Greenspace Morphology and Preterm Birth: A State-Wide Study in Georgia, United States (2001–2016)

Abstract

Background:

Residential greenness is linked to birth outcomes. However, the role of greenspace morphology remains poorly understood. Additionally, evidence is lacking regarding whether these relationships vary by subpopulation.

Objective:

We examined the association between preterm birth and residential greenspace morphology, including percentage, shape, connectedness, aggregation, closeness, and fragmentation.

Methods:

We analyzed 2,063,444 singleton live births between 2001 to 2016 in Georgia, USA. Thirty-meter resolution landcover data from National Land Cover Databased (2001–2016) were obtained to calculate greenspace morphology metrics for 1,953 census tracts in Georgia. A two-stage logistic regression examined associations between each greenspace morphology metric and preterm birth at individual level. Stratified analysis was conducted by maternal race, ethnicity, education, urbanicity, poverty rate, and greenspace percentage.

Results:

Higher greenspace percentage, aggregation, closeness, shape complexity, connectedness, and lower fragmentation were linked to a lower risk of preterm birth. After adjusting for poverty rate, associations with morphology attenuated, except for fragmentation [odds ratio (OR) = 1.014; 95% confidence interval (CI): 1.0001, 1.026] across the entire population. Strongest associations were found among black mothers and in high-poverty areas. Specifically, the odds of preterm birth in the highest quartile of greenspace percentage were 0.962 (95% CI: 0.933, 0.991) times the odds in the lowest quartile. Additionally, a lower risk of preterm birth was associated with higher greenspace aggregation (OR = 0.969; 95% CI: 0.947, 0.992), and a higher risk of preterm birth was associated with higher fragmentation (OR = 1.028; 95% CI: 1.009, 1.047), both in the black mothers group. In tracts with a high poverty rate, a lower risk of preterm birth associate with higher greenspace percentage (OR = 0.953; 95% CI: 0.910, 0.999), aggregation (OR = 0.976; 95% CI: 0.955, 0.997), and lower fragmentation (OR = 0.976; 95% CI: 0.958, 0.994). The association with greenspace morphology was most pronounced in census tracts with a medium level of greenspace percentage.

Discussion:

Our study complements other studies by showing the importance and protective effects of greenspace morphology. The observed effects are particularly prominent in census tracts characterized by a moderate level of greenspace percentage, high poverty rates, and among black women. Our findings suggest the need for tailored greenspace planning strategies based on varying levels of greenness in different areas. For locations with low greenness, increasing the greenspace percentage may be prioritized. In areas with a medium level of greenness, strategic enhancement of greenspace morphology is recommended. For areas with high greenness, the focus should be on improving spatial closeness of greenspace. https://doi.org/10.1289/EHP14571

Introduction

Preterm birth and its complications stand as the leading cause of death among children under 5 years of age globally.¹ In the United States, $\sim 1$ in every 10 infants born in 2022 were preterm, with notable disparities by race: the rate among African-American women (14.6%) was $\sim 50$ percent higher than that among white or Hispanic women (9.4% and 10.1%, respectively).² Infants born prematurely often exhibit underdeveloped lungs, skin, digestive, and immune systems, placing them at an elevated risk for neonatal complications, such as low birth weight, jaundice, breathing and feeding difficulties, infections, and challenges related to digestion and vision.³ Premature infants typically undergo hospitalization, experience higher rates of rehospitalization, and necessitate complex care upon discharge.⁴ Notably, the lifetime costs associated with preterm birth are estimated to be 10 times higher than those associated with a full-term birth in the US,⁵ and preterm births account for 61% of all maternal and neonatal costs.⁶ These factors underscore the ongoing significance of preterm birth as a major public health concern.

Apart from medical, social, and behavioral risk factors associated with preterm birth, environmental factors, such as exposure to greenspace, have been linked to improved health outcomes, including those related to childbirth.^7–9 Experimental studies indicate that visual exposure to greenspace can reduce stress, depression, anxiety, fatigue, and the time required for recovery from surgery.^10–13 Additionally, it has been associated with enhancements in mood and cognitive functions, including attention, concentration, and memory.^14–16 Observational studies at the population level have consistently shown that neighborhood greenspace is correlated with better mental, cardiovascular, and respiratory health, increased physical activity, lower rates of obesity and diabetes, improved overall quality of life, enhanced immune system function, reduced allergies, and better pregnancy outcomes.^17–26

Proximity to residential greenspace is linked to higher birth weight and lower odds of having a small-for-gestational-age baby.²⁷ Furthermore, a higher level of residential greenness measured by the normalized difference vegetation index (NDVI) has been associated with increased birth weight^28,29 and a decreased risk of preterm birth,³⁰ low birth weight,³¹ and small for gestational age infants.³² These effects remain robust even after adjusting for factors such as air pollution and traffic,³³ noise exposures, neighborhood walkability, neighborhood social economic status,³⁴ and park proximity.³⁵ Notably, one study found that such associations were only present for the black population but not for the white population,³⁶ and two other studies reported that the association was only present for mothers residing in high-density urban areas.^37,38 Apart from proximity and NDVI, an increase in the number of street trees has been associated with a decreased risk of preterm birth.³⁹ Moreover, a greater percentage of greenspace land availability has been linked to a decreased risk of preterm birth.^40,41 However, some studies have reported that the association between greenspace and birth outcomes disappears after controlling for individual and other neighborhood factors,^42,43 while others have also reported null associations.⁴⁴

All previous studies, which document greenspace and birth outcome relationships, however, have focused on quantifying the influence of greenness magnitude.⁴⁵ A crucial need remains to address where and how to invest in urban nature for optimal health benefits, especially in situations where increasing greenspace is practically challenging due to limited land availability.⁴⁶ It is thus necessary to examine the role of greenspace spatial morphology, including the fragmentation and connectivity of natural features, to help answer this question.⁴⁷ A number of studies have reported that more connected, aggregated, coherent, and complex shaped urban parks are associated with a lower mortality and morbidity risk of noncommunicable diseases, even when the greenness magnitude is comparable.^48–52 However, to our knowledge, no studies have yet explored the association between greenspace morphology and birth outcomes. This study presented the first, to our knowledge, exploration of residential greenspace morphology, encompassing factors such as percentage, shape, connectedness, aggregation, proximity, and fragmentation, in relation to preterm birth. Due to well-documented health disparities, we also investigated whether heterogeneous associations exist between subpopulations defined by maternal characteristics.

