas (km2) (500m buffer)
Q1	Reference	Reference	Reference	Reference
Q2	−0.034 (−0.149, 0.082)	0.059 (−0.054, 0.171)	0.014 (−0.102, 0.130)	0.003 (−0.107, 0.113)
Q3	−0.004 (−0.122, 0.113)	−0.111 (−0.228, 0.005)	−0.053 (−0.172, 0.066)	−0.024 (−0.137, 0.089)
Q4	−0.010 (−0.125, 0.105)	−0.012 (−0.125, 0.101)	−0.019 (−0.135, 0.097)	0.027 (−0.083, 0.138)
Access Index of park (km) (500m buffer)
Q1	Reference	Reference	Reference	Reference
Q2	−0.038 (−0.153, 0.077)	0.049 (−0.064, 0.163)	−0.006 (−0.123, 0.110)	−0.028 (−0.139, 0.082)
Q3	−0.022 (−0.141, 0.096)	−0.045 (−0.162, 0.072)	−0.026 (−0.145, 0.093)	−0.037 (−0.150, 0.076)
Q4	−0.033 (−0.148, 0.081)	−0.049 (−0.161, 0.064)	−0.055 (−0.170, 0.060)	0.029 (−0.081, 0.139)
Distance to nearest park (km)
Q1	Reference	Reference	Reference	Reference
Q2	0.038 (−0.075, 0.152)	−0.088 (−0.199, 0.024)	−0.009 (−0.123, 0.105)	0.114 (0.005, 0.223)b
Q3	0.142 (0.027, 0.257)b	−0.053 (−0.166, 0.060)	0.029 (−0.087, 0.145)	0.084 (−0.027, 0.195)
Q4	0.156 (0.037, 0.275)b	0.013 (−0.105, 0.131)	0.031 (−0.089, 0.151)	0.013 (−0.100, 0.126)

^aEffect size adjusted for participants’ individual-level covariates, PM_2.5, temperature, and ZCTA-level socioeconomic variables.

^bSignificant results at significance level of 0.05.

Given the significant results in the main analysis above, in the following analyses, we focused on PTB risks associated with vegetation levels (1000m) and total number of parks (500m). Significant associations for PTB (OR=0.63, 95% CI: 0.40–0.98) were found in the third-highest exposure groups (Q2) of 1000m vegetation during the third trimester, and the total number of parks (500m) was significantly associated with decreased PTB risks in all trimesters (Table S2). Estimated risks for most sub-groups were not significant (Figure S6). Associations were present for 1000m vegetation levels and PTB in pregnant women with post-graduate degrees (OR=0.53, 95% CI: 0.30–0.93), Asian participants (OR=0.25, 95% CI: 0.10–0.65), and those who lived in communities with higher percent Black populations (OR=0.66 (95% CI: 0.43–1.00) and higher median household income (OR=0.62, 95% CI: 0.40–0.97). Associations for the number of parks (500m) were significant in the Hispanic racial group (OR=0.57, 95% CI: 0.33–0.99) (Figure S6).

Results of sensitivity analyses were robust to: including seven participants who were originally excluded due to missing residential address during pregnancy, including variables of maternal depression and physical activity levels, and using network distance for park metrics (Tables S3–S6). The PTB-vegetation associations were found for 500m and 2000m buffers (Table S7). Associations were not found for the number of parks in 1000m buffers and PTB (Table S8). Similar to the main results (Table 4), no associations were found for Z-score birth weight based on the birthweight distribution within the cohort (Table S9).

Discussion

We examined associations between prenatal greenspace exposure and fetal and neonatal health outcomes using a prospective cohort in NYC, NY. We found significantly decreasing risks for PTB in pregnant women with higher vegetation exposure during pregnancy, but not for TLBW or EFW. Participants with more parks near home (within 500m) during pregnancy had lower PTB risks.

Our findings for vegetation index and PTB were similar to previous findings (40–43), but contrasts a birth certificate-based study in NYC, NY, showing no significant PTB-vegetation associations (34). Review articles of 11 studies concluded that the strength of evidence for the associations between PTB and vegetation index was moderate given that the overall results for the associations were not significant (14,44). Varying results may relate to differences in study populations (e.g., smoking, socioeconomic status).

We found that PTB was associated vegetation levels with 500–2000m buffers but not with 100-m or 250-m buffers (Table S7). Earlier work is unclear regarding what buffer size of greenspace is associated with adverse birth outcomes (14). Some studies found associations across different buffer sizes for greenness, but some found associations for some buffer sizes and not others (43,45). Potential reasons may relate to varying mechanisms by buffer sizes of greenspace. Vegetation in a smaller circle (e.g., 100–250m buffers) represents immediate access to greenspace, which may be relevant for psychological benefits and protection against noise, heat, and air pollution (46). Larger buffers may be large-scale greenspace (e.g., park, forest) with enhanced environment, outdoor activities, stress relief, and sociocultural activities (47). Our results found that PTB was associated with the number of parks in 500m buffers but not with 1000m buffers, which suggests that parks closer to home are more beneficial. The greenspace’s benefits for infant health by buffer zones may also be influenced by urbanicity, pollution levels, and built environment (e.g., transportation), which should be explored in future analyses.

Lower socioeconomic status may lead to less healthcare access and higher environmental hazard exposure, making greenspace more important health resource for disadvantaged sub-populations (43). In previous literature, PTB-vegetation associations were higher for less educated pregnant women (43,48). However, our analyses found decreased PTB risks for higher vegetation exposure among pregnant women with higher education and higher community-level household income. Our findings may differ as our study population had relatively higher education (27.9% with post-graduate degrees). We found PTB-vegetation associations for communities with higher percentages of Black population, aligning with previous studies showing more benefits for Black communities (38).

