Neighborhood socioeconomic status, green space, and walkability and risk for falls among postmenopausal women: The Women’s Health Initiative

Abstract

Purpose:

This study estimated associations between neighborhood socioeconomic status (NSES), walkability, green space, and incident falls among postmenopausal women and evaluated modifiers of these associations, including study arm, race and ethnicity, baseline household income, baseline walking, age at enrollment, baseline low physical functioning, baseline fall history, climate region, and urban-rural residence.

Methods:

The Women’s Health Initiative recruited a national sample of postmenopausal women (50–79 years) across 40 U.S. clinical centers and conducted yearly assessments from 1993 to 2005 (n=161,808). Women reporting a history of hip fracture or walking limitations were excluded, yielding a final sample of 157,583 participants. Falling was reported annually. NSES (income/wealth, education, occupation), walkability (population density, diversity of land cover, nearby high-traffic roadways), and green space (exposure to vegetation) were calculated annually and categorized into tertiles (low, intermediate, high). Generalized estimating equations assessed longitudinal relationships.

Results:

NSES was associated with falling before adjustment (high vs low OR=1.01, 95% CI: 1.00,1.01). Walkability was significantly associated with falls after adjustment (high vs low OR=0.99, 95% CI: 0.98,0.99). Green space was not associated with falling before or after adjustment. Study arm, race and ethnicity, household income, age, low physical functioning, fall history, and climate region modified the relationship between NSES and falling. Race and ethnicity, age, fall history, and climate region modified relationships between walkability and green space and falling.

Conclusions:

Results did not show strong associations of NSES, walkability, or green space with falling. Future research should incorporate granular environmental measures that may directly relate to physical activity and outdoor engagement.

INTRODUCTION

Approximately 25% of older adults in the United States (U.S.) experience at least one fall annually, and this is projected to increase significantly by 2030 (Older Adult Falls, n.d.; World Health Organization, 2013). It is estimated that 20% of falls result in serious injury, approximately 3 million older adults annually are hospitalized because of falls, and 27,000 annual falls result in death (Bergen et al., 2016; Center for Disease Control, 2020). Importantly, about 70.5% of falls in older populations occur among women (Stevens, 2005).

Environments are crucial for facilitating or inhibiting physical activity (PA) with age (Michael & Yen, 2014; van Hoof et al., 2021), and this has implications for fall prevention (Curl et al., 2020; Li et al., 2006). Although most falls occur outdoors (Bergland et al., 2003; Li et al., 2006; Rubenstein, 2006), few interventions to prevent falls among older adults incorporate outdoor factors (Ganz & Latham, 2020). This is surprising, because outdoor falls are more likely to have environmental causes compared to indoor falls, which are often related to poor health (Bergland et al., 2003; Li et al., 2006). To inform intervention efforts, research is needed on outdoor factors impacting falling in older adults.

Few prospective studies have examined falling and outdoor environments to establish a longitudinal relationship (Lo et al., 2016; Nicklett et al., 2017; Talbot et al., 2005). One study found that perceived neighborhood safety and the absence of trash resulted in reduced falling odds (Nicklett et al., 2017). Additional research reported that falls mostly occurred while walking and were caused by balance or gait impairment and environmental causes, but most participants could not recall specific environmental causes (Talbot et al., 2005). Participant reports of falling causes and context suffered from poor recall, indicating the need to use objective measures (Cummings et al., 1988; Talbot et al., 2005). Lastly, research has identified a relationship between incident falls and objectively-measured neighborhood disadvantage, including prevalence of poverty and female-headed households (Lo et al., 2016). Neighborhood disadvantage, or low neighborhood socioeconomic status (NSES), is normally comprised of indicators of income, occupation, and education (Pickett, 2001). Research establishing its relationship with falling only examined income-based measures (Lo et al., 2016). Research has also mostly focused on hazardous outdoor fall risk factors (e.g., low NSES) and has overlooked environmental factors that may protect against falling and are associated with NSES and outdoor activity, such as walkability and green space (Adkins et al., 2017; Dalton et al., 2016; Edwards & Dulai, 2018; King & Clarke, 2015; Wen et al., 2013).