Methods

Study Area and Birth Outcomes

We procured individual level birth certificate data spanning the years 2001 to 2016 from the Office of Health Indicators for Planning within the Georgia Department of Public Health. These birth records in Georgia encompass self-reported maternal demographics and risk factors linked to adverse birth outcomes. The data is considered identifiable. Preterm birth was defined as a live birth occurring at $<37$ weeks of gestation, as determined by the clinical estimate. In cases where the clinical estimate was unavailable, gestational age was calculated using the last menstrual period date. Clinical estimates, derived through clinical assessments primarily using ultrasound, are contrasted with self-reported last menstrual period data, prone to recall bias and digit preference (e.g., preference for reporting dates such as the first or 15th of each month). Given our study’s scope covering birth certificates as well as clinical estimates, we prioritized clinical estimates for assessing preterm birth. Only 0.3% ($n=\mathrm{6,277}$) of births lacked clinical estimates, in which instances, last menstrual period–based estimates were employed. Conception date was estimated by adding 14 d to the estimated first day of gestation. Due to the absence of greenspace data for the year 2000, our analysis was limited to pregnancies with conception dates between 1 January 2001, and 31 December 2016. Singleton pregnancies with gestational ages ranging from 27 to 42 wk were included. Additionally, we excluded births with missing data. Specifically, births that are lacking census tract information regarding maternal residence and those with missing values in social demographic variables. We also excluded records where maternal age was $<15$ years or $>49$ years and births with a birth weight exceeding $\mathrm{7,500}\mathrm{g}$ and below the first quantile ($\mathrm{1,953}\mathrm{g}$). This left us with 2,040,007 singleton live births. The study was approved by the Institutional Board by Emory University (Study ID: 00102330). Informed consent was waived because this study is a secondary data analysis and did not involve contacting participants.

Quantification of Greenspace Morphology

In this study, we acquired $30\text{-m}$ resolution National Land Cover Database imagery for the years 2001, 2004, 2006, 2008, 2011, 2013, and 2016 from the United States Geological Survey via The Multi-Resolution Land Characteristics (MRLC) consortium platform.⁵³ The MRLC consortium platform represents a collaborative effort among federal agencies dedicated to coordinating and producing consistent and pertinent land cover information on a national scale for diverse applications within the field. The obtained imagery encompasses various classes and subcategories, delineated as water, developed areas, barren land, forests, shrublands, herbaceous areas, planted areas, and wetlands. Our classification methodology involved categorizing low-, medium-, and high-intensity developed areas, barren areas, and water spaces as nongreen land cover. The remaining classes, characterized predominantly by vegetated land, were classified as green land cover using ArcGIS 10.8, based on a $30\text{-}\times 30\text{-m}$ grid. For the purposes of this investigation, greenspace was operationally defined to encompass grasslands, shrublands, forests, and developed open spaces that were predominantly characterized by vegetation in the form of lawn grasses. These open spaces consist of over 80% vegetated land, typically including parks, golf courses, and urban vegetation planted for recreation or esthetic reasons.

We employed a comprehensive methodology to quantify greenspace morphology for 1,953 census tracts in Georgia (Figure 1). Based on the 2016 American Community Survey 5-year estimates, the average population of census tracts in Georgia is 3,067, with an interquartile range of 2,884. In Georgia, census tracts have an average area of $78.2{\mathrm{km}}^2$ with an interquartile range of $94.8{\mathrm{km}}^2$. For each tract, we established a $0.5\text{-mi}$ Euclidean buffer extending from all edges of each tract, reflecting an approximate 10-min walking distance, ensuring genuine real-world exposure to greenspace within the respective tracts. In densely populated tracts, these buffers may overlap. However, since our focus is on capturing genuine greenspace exposure, it’s possible that residents in neighboring tracts share exposure to the same areas of overlap. Therefore, the overlap is not a significant concern for this study. This buffer method, widely used in environmental and health studies, has been validated for capturing residential exposure to greenspace.^51,54–57 Data processing utilized a custom Python 3.1 code for ArcGIS 10.8. Subsequently, we computed six landscape metrics for each buffered tract using the landscapemetrics package in R. The selection of metrics was driven by three specific goals: first, to use metrics that have been consistently validated in prior studies and are linked to chronic conditions known as risk factors for preterm birth^48,51,58; second, to pinpoint spatial morphological features of greenspaces that enhance ecological functions, such as cooling and air pollution reduction, which are proven to impact birth outcomes^59–62; and third, to identify morphological attributes that are easily understandable by city planners and policymakers. Metrics chosen include fragmentation, connectedness, aggregation, distantness, shape complexity, and the percentage of land area that is considered greenspace (see Figure 2). In the absence of land cover data for certain years, we estimated greenspace morphology metrics with available data by employing a linear interpolation between years. Considering that the process of greenspace-related land planning and construction is typically incremental and that there is only a 1- to 2-year gap between the available data points, we believe that linear interpolation is a reasonable approach in this scenario. Therefore, the greenspace morphology metrics data were compiled annually from 2001 to 2016 at the census tract level. The tract-level greenspace morphology metrics were then linked to each mother at the individual level based on their geographic locations. An illustration of the greenspace spatial morphology captured by these indices is presented in Figure 2, while detailed formulas for metric calculation are available in Figure 3.