Study results can vary by exposure assessments for greenspace. Only a few studies used park indices for birth outcomes, with inconclusive results. Studies suggested that the associations of some health outcomes (e.g., mortality) with greenspace varied by the fragmentation of greenspace (e.g., number of parks) within an area (49). Our findings on PTB and park numbers align with these results. Greenspace scale (e.g., park area) may impact health outcomes (e.g., mortality), but we did not find associations between size and accessibility of parks and infant outcomes (50). Also, a study in Israel examined prenatal exposure to greenspace >5000m² in 300m residential buffers, finding no association with PTB (7). A cohort study in Lithuania suggested that longer distance to parks was associated with PTB and TLBW (10), whereas associations between home-park distance and PTB were not observed in Ontario, Canada (11). Large parks in our study (e.g., Central Park) may benefit both nearby and distant populations because of their scales, recreational activities, and public access, which may contribute to the lack of associations between most of our park metrics in 500m buffers or increased fetal weight (20-week gestation) for increased park distance. We also did not consider park attractiveness (e.g., facilities, amenities, safety), which was suggested to be more important than park size or proximity to promote physical activities (51).

We did not find consistently decreased TLBW risks or low birthweight among higher greenspace exposure groups. Previous literature suggested mixed results for vegetation exposure and birthweight (14,44,52). We did not find associations of fetal weight with vegetation or parks. Currently, studies for greenspace and fetal growth measures are scarce. A cohort study for Beijing found that higher EFW was associated with vegetation level in 500m buffers of maternal addresses at delivery (12). A study of four cohorts in Spain did not find associations between prenatal vegetation exposure and EFW (7–42 gestational weeks) (13). Overall, study results are inconclusive for greenspace and fetal growth biometrics.

Our study has several strengths. To our knowledge, this is the first US study to examine the impacts of greenspace on EFW, using multiple indicators based on quantity and proximity of greenspace. Maternal data were prospectively collected during follow-up visits, reducing recall bias. Further, we assessed prenatal exposure corresponding to changes in maternal residential locations throughout pregnancy. Most related epidemiological studies examined greenness based on a certain time (e.g., at delivery or recruitment) (12,36,53,54), except for a few studies considering maternal residential mobility in greenspace exposure assessments (41,55,56). Exposure over long-time periods is likely to be affected by residential mobility during pregnancy (57). Even if information on residential mobility during pregnancy may be incomplete (e.g., unreported addresses between follow-ups) in our data, our exposure measurements were based on a longitudinal approach throughout pregnancy and critical windows (e.g., trimester). Since our exposure estimates considered residential history, we expect the exposure-response relationships to be less affected by exposure misclassification. We used geocoded maternal residences, improving on birth registry-based studies that rely on administrative areas (e.g., ZIP code, county). Our analyses adjusted for environmental exposures (air pollution and temperature) (47), associated with adverse birth outcomes (32,58,59).

This study has limitations. We did not have information on participants’ daily time-activity patterns or their use or interaction with greenspace. We also lacked data on participants’ daily time-activity patterns, which remains a future research area. Recent studies used GPS devices to measure time-activity patterns and suggested links between greenspace visits and promoting physical activity during pregnancy (60). Studies also found that mobility patterns and means of travel (e.g., pedestrian vs. vehicle-based) can vary by maternal employment status or education (61), potentially leading to differences in accessibility to parks. Hence, greenspace exposure assessments based on stationary areas (e.g., home) can have misclassification for exposures. Our results based on a cohort in NYC may not be generalizable to other populations due to differences in population and greenspace. For instance, the NYC population is highly diverse. The urbanicity level of the city is high, which can be highly associated with environmental pollution and greenspace. The residential mobility patterns during pregnancy, considered in our exposure assessments, may vary in other populations. Future studies in different regions can be useful to establish comprehensive results as many factors can differ by location (e.g., population characteristics, co-exposures, healthcare, nature of greenspace).

This study found lower PTB risks associated with higher prenatal vegetation exposure and a larger number of parks near home, suggesting that higher levels of greenness near home are beneficial for infant maturity. Potential effect modification by racial regional and maternal socioeconomic status was suggested. Our findings provide epidemiological evidence for a better understanding of which greenspace features are associated with reduced risks of low fetal development and can aid decision-making for urban planning. Future studies are needed to validate the usefulness of the proximity- and quantity-based greenspace metrics as a proxy of physical activities or visits to greenspace in addition to considering various greenspace characteristics including quality and attractiveness of greenspace. Also, future studies are needed to assess exposure incorporating residential mobility of pregnant women.

Abbreviation

EFW

estimated fetal weight.

EVI

enhanced vegetation index.

LBW

low birthweight.

OR

odds ratio.

PTB

preterm birth.

TLBW

term low birthweight.

ZCTA ZIP

Code Tabulation Areas.

Data Availability Statement:

The participants of this study did not give written consent for their health data to be shared publicly, so due to the sensitive nature of the research supporting data and computing code are not available. The environmental data used in this study are publicly available from the US Environmental Protection Agency [available at https://aqs.epa.gov/aqsweb/airdata/download_files.html], the Geofabrik [https://download.geofabrik.de/north-america/us.html], and the United States Geological Survey [available at https://www.usgs.gov/landsat-missions/landsat-enhanced-vegetation-index].