Walkability, defined as how supportive for walking a space is, is associated with PA among older adults (Chudyk et al., 2017; Orstad et al., 2018). Walkability may have an impact on falling because walking is the most reported exercise in older adults (Y.-S. Lee, 2005), and most falls occur during outdoor activity (Bath & Morgan, 1999; Bergland et al., 1998, 2003). Neighborhood walkability is often conceptualized using “3 Ds”: density, diversity, and design (Cervero & Kockelman, 1997; Smith et al., 2008). Population density is associated with increased outdoor activity (Glazier et al., 2014; Rodríguez et al., 2009; Tanishita & van Wee, 2017), diversity of land use is associated with walking-friendly and economically and socially vibrant communities (Fina, 2016; Noordzij et al., 2021; Rodríguez et al., 2009; Yue et al., 2017), and design (or urban layouts that reduce exposure to hazards) is related to outdoor activity among older adults (Ewing & Handy, 2009; H.-S. Lee et al., 2011; Van Cauwenberg et al., 2012; Wilmut & Purcell, 2021). Past research shows mixed relationships between perceived walkability and falling (Merom et al., 2015). As well, one prior study demonstrated that objectively measured residential density was negatively associated with falling (S. Lee et al., 2017), but research is needed to establish the relationship between objectively measured overall walkability and falling.

Green space, defined as areas partially or fully covered by vegetation, has a positive association with older adult health (A. C. K. Lee & Maheswaran, 2011; Zandieh et al., 2019). Green space is related to area-level SES (Hoffimann et al., 2017; Schüle et al., 2019) and promotes outdoor activity (Besser & Mitsova, 2021; Dalton et al., 2016; Zandieh et al., 2019) and social interaction (Maas et al., 2009). While green space impacts health and well-being with age, research is needed to understand if it is related to falling because higher levels of green space are often located in less walkable neighborhoods (Shuvo et al., 2021) and green spaces are viewed as destinations (Finlay et al., 2015; Zandieh et al., 2019). Research is needed to understand if green space has an independent relationship with falling risk.

Limited research has examined the modifying role of factors associated with falling, including hormone or dietary intervention, race, ethnicity, income, walking, age, physical functioning, fall history, climate region, or urban-rural residence in the relationship between outdoor environments and falling. To start, falling may be less pronounced among those experiencing hormone therapy or dietary modifications (Larocque et al., 2015; Randell et al., 2001). Next, nationally representative research has demonstrated that falling incidence is higher among Non-Hispanic White populations compared to other racial or ethnic populations (Han et al., 2014), and Non-Hispanic White people are more likely to live in higher NSES, walkable, and greener neighborhoods (Conderino et al., 2021; Osypuk et al., 2009). Low-income individuals have higher fall risk (Bergen et al., 2016), and the negative impact of low NSES on many health outcomes may be modified by income status (do Nascimento et al., 2022; Miller et al., 2020). Regular walking is associated with lower serious fall risk (Gregg et al., 2000), and individuals who walk often spend more time outdoors (Kerr et al., 2012). For age, older adults have higher fall risk and spend less time outdoors compared to younger counterparts (Freeman et al., 2019; Morrison et al., 2016). Those with functional limitations are more likely to fall and spend less time outdoors (Mangani et al., 2008; Seo et al., 2021). Those with a history of falls have a higher fall risk (Rubenstein, 2006) and fear of falling (S. Lee et al., 2018). Physical health/functioning and fall history may also impact perceptions of and engagement with outdoor environments (Merom et al., 2015). Climate-related hazards may increase falling risk, and those experiencing weather-related barriers may have less outdoor engagement and experience increased hazards (Clarke et al., 2015; Mondor et al., 2015). Rural residents also show higher fall risk (Moreland et al., 2020), and urban vs. rural neighborhoods have characteristically different socioeconomic composition, walkability, and green space (Berry et al., 2017; Holt, 2007; Xu & Wang, 2015).

Relationships between outdoor environmental indicators that positively impact older adult health, such as walkability and green space, and falling are understudied. Also, research on the relationship between NSES and falling has not included a comprehensive NSES measure or a national sample. To inform prevention efforts, it is crucial to identify subgroups whose falling risk is most impacted by NSES, walkability, and green space. Therefore, we propose two specific aims: (1) to estimate associations between neighborhood socioeconomic status (NSES), walkability, green space, and incident falls among postmenopausal women, and (2) to evaluate modifiers of these associations, including study arm participation, race, ethnicity, baseline household income, baseline walking, age at enrollment, baseline low physical functioning, baseline fall history, climate region, and urban-rural residence.