Figure 1.

Illustration of greenspace morphological metrics data processing procedure. This figure shows the steps involved in processing greenspace morphology data. First, a half-mile buffer was created around each tract of land. This buffer was then used to crop the corresponding greenspace map. The cropped map is subsequently utilized for calculating various greenspace morphological metrics. The figure was created using a land cover base map from the National Land Cover Database and processed with PowerPoint software.

Figure 2.

Illustration of the landscape metrics selected in reflecting greenspace spatial morphology. The six metrics each capture a distinct spatial characteristic: percentage, fragmentation, connectedness, aggregation, shape complexity, and closeness. (Some images in this table are directly adopted from a previous paper,⁵⁰ with additional images created and provided by the authors of the current study for the purpose of enhancing the comprehensiveness of the data presentation. Copyright 2024 Huaqing Wang. Used with permission.) Figure was created by using AutoCAD and Adobe Illustrator.

Figure 3.

Detailed formula for calculating greenspace morphology metrics. (Information in this figure is referenced from the FRAGSTATS HELP manual.⁶³)

Covariates

We controlled for individual level covariates previously identified as predictors of adverse birth outcomes. A comprehensive list of these variables is provided in Table 1. Given the substantial impact of maternal age on birth outcomes, with both younger and older maternal ages associated with adverse effects,^64,65 we incorporated maternal age as a categorical variable in our model. Age was grouped into 16–19, 20–24, 25–29, 30–34, 35–39, 40–44, and 45–49. Higher educational attainment is often correlated with improved health practices and greater prenatal care engagement, which can mitigate risks of preterm birth⁶⁶; thus, maternal education levels were also included as a categorical variable (categorized as $<$ high school, high school or equivalent, some college/associate degree, bachelor’s or above). Race and ethnicity were accounted for, recognizing that African American and Hispanic women experience higher rates of preterm birth compared to non-Hispanic white women.^67,68 Race is categorized as white, black, Asian/Native Hawaiian/Pacific Islander, and other. Maternal ethnicity is categorized as Hispanic and non-Hispanic. Additionally, marital status was controlled (married, unmarried) as married women generally have lower rates of preterm birth compared to unmarried women.⁶⁹ Parity was included based on its association with preterm birth risk, with nulliparity being a notable factor.⁷⁰ Therefore, we categorized parity as one and greater or equal to two. We adjusted for tobacco use (yes, no) during pregnancy, given its known link to preterm birth.⁷¹ Finally, year of conception, which represents the year in which each individual (birth) was conceived, is also controlled. Additionally, temporal covariates included in the models were the season of conception (Spring: March–May, Summer: June–August, Autumn: September–November, Winter: December–February). This is because of its established association with preterm birth.⁷² These data were obtained through the birth certificate record from the Office of Health Indicators for Planning within the Georgia Department of Public Health.

Table 1.

Maternal characteristics of Georgia singleton live birth conceived between 2001 and 2016, stratified by preterm and term birth status.

Maternal characteristics	Preterm births	Term births	Overall
	($n=\mathrm{164,369}$)	($n=\mathrm{1,875,638}$)	($n=\mathrm{2,040,007}$)
Age
15–19	18,450 (11.2%)	183,747 (9.8%)	202,197 (9.9%)
20–24	43,989 (26.9%)	490,117 (26.1%)	534,106 (26.2%)
25–29	42,607 (25.9%)	526,127 (28.1%)	568,734 (27.9%)
30–34	35,014 (21.3%)	429,786 (22.9%)	464,800 (22.8%)
35–39	18,912 (11.5%)	202,096 (10.8%)	221,008 (10.8%)
40–44	5,048 (3.1%)	41,491 (2.2%)	46,539 (2.3%)
45–49	338 (0.2%)	2,214 (0.1%)	2,552 (0.1%)
Education
$<$High school	8,244 (5.2%)	99,849 (5.5%)	108,093 (5.5%)
High school/general educational development	27,950 (17.6%)	263,748 (14.5%)	291,698 (14.8%)
Some college/AA	51,907 (32.8%)	547,677 (30.2%)	599,584 (30.3%)
Bachelor’s degree or higher	70,389 (44.4%)	904,172 (49.8%)	974,561 (49.4%)
Missing	5,879	60,192	66,071
Race
White	87,091 (53.5%)	1,214,375 (62.4%)	1,214,375 (60.1%)
Black	66,754 (41.0%)	605,046 (31.1%)	671,800 (33.2%)
Asian/non-Hispanic/Pacific Islander	4,936 (3.0%)	74,050 (3.8%)	78,986 (3.9%)
American Indian/Alaska Native	267 (0.2%)	3,418 (0.2%)	3,685 (0.2%)
Other	3,811 (2.3%)	48,307 (2.5%)	52,118 (2.6%)
Missing	1,510	17,533	19,043
Ethnicity
Non-Hispanic	139,417 (87.3%)	1,554,135 (84.6%)	1,693,552 (84.9%)
Hispanic	20,208 (12.7%)	281,464 (15.4%)	301,672 (15.1%)
Missing	4,744	40,039	44,783
Marital status
Unmarried	80,879 (49.5%)	790,713 (42.3%)	871,592 (42.8%)
Married	82,689 (50.5%)	1,079,924 (57.7%)	1,162,613 (57.2%)
Missing	801	5,001	5,802
Parity
First	63,835 (38.8%)	755,272 (40.3%)	819,107 (40.2%)
Second or more	100,534 (61.2%)	1,129,366 (59.7%)	1,220,900 (59.8%)
Conception season
Spring	39,952 (24.3%)	454,940 (24.3%)	494,892 (24.3%)
Summer	42,869 (26.1%)	484,004 (25.8%)	526,873 (25.8%)
Autumn	40,881 (24.9%)	476,167 (25.4%)	517,048 (25.3%)
Winter	40,667 (24.6%)	460,527 (24.6%)	501,194 (24.6%)
Tobacco use during pregnancy
No	148,096 (91.2%)	1,744,573 (93.6%)	1,892,669 (93.4%)
Yes	14,202 (8.8%)	118,542 (6.4%)	132,744 (6.6%)
Missing	2,071	12,523	14,594