METHODS

Study Setting and Sample

This study used data from the Women’s Health Initiative (WHI) randomized controlled trial and prospective observational study. As is described elsewhere (Anderson et al., 2003), women were recruited at 40 U.S. clinical centers, 1993–1998. Clinical centers were in diverse locations throughout the United States, with 12 centers in the West region, 9 in the Midwest region, 8 in the Northeast region, and 11 in the South region (Women’s Health Initiative, n.d.-a). Eligible women were: aged 50–79 years, postmenopausal, without a medical condition associated with a predicted survival <3 years, and planning to reside in the same geographic area for at least 3 years (Hays et al., 2003; The Women’s Health Initiative Study Group, 1998). Racial and ethnic groups were enrolled proportionately to the target population (Hays et al., 2003). Self-report of falls was discontinued during the WHI extension phase, so follow-up was until 2005 (Women’s Health Initiative, n.d.-b). Data included women from both the WHI Observational Study (OS; n =93,676) and Clinical Trials (CT; n = 68,132). Women reporting a history of hip fracture or high limitations with walking a block or more at baseline were excluded, yielding a final sample of 157,583 participants (Braithwaite et al., 2003).

Participant Measures

Participant-level measures were obtained using annual questionnaires (Anderson et al., 2003; Hays et al., 2003; The Women’s Health Initiative Study Group, 1998). For fall assessments, women were asked annually how many times they fell onto the floor/ground (excluding falls due to sport activities, without reporting whether falls occurred outdoors) (Cauley et al., 2017). Count of falls was changed into a binary outcome variable (no, any falls). Multiple and single fallers differ in PA or functional status (Bjerk et al., 2018; Cimilli Ozturk et al., 2017; Dai et al., 2018), so walking levels and physical limitations were treated as effect modifiers. Sensitivity analyses in Appendix A confirm that differences between multiple and single falls were not concealed in our main analyses.

Study arm participation (OS; CT) was assigned at enrollment. WHI CT participants were randomized in one or more of three randomized controlled trials: dietary modification, hormone therapy, and calcium/vitamin D supplementation (described elsewhere) (Anderson et al., 2003). Age (in years), household income level (≤$34,999, $35,000-$74,999, ≥$75,0000, missing), race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, American Indian/Alaska Native, Asian (ancestry was Chinese, Indo-Chinese, Korean, Japanese, Pacific Islander, or Vietnamese), and other) (Garcia et al., 2022), and marital status (presently married, divorced/separated, widowed, never married, marriage-like relationship) were self-reported at enrollment (Hays et al., 2003). Walking outside for at least 10 minutes was reported in minutes, and converted to metabolic equivalent of task-hours per week (MET-hr/wk) (LaMonte et al., 2017, 2019), then grouped into tertiles. Physical functioning limitations were self-reported at baseline using indicators of physical function in moderate/vigorous activity; strength to lift, carry, stoop, bend, climb; walking various distances; with 3-point Likert scale response options (Fugate Woods et al., 2005). Scores were out of 100 and scores below 75 met criteria for low physical functioning (Fugate Woods et al., 2005). Falling history at baseline was self-reported for the previous year, excluding falls due to sports (Anderson et al., 2003; Bea et al., 2017). Geocoded participant address-specific Rural-Urban Commuting Area (RUCA) codes were used to categorize residence as urban, large rural city/town, small rural town, isolated small rural town, or missing (United States Department of Agriculture, 2020). RUCA data source years were 1990 and 2000. Geographic procedures relied on geocoded address point (coordinate) in polygon joins using ArcGIS 10.2.2 software (ESRI, n.d.). Climate region includes four regions (Northeast, Midwest, West, South), divided into climactically similar regions using latitude (lower=<35 degrees; middle=35–40 degrees, upper=>40 degrees): Northeast, South-lower, South-upper, Midwest-lower, Midwest-upper, West-lower, West-middle, West-upper.

Neighborhood Measures

Participants reported residential addresses at baseline and throughout follow-up, and these were geocoded (Whitsel, 2004; Whitsel et al., 2006). NSES included dimensions of income, wealth, education, and occupation (Diez Roux et al., 2001). Income was measured using median household income; percentage of households with interest, dividends/rent income; and median value of owner-occupied housing. Education was the percentage of adults with high school and college degrees. Occupation was the percentage of adults with professional, managerial, or executive occupation. Census tract-level data for all measures were collected from the U.S. Census of Population and Housing for 2000 and American Community Survey for 2005–2009 using five-year estimates and merged with participant-level data using SAS 9.4 software. NSES was calculated by natural log-transforming the median household income and median value of housing units, z-transforming all six measures, and summing them to create a summary z-score, separately for OS and CT participants annually (Diez Roux et al., 2004; Krieger, 2002; Rossi & Gilmartin, 1980). Z-scores were categorized into tertiles (low, intermediate, high).

Neighborhood walkability was measured based on the “3 D’s” (i.e., density, diversity, and design) associated with walking (Cervero & Kockelman, 1997). Density was measured using U.S. Census of Population and Housing attributes data from 2000 (U.S. Census Bureau., 2020), as the ratio of population count to land area (i.e., persons per kilometer²) within a geocoded participant address-centric Euclidean buffer of 5.0 miles (calculated using ArcGIS 10.2.2 software). Diversity was measured using Euclidean areal measures of exposure to developed land and cross-walked over National Land Cover Databases (NLCD) (National Geospatial Program, n.d.; United States Geological Survey, n.d.). NLCD data were used from 1992, 2001, and 2006 to create 500-meter geocoded participant address-centric Euclidean buffers of land cover types (i.e., water, developed, barren, forest, shrubland, herbaceous, planted/cultivated, wetlands) using ArcGIS 10.2.2 software (Multi-Resolution Land Characteristics (MRLC) Consortium, n.d.). The Herfindahl–Hirschman Index (HHI) was used to assess diversity of land cover (Jurgis, 2016). Design (i.e., exposure to high traffic roadways) was based on ESRI mapped and Census feature class coded A1-A3 roadways (annual Street Map data 1993–2000, 2003, ESRI & Maps data 2001–2002, 2004–2005) (TIGER Census Feature Class Codes, n.d.). A1-A3 roadways include interstate/toll highways with interchanges, national and regional highways without limited access, and state highways, and some county highways that connect smaller towns, subdivisions, and neighborhoods (TIGER Census Feature Class Codes, n.d.). Using ArcGIS 10.2.2 software, geocoded participant address-centric Euclidean buffers of 500 meters were employed to identify the total length in meters of A1-A3 roads within the empirically supported distances (Noordzij et al., 2020; Portegijs et al., 2017; Timmermans et al., 2021). Road lengths were summed to estimate total meters of A1-A3 roadways proximal to participant addresses annually (1993–2005). Walkability components were standardized separately (using PROC STANDARD, SAS 9.4 software), then averaged to create composite scores that were categorized into tertiles (low, intermediate, and high).

Neighborhood green space was ascertained using Euclidean areal measures of exposure to trees/vegetation that were cross-walked over NLCD (Multi-Resolution Land Characteristics Consortium, n.d.; United States Geological Survey, n.d.). Similar to the diversity calculation, NLCD data (from 1992, 2001, and 2006) were used to create 500-meter geocoded participant address-centric Euclidean buffers of land cover types (i.e., water, developed, barren, forest, shrubland, herbaceous, planted/cultivated, wetlands) using ArcGIS 10.2.2 software (Multi-Resolution Land Characteristics (MRLC) Consortium, n.d.). As in past green space research, fractions of the total buffer area (in meters²) occupied by each land cover type determined the areal percentage of vegetative land types, including forest, shrubland, planted/cultivated, and herbaceous grasslands/wetlands (Dalton et al., 2016; de Vries et al., 2003; Maas et al., 2006). Estimates were standardized (using PROC STANDARD, SAS 9.4 software) and converted into tertiles (low, intermediate, and high).

Data on NSES, walkability, and green space were merged with annual fall data. For years with missing data on environmental exposures, it was assumed that the environmental exposure did not change since the most recent data collection time point with available data and data from that year was used. Large buffer sizes were used to minimize missingness for walkability and green space exposures, so sensitivity analyses were completed using smaller (i.e., 100-meter buffer for green space, diversity of land cover, and exposure to A1-A3 highways and .75-mile buffer for population density) buffers to justify our approach (Appendix B).