In a secondary analysis, we further adjusted for the percentage of the population that is considered poverty at the census tract level. Census tract-level poverty data were linked to each mother at the individual level based on her address. Such poverty rate data were obtained from American Community Survey (ACS) from three mutually exclusive periods (2005–2009, 2010–2014, 2015–2019). Because ACS data were not available before 2005, for births with conception date in 2001–2004 were assigned the values from 2005–2009. The ACS measures poverty by collecting data on household income over the past 12 months and comparing it to the federal poverty thresholds, which are updated annually.

Statistical Analysis

We employed a two-stage logistic regression analysis to explore the relationships between greenspace morphology metrics and preterm birth. In the first stage, we analyzed such associations at the individual level for each of the 159 counties. This resulted in county-specific log odds ratios for each county independently. In the second stage, we aggregated these 159 risk estimates across counties using a random-effect meta-analysis. We initially considered conducting the first-stage analysis at the census tract level; however, this approach was ultimately dismissed due to insufficient sample sizes and the low incidence of preterm birth events within individual tracts. We then adopted counties for the analysis, as counties are commonly employed in spatial analysis due to their well-defined geographical and administrative boundaries, facilitating policy implementation and comparative analyses. The urbanicity variable, which was used to stratify the data into quantiles for subsequent analysis, was also defined at the county level. Our two-stage logistic regression approach allows for the incorporation of county-specific intercepts as fixed effects, capturing variations across counties while maintaining computational efficiency, particularly during sensitivity analyses. Additionally, this approach enables us to discern how the effects of confounders differ within each county, enhancing control over confounding factors compared to single-model random-intercept analysis. We did not include population density as a controlled variable, as urbanicity is conceptually regarded as equivalent to population density.

We employed stratification to investigate potential variations in the relationships between greenspace morphology and preterm birth across subpopulations. The data was stratified based on maternal race (black vs. non-black), maternal ethnicity (Hispanic vs. non-Hispanic), maternal education ($\le \text{high school}$ vs. $>\text{high school}$), pregnancies from urban counties (first through third quantile group vs. fourth quantile group), pregnancies residing in high-poverty census tracts (first through third quantile group vs. fourth quantile group), and pregnancies residing in high urbanicity counties (first through third quantile group vs. fourth quantile group). These quantiles serve as the effect modifiers. We initially examined the relationships between greenspace morphology and preterm birth across each racial category and found that, compared to the black group, the other racial categories were largely similar and less distinct (Figure S1). Furthermore, the smaller sample sizes of the Asian and other groups posed challenges for data interpretation and presentation. As a result, we stratified participants into two groups: black and non-black. All covariates cited in the primary model were adjusted for within every stratification subgroup, except for the one subjected to stratification. These covariates included maternal age, education, race, ethnicity, marital status, birth parity, smoking, year of conception, season of conception, and poverty level.

Additionally, we explored two approaches to assess the combined impact of greenspace morphology and greenness level, measured by the percentage of greenspace—a widely used metric in prior studies.^7,41,73 In the initial approach, we implemented two-exposure models that incorporated greenspace percentage and each of the other five morphology measures independently, considering the strong correlation among the greenspace morphology metrics. In this dual-exposure analysis, our aim was to quantify the independent contribution of greenspace morphology metrics beyond the level of greenness. In the second approach, we investigated whether the level of greenness modifies the effect of greenspace morphology. For ease of interpretation, we categorized the greenspace percentage measures into tertiles using the 33rd and 66th quantile values and assessed the association of the other five greenness morphology measures within each greenspace percentage tertile. Furthermore, a sensitivity analysis was performed using only the years 2001, 2004, 2006, 2008, 2011, 2013, and 2016, that comes with the land cover data, excluding the interpolated years. We also incorporated an additional sensitivity analysis to investigate the relationships between greenspace morphology and preterm birth in the five most densely populated counties in Atlanta and 20 counties across Georgia.

To address between-county heterogeneity (Table S1), we utilized a random-effects meta-analysis to aggregate county-specific log odds ratios. In each analysis, county-specific estimates with a standard error $>100$ were excluded due to small sample size. Pooled odds ratios (ORs) were reported for an interquartile range (IQR) increment for each morphology metric, enhancing the comparability of risk estimates across exposure measures with varying variabilities and scales. IQR values were computed based on all pregnancies across the state. All statistical analyses were conducted using R software, version 4.2.0. The R source code used for the statistical analysis is available in the supplemental data (supplementary file “Supplementary_SourceCode_TwoStageLogisticRegression.R”). All figures and maps were created using the ggplot package in R, unless otherwise noted. We considered an alpha value of $<0.05$ to be statistically significant.