Statistical Analyses

Longitudinal relationships between NSES, walkability, green space, and fall incidence (0/any falls) were assessed using generalized estimating equations (GEE) with a specified binary outcome distribution and a logit link function (Hubbard et al., 2010). Models were presented before and after adjustment and included a repeated statement with independent covariance to account for within participant correlations. Quasi-likelihood-based criteria values for the models with an independence correlation structure and with an exchangeable working correlation structure were compared to ensure independent correlation structure was appropriate for this data. Adjustment variables, chosen a priori, included age at enrollment, race and ethnicity, climate region, urban-rural residence, baseline household income, and baseline marital status.

Interaction terms (also chosen a priori) were assessed separately for the relationship between each neighborhood exposure and study arm participation, race and ethnicity, baseline household income, baseline walking, age at enrollment, baseline low physical functioning, baseline fall history, climate region, and urban-rural residence. If effect modification was found based on product terms (α=0.005, Bonferroni-corrected threshold), stratified models were presented. Analyses were completed using SAS 9.4 software.

RESULTS

Most participants had no falling history (64.9%), were 60–69 years of age (44.9%), identified as Non-Hispanic White (82.7%), were married (60.7%), reported no walking outside for at least 10 minutes at a time (30.5%), and did not report low physical functioning (69.4%) at baseline (Table 1). Most participants lived in the Northeast (22.8%) and in urban locations (82.0%) (Table 1).

Table 1.

Demographic and physiologic characteristics and fall outcomes at baseline (N=157,583).