Results

The study encompassed a total of 2,040,007 singleton live births within 1,953 census tracts, with 164,369 (8.9%) classified as preterm. In Table 1, we present summary statistics of maternal characteristics, stratified by preterm and full-term births. Higher rates of preterm birth were associated with established maternal risk factors, including maternal age,^74,75 lower educational attainment,^76,77 black race,⁷⁸ non-Hispanic ethnicity,⁷⁹ unmarried status,⁶⁹ and tobacco use during pregnancy.⁸⁰ The detailed results of the first-stage analysis at the individual level for each of the 159 counties are provided in Supplementary Excel Tables S1–S6. Table 2 provides a summary of the census tract-level greenspace morphology metrics. Among these metrics, greenspace percentage exhibited the greatest variability between tracts, while cohesion demonstrated the smallest variability. Box plots and histograms depicting each greenspace metric are displayed in Figures S2 and S3. Supplementary Figure S4 depicts exposure choropleth maps for the year 2010, encompassing the entire state of Georgia and the five-county Atlanta metropolitan area. These maps highlight significant spatial variability in exposure both across the state and within urban areas. Figure 4 presents pairwise correlations between greenness exposure metrics at census tract level. Notably, the percentage of greenspace exhibited positive correlations with aggregation ($r=0.95$), cohesion ($r=0.65$), and complexity ($r=0.52$), and negative correlations with fragmentation ($r=-0.90$) and distantness ($r=-0.57$).

Table 2.

Summary statistics of census tract level greenspace morphology metrics in Georgia state, 2001–2016 (census tracts, $n=\mathrm{1,953}$).

Greenspace morphology metrics	Minimum	Q1	Median	Q3	Maximum	SD	IQR
Percentage (%)	0.9	49.0	66.4	88.7	99.8	22.9	39.8
Aggregation (%)	30.1	81.4	89.2	96.1	99.8	9.2	14.7
Closeness (meter)	60.0	69.7	72.9	77.3	229.4	8.8	7.5
Shape complexity (unit)	1.1	6.6	9.1	11.4	21.7	3.6	4.8
Fragmentation (number per 100 hectares)	0.0	0.7	3.8	7.9	30.8	4.8	7.2
Cohesion (unit)	36.3	98.2	99.5	99.9	100.0	3.6	1.7

Note: Percentage (PLAND), the percentage of greenspace in a census tract; Aggregation (AI), the measure of the aggregation of greenness distribution. The higher, the more spatially aggregated; Closeness (ENN), the mean Euclidean nearest neighbor distance—whether the greenspaces in a census tract are spatially close to each other. Shape complexity (SHAPE), the measure of the shape complexity of greenspace distribution. The higher, the more complex the shape; Fragmentation (PD), the fragmentation of greenspace. The higher, the more fragmented; Connectedness (COHESION), the measure of connectedness of greenspace in a census tract. The higher, the more connected. IQR, interquartile range; SD, standard deviation.

Figure 4.

Pairwise Pearson’s correlations greenspace morphology metrics at census tract level. Georgia, 2001–2016. $n=\mathrm{1,953}$ census tracts. Note: Percentage (PLAND), the percentage of greenspace in a census tract; Aggregation (AI), the measure of the aggregation of greenness distribution. The higher, the more spatially aggregated; Distantness (ENN), the mean Euclidean nearest neighbor distance—whether the greenspaces in a census tract are spatially close to each other; Shape complexity (SHAPE), the measure of the shape complexity of greenspace distribution. The higher, the more complex the shape; Fragmentation (PD), the fragmentation of greenspace. The higher, the more fragmented; Connectedness (COHESION), the measure of connectedness of greenspace in a census tract. The higher, the more connected.

After excluding missing values when “percentage” is the greenspace morphology in the model, we analyzed a total of 1,988,625 births. This includes 659,476 births for the black subpopulation and 1,329,149 births for the non-black subpopulation, 300,786 births for the Hispanic subpopulation and 1,687,839 births for the non-Hispanic subpopulation, and 392,422 births for individuals with education less than high school and 1,538,593 births for those with education greater than high school. The Q1–Q3 urbanicity group (urbanicity below 58.30%) includes 469,320 births, while the Q4 group (urbanicity between 58.30% and 99.75%) includes 1,519,305 births. In relation to poverty, 1,492,186 births occurred in Q1–Q3 census tracts, where $<20.60\%$ of the population lives below the poverty line, while 496,439 births occurred in Q4 tracts, where 20.63% or more of the population lives below the poverty line (Supplementary Table S2).

Figure S5 and Figure 5 displays the overall association between greenspace percentage and preterm birth, alongside the association stratified by maternal and neighborhood characteristics. Numerical values are detailed in Supplementary Table S3. Higher neighborhood greenspace percentage associated with a reduced risk of preterm birth [OR = 0.966; 95% confidence interval (CI): 0.952, 0.981]. This association was more pronounced among black mothers (OR = 0.992; 95% CI: 0.909, 0.956), predominantly influenced by census tracts located in high-urbanicity counties with greater economic disadvantage. In Figure 6 and Figure S6, the overall odds ratios are presented between six greenspace morphology metrics and preterm birth. Consistent protective associations were observed for higher level of percentage, aggregation, shape complexity, connectedness, and lower values of distantness as well as fragmentation of greenspace morphology concerning preterm birth. Stratified analysis showed such associations are stronger among black mothers, those with education beyond high school, and mothers in Q4 urbanicity areas. Significant findings were also observed in both Hispanic and non-Hispanic groups, with similar trends in odds ratios.

Figure 5.

Odds ratios and 95% confidence intervals (CI) from primary single exposure model and stratification analysis for preterm birth per interquartile range (IQR) increase in census tract greenspace percentage among Georgia live births, 2001 to 2016. Covariates controlled for in the analysis included age, education, race, ethnicity, marital status, parity, conception season, tobacco use during pregnancy, urbanicity, and poverty, except the stratified variable. Greenspace Percentage $\text{IQR}=39.760$, and $n=\mathrm{1,988,625}$. (The numeric data generated for this figure can be found in Supplement Table S3. Detailed sample sizes for each stratification can be found in Table S2.)

Figure 6.

Odds ratios (OR) and 95% confidence intervals for preterm birth per interquartile range (IQR) change in census tract level greenspace morphology among Georgia live births, 2001 to 2016. Covariates controlled for in the analysis included age, education, race, ethnicity, marital status, parity, conception season, tobacco use during pregnancy, urbanicity, and poverty. All models have 1,988,625 births for analysis. Greenspace Percentage $\text{IQR}=39.760$; Aggregation $\text{IQR}=14.474$; Distantness $\text{IQR}=7.542$; Shape $\text{IQR}=4.766$; Fragmentation $\text{IQR}=7.214$; Connectedness $\text{IQR}=1.716$. (The numeric data generated for this figure can be found in Supplement Table S8.). Note: Percentage (PLAND), the percentage of greenspace in a census tract; Aggregation (AI), the measure of the aggregation of greenness distribution. The higher, the more spatially aggregated; Distantness (ENN), the mean Euclidean nearest neighbor distance—whether the greenspaces are spatially close to each other; Shape complexity (SHAPE), the measure of the shape complexity of greenspace distribution. The higher, the more complex the shape; Fragmentation (PD), the fragmentation of greenspace. The higher, the more fragmented; Connectedness (COHESION), the measure of connectedness of greenspace in a census tract. The higher, the more connected.

In our secondary analysis, with additional adjustment for census tract poverty level, we observed an attenuation in the overall statewide associations between greenspace morphology and preterm birth outcomes except the fragmentation (Supplementary Table S3 and Figure S7). Higher greenspace fragmentation retained statistical significance (OR = 1.014; 95% CI: 1.001, 1.026) across the entire population. Notably, among black mothers, the protective associations of greenspace morphology remained. Specifically, per IQR higher in greenspace percentage (OR = 0.962; 95% CI: 0.933, 0.991) and aggregation (OR = 0.969; 95% CI: 0.947, 0.992), there was an association with a lower risk of preterm birth. Per IQR higher in fragmentation (OR = 1.028, 95% CI: 1.009, 1.047) in greenspace morphology, there was an association with a higher risk of preterm birth. In the Hispanic mother group, greenspace percentage and connectedness remained significantly associated with preterm birth with poverty level controlled. The odds ratios per IQR increase in overall percent greenspace and connectedness were 0.944 (95% CI: 0.895, 0.997) and 0.988 (95% CI: 0.977, 0.998), respectively. Among tracts with a higher percentage of the poverty population, higher values in greenspace percentage (OR = 0.953; 95% CI: 0.910, 0.999) and aggregation (OR = 0.976; 95% CI: 0.955, 0.997) were associated with a lower risk of preterm birth, while higher value in greenspace fragmentation (OR = 1.025; 95% CI: 1.006, 1.048) was associated with a higher risk of preterm birth.

In the two-exposure model (Supplementary Table S4), we observed minimal additional protective effects of morphology beyond percent greenness. However, when greenspace percentage was categorized into low, medium, and high levels, a clear joint effect of greenspace percentage and morphology on preterm birth risk emerged (Figure 7; Table S5). A consistent U-shaped joint effect was observed. Specifically, the association with greenspace morphology was most pronounced in census tracts with a medium level of greenspace percentage (in the second tertile, between 55.24% and 82.36%). For instance, higher fragmentation had an odds ratio of 1.029 (95% CI: 1.010, 1.049) in the second tertile of percent greenness but was null in the other two tertiles (Supplementary Table S6). The sensitivity analysis using only the years associated with land cover data showed that the statewide overall associations between greenspace measures and preterm birth matched our initial findings. Nonetheless, these estimates exhibited broader confidence intervals and lost statistical significance given an $\sim 50\%$ reduction in sample size. Similarly, the sensitivity analyses conducted for both the 5-county and 20-county subsets unveiled comparable overall trends. However, some of the associations observed became statistically insignificant (refer to Supplementary Figure S8 and Figure S9).

Figure 7.

Joint effects of greenspace percentage and greenspace morphology on the risk of preterm birth in Georgia, 2001–2016. Covariates controlled for in the analysis included age, education, race, ethnicity, marital status, parity, conception season, tobacco use during pregnancy, urbanicity, and poverty. Stratified by greenspace percentage. Lower one-third quantile group of the greenspace percentage (0.889–55.273) has 662,893 births, medium one-third quantile group (55.273–82.363) has 662,867 births, and upper one-third quantile group (82.363–99.761) has 662,826 births. (The numeric data generated for this figure can be found in Supplement Tables S6 and S9.). Note: Percentage (PLAND), the percentage of greenspace in a census tract; Aggregation (AI), the measure of the aggregation of greenness distribution. The higher, the more spatially aggregated; Distantness (ENN), the mean Euclidean nearest neighbor distance—whether the greenspaces are spatially close to each other; Shape complexity (SHAPE), the measure of the shape complexity of greenspace distribution. The higher, the more complex the shape; Fragmentation (PD), the fragmentation of greenspace. The higher, the more fragmented; Connectedness (COHESION), the measure of connectedness of greenspace in a census tract. The higher, the more connected. CI, confidence interval; IQR, interquartile range.

Discussion

Our study complements other studies by showing the potential protective effects of greenspace morphology. To the best of our knowledge, this study represents the first exploration of the association between greenspace morphology and preterm birth outcomes. Our findings suggest that pregnant women residing in urban areas characterized by a higher greenspace percentage, more aggregated and connected greenspaces, spatially close distribution, and complex shapes are associated with a reduced risk of preterm birth. Notably, this association is strongest for black women and individuals living in areas marked by a high poverty rate. The relationship between greenspace morphology and preterm birth is most evident in urban areas with a medium level of greenspace percentage. These significant associations were discerned through the analysis of $30\text{-m}$ resolution land cover data in the state of Georgia, utilizing established landscape metrics and drawing on reliable health outcomes data sources.

Our study builds upon existing evidence by first reaffirming the inverse relationship between the overall greenspace percentage and preterm birth, aligning with findings from prior investigations.^40,41 Further, to the extent of our knowledge, while no prior studies have explored the link between greenspace morphology and preterm birth, our results regarding aggregated, connected, spatially close, and complex-shaped greenspaces are consistent with observations from other domains. In our study, greenspace is defined as all land cover categories in the national land cover database that are predominantly covered by vegetation. This includes forests, shrublands, herbaceous areas, planted areas, wetlands, and other similar categories. Census tracts characterized by connected, aggregated, and complex-shaped and less fragmented greenspace morphology have previously been associated with reduced risks of all-cause, cardiac, respiratory, and neoplasm mortality,⁵¹ as well as impacts on life expectancy.⁸¹ Additionally, these morphological features are linked to the prevalence of various health outcomes, including poor mental health, heart disease, stroke, diabetes, chronic obstructive pulmonary disease (COPD), and insufficient leisure time physical activity, even when greenness magnitudes are comparable.⁵⁰ Moreover, increased aggregation and connectedness of greenspace morphology have shown associations with a reduced risk of death from colon cancer,⁵² lower frailty among older adults,⁵⁸ decreased diagnosis of schizophrenia,⁸² reduced mental distress,⁸³ and lower obesity among Hispanic children.⁸¹

Furthermore, we observed that the association between greenspace morphology and preterm birth weakened after adjusting for the poverty rate but remained significant for black women and in areas with high poverty rates. This attenuation may be attributed to a notable correlation between poverty rates and greenspace availability in the State of Georgia.⁸⁴ Wealthier individuals might choose to reside in areas with more favorable greenspace, whereas economically disadvantaged individuals may not have equivalent access to such amenities in their neighborhoods.⁸⁵ Additionally, individuals with higher socioeconomic status may possess greater resources related to health and health care, which could exert a more significant influence on preterm birth compared to greenspace.⁸⁶ Further, black communities often reside in regions characterized by heightened levels of environmental pollutants, such as air pollution.⁸⁷ In these circumstances, greenspaces and their morphology might hold the potential to provide more substantial alleviation against adverse health impacts. A previous study from Australia indicated that neighborhoods with lower income levels have less greenspace.⁸⁸ Similarly, across the United States, block groups with lower income and a majority of persons of color tend to have less greenery and fewer parks.⁸⁹ We noted that a higher percentage, aggregation, and less fragmented greenspace morphology were associated with a lower risk of preterm birth in high-poverty-rate areas and for black women even after adjusting for poverty rate. This consistency with previous observations regarding racial differences is supported by studies reporting that greenness levels, measured using the normalized difference vegetation index (NDVI) at a $30\text{-m}$ resolution, were significantly associated with higher birth weight only in the black population and not in the white population.³⁶ Another study reported an association between NDVI and birth weight only among individuals with the lowest education levels.⁹⁰ Our study contributes value by highlighting that not only does the level of greenness matter, but the spatial morphology of greenspace also plays a role.

One novel finding is that we have uniquely identified that the associations between greenspace morphology and preterm birth are most prominent when there is a medium level of greenness, as measured by the percentage of greenspace. This underscores the importance of morphology, particularly when a moderate amount of greenness is present in a census tract. In areas with a moderate level of greenspace, residents might have sufficient but not overwhelming access to greenspaces. This optimal exposure might allow the distinct features of the greenspace morphology to more significantly impact health outcomes. In areas with a lower percentage of greenspace, the distribution may be too sparse to offer much variation in morphology, which, alongside its scarcity, limits its effectiveness. Conversely, areas with a high percentage of greenspace often feature large, well-connected parks that are cohesive enough, which could potentially overshadow the importance of specific morphological traits. This discovery may be attributed to the utilization of $30\text{-m}$ resolution land cover data, enabling the capture of each park with a size $>900$ square meters. This finding aligns with, yet differs slightly from, two previous studies that reported significant relationships between greenspace morphology and mortality and morbidity while controlling for greenness, measured by the total area of greenspace, at the census tract level.^50,51 The variance may be attributed to these two studies using 1-m-high resolution imagery for greenspace morphology calculations, allowing them to capture greenspaces with a size down to $1{\mathrm{m}}^2$. The abundance of greenspace parcels between 1 and $900{\mathrm{m}}^2$ may influence these results. Additionally, further study is needed to better understand the reasons behind these variations.

The association between greenspace morphology and preterm birth may be attributed to two primary factors, the first being the ecological functions involved. Landscape morphology influences ecological services such as the reduction of air pollution and cooling effects,^91,92 which impact birth outcomes. The density and shape of greenspace influence particulate matter with aerodynamic diameter $\le 2.5\mathrm{\mu }\mathrm{m}$ (${\mathrm{PM}}_{2.5}$), ${\mathrm{NO}}_2$, and ${\mathrm{SO}}_2$ levels.⁹³ Larger, aggregated, and less fragmented greenspace are linked to reduced air pollution and cooler temperatures.⁶⁰ Additionally, the presence of large, complex-shaped greenspace is associated with lower particulate matter concentration.⁹⁴ Air pollution, known to increase toxic chemicals in the blood and induce immune system stress, can weaken the placenta surrounding the fetus, potentially leading to preterm birth.^95,96 Numerous studies have reported associations between air pollution and adverse pregnancy outcomes.^26,97–99 A systematic review also highlighted an association between heat, ozone, or fine particulate matter and adverse pregnancy outcomes.⁶¹ Another study identified ${\mathrm{PM}}_{2.5}$ concentration as a mediator in the association between greenspace morphology and noncommunicable diseases.⁵⁰ Additionally, two studies reported that air pollution acts as a mediator in the association between greenspace fragmentation and respiratory mortality.^100,101 Except for air pollution, the cooling effect of greenspace may also contribute to better birth outcomes. Clustered vegetation effectively reduces surface temperatures compared to dispersed and fragmented patterns.¹⁰² Aggregated greenspace morphology and the size of greenspace notably impact land surface temperature reduction.¹⁰³ High ambient temperature has been linked to adverse birth outcomes.¹⁰⁴ Furthermore, connected, aggregated, cohesive, and complex-shaped greenspace are associated with higher biodiversity levels, promoting improved well-being and reduced risk of inflammatory disorder.^105–107 The balance of inflammatory cytokines predicts birth weight.¹⁰⁸ Further investigations are warranted to explore the extent to which the ecological functions of greenspace morphology are linked to pregnancy outcomes.

The second potential pathway linking greenspace morphology and preterm birth could involve the behavior of pregnant mothers. Greenspace morphology might influence how likely pregnant women are to be exposed to and utilize greenspace, thereby contributing to birth outcomes. However, these mechanisms require further exploration. Current evidence suggests that leisure-time physical activity mediates the association between greenspace morphology and the prevalence of noncommunicable diseases.⁵⁰ The strength of this mediating effect is $\sim 35$ times greater than the mediating effect of ${\mathrm{PM}}_{2.5}$ concentration.⁵⁰ Since higher leisure-time activity is associated with a reduced risk of preterm birth,¹⁰⁹ it could potentially serve as a mediator between greenspace morphology and birth outcomes. Connected parks might provide an opportunity for pregnant women to spend more time in greenspace by walking or biking from one park to another without leaving greenspace, improving the mental health of pregnant women,¹¹⁰ and potentially contributing to positive birth outcomes. Nevertheless, additional studies are needed to investigate the mediating or moderating role of human behavior in the observed relationship between greenspace morphology and preterm birth.

From the perspective of landscape and city designers, establishing green belts along streets to connect existing parks might be a practical approach for achieving an optimal greenspace distribution. The addition of isolated small lawn parcels in front of each building may not be as advantageous as enhancing a preplanned or existing relatively large park within a community. In cases where fragmented lawns already exist, planting trees in the gap areas represents a practical and cost-effective means of spatially linking them by providing a substantial canopy. Moreover, transforming a large park to create a more complex shape, with increased entry points and border areas, may enhance park accessibility to a broader population and thereby contribute to favorable birth outcomes. Our findings support the concept that in areas with low levels of greenness, augmenting greenness by incorporating parks larger than $900{\mathrm{m}}^2$ is likely to be beneficial. In areas with a medium level of greenness, strategic planning and design of parks to promote favorable greenspace morphology by increasing aggregation, shape complexity, connectedness, and reducing fragmentation might be advantageous. For areas with higher levels of greenness, improving spatial closeness in greenspace might be beneficial for birth outcomes.

Our study is subject to certain limitations. First, the absence of a direct measure of exposure to greenspace means that we cannot determine whether pregnant women visited or spent time in these greenspaces. Our findings based on aggregate census tract level greenspace data may not accurately reflect individual-level relationships between greenspace and birth outcomes. Individual-level studies are required to investigate the effects of greenspace morphology on real-world exposure to nature. Second, our assessment focused on all vegetated land cover, overlooking variations in type, accessibility, and quality of greenspace. Thirdly, our study was limited to mothers residing in the state of Georgia, which restricts the generalizability of our findings. For instance, our study population comprised a significant proportion of African American women (33.2%), making it challenging to extrapolate these results to populations in states like California, where African Americans constitute only around 7% of the population.¹¹¹ Additionally, Georgia is known for its extensive green and forested areas, with nearly 67% of the state covered by forests.¹¹² This makes our results difficult to apply to other states such as North Dakota (1.72% forest coverage), Nebraska (3.2%), and South Dakota (3.93%), which have notably limited forest coverage.¹¹³ Further, this research focuses solely on live births, despite many pregnancies ending in losses (such as spontaneous abortions and stillbirths). Therefore, the interpretations of the associations presented here should be limited to pregnancies that culminated in live births. Finally, our study utilized $30\text{-m}$ resolution land cover data, which cannot capture greenspaces smaller than $900{\mathrm{m}}^2$ and cannot account for seasonal variations in greenspace exposure, as may occur in the state of Georgia. Exploring studies based on finer resolution and temporal greenspace maps is valuable, as they may provide evidence for health-promoting greenspace projects concerning smaller pocket parks, individual trees, and seasonal variations.

In conclusion, our study demonstrates that neighborhood greenspace morphology, specifically characterized by spatial closeness, connectedness, aggregation, cohesion, and complex shapes, is associated with a lower risk of preterm birth. This effect is particularly pronounced in urban areas with a medium level of greenspace amount, in census tracts with a high poverty rate, and among black women. Our findings suggest that not every greenspace contributes equally to health benefits in terms of birth outcomes. Investing in urban nature to enhance spatial closeness, connectedness, aggregation, cohesion, and shape complexity, and filling in spatial gaps between greenspaces at the neighborhood scale might lead to improved birth outcomes. These insights carry significant practical and policy implications for optimizing the spatial arrangement of greenspace at the neighborhood level. They also provide guidance for city and landscape planners, emphasizing the importance of alternative greenspace designs in promoting healthy communities.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.