Variable	Overall	Two or more falls	One fall	No falls	Missing fall data
	# (percent)	# (percent)	# (percent)	# (percent)	# (percent)
Total	157,583 (100)	18,552 (11.8)	30, 252 (19.2)	102,294 (64.9)	6,485 (4.1)
Socioeconomic status
Missing	96 (0.1)	12 (12.5)	19 (19.8)	65 (67.7)	0 (0)
Low	52,448 (33.3)	6,653 (12.7)	9,808 (18.7)	33,769 (64.4)	2,218 (4.2)
Intermediate	52,543 (33.3)	6,034 (11.5)	10,091 (19.2)	34.250 (65.2)	2,168 (4.1)
High	52,496 (33.3)	5,853 (11.1)	10,3334 (19.7)	34,210 (65.2)	2,099 (4.0)
Walkability
Missing	87 (0.1)	11 (12.6)	16 (18.4)	60 (69.0)	0 (0)
Low	52,483 (33.3)	6,221 (11.8)	10,070 (19.2)	33,855 (64.5)	2,337 (4.5)
Intermediate	52,478 (33.3)	6,095 (11.6)	10,175 (19.4)	34,215 (65.2)	1,993 (3.8)
High	52,535 (33.3)	6,225 (11.9)	9,991 (19.0)	34,164 (65.0)	2,155 (4.1)
Green space
Missing	3308 (2.1)	265 (8.0)	525 (15.9)	2,515 (76.0)	3 (0.1)
Low	51,547 (32.7)	6,157 (11.9)	9,996 (19.4)	33,898 (65.8)	1,496 (2.9)
Intermediate	50,906 (32.3)	5,770 (11.3)	9,813 (19.3)	32,927 (64.7)	2,396 (4.7)
High	51,822 (32.9)	6,360 (12.3)	9,918 (19.1)	32,954 (63.6)	2,590 (5.0)
Study arm participation
Dietary clinical trial arm	9,659 (6.1)	1,011 (10.5)	1,769 (18.3)	5,996 (62.1)	883 (9.1)
Hormone therapy clinical trial arm (Estrogen + Progesterone)	2,811 (1.8)	313 (11.1)	550 (19.6)	1,797 (63.9)	151 (5.4)
Hormone therapy clinical trial arm (Estrogen alone)	2,001 (1.3)	254 (12.7)	400 (20.0)	1,188 (59.4)	159 (8.0)
Calcium/Vitamin D clinical trial arm	35,542 (22.5)	4,202 (11.8)	6,655 (18.7)	21,959 (61.8)	2,726 (7.7)
Clinical trial control group	16,528 (10.5)	1,829 (11.1)	3,019 (18.3)	10,270 (62.1)	1,410 (8.5)
Observational study arm	91,042 (57.8)	10,943 (12.0)	17,859 (19.6)	61,084 (67.1)	1,156 (1.3)
Age
50–59	52,614 (33.4)	6,486 (12.3)	9,937 (18.9)	33,443 (63.6)	2,748 (5.2)
60–69	70,835 (44.9)	8,066 (11.4)	13,509 (19.1)	46,600 (65.8)	2,660 (3.8)
70–79+	34,134 (21.7)	4,000 (11.7)	6,806 (19.9)	22,251 (65.2)	1,077 (3.2)
Baseline household income
Missing	10,486 (6.7)	1,202 (6.5)	1,896 (6.3)	7,003 (6.8)	385 (5.9)
≤$34,999	59,711 (37.9)	7,470 (40.3)	11,410 (37.7)	38,108 (37.3)	2,723 (42.0)
$35,000 – $74,999	59,724 (37.9)	6,745 (36.4)	11,604 (38.4)	38,883 (38.0)	2,492 (38.4)
≥$75,000	27,662 (17.5)	3,135 (16.9)	5,342 (17.7)	18,300 (17.9)	885 (13.7)
Race and ethnicity
Missing	400 (0.3)	45 (11.3)	70 (17.5)	267 (66.7)	18 (4.5)
Non-Hispanic White	130,377 (82.7)	15,604 (12.0)	25,593 (19.6)	83,700 (64.2)	5,480 (4.2)
Black or African American	13,958 (8.9)	1,511 (10.8)	2,392 (17.2)	9,562 (68.5)	493 (3.5)
Hispanic/Latino	6,258 (4.0)	755 (12.1)	1,111 (17.7)	4,022 (64.3)	370 (5.9)
Asian or Pacific Islander	4,121 (2.6)	298 (7.2)	659 (16.0)	3,115 (75.6)	49 (1.2)
American Indian or Alaskan Native	673 (0.4)	101 (15.0)	142 (21.1)	409 (60.8)	21 (3.1)
Other	1,796 (1.1)	238 (13.2)	285 (15.9)	1,219 (67.9)	54 (3.0)
Baseline marital status
Missing	741 (0.5)	85 (11.5)	139 (18.8)	481 (64.9)	36 (4.9)
Presently married	95,631 (60.7)	10,540 (11.0)	17,947 (18.8)	62,946 (65.8)	4,198 (4.4)
Divorced or separated	25,003 (15.9)	3,345 (13.4)	4,912 (19.7)	15,835 (63.3)	911 (3.6)
Widowed	26,770 (17.0)	3,312 (12.4)	5,351 (20.0)	17,091 (63.8)	1,016 (3.8)
Never married	6,879 (4.3)	948 (13.8)	1,403 (20.4)	4,290 (62.4)	238 (3.5)
Marriage-like relationship	2,559 (1.6)	322 (12.6)	500 (19.5)	1,651 (64.5)	86 (3.4)
Climate region
Northeast	35,922 (22.8)	4,177 (11.6)	6,662 (18.6)	23,425 (65.2)	1,658 (4.6)
South – lower	26,251 (16.7)	2,872 (10.9)	4,802 (18.3)	17,355 (66.1)	1,222 (4.7)
South – upper	14,442 (9.2)	1,732 (12.0)	2,791 (19.3)	9,406 (65.1)	513 (3.6)
Midwest – lower	7,272 (4.6)	1,023 (14.1)	1,526 (21.0)	4,700 (64.6)	23 (0.3)
Midwest – upper	27,460 (17.4)	3,387 (12.3)	5,717 (20.8)	17,171 (62.5)	1,185 (4.3)
West – lower	23,535 (14.9)	2,470 (10.5)	4,258 (18.1)	15,943 (67.7)	864 (3.7)
West – middle	15,596 (9.9)	1,946 (12.5)	3,086 (19.8)	9,952 (63.8)	612 (3.9)
West – upper	7,105 (4.5)	945 (13.3)	1,410 (19.9)	4,342 (61.1)	408 (5.7)
Baseline walking levels (MET- hours/wk)
Missing	7348 (4.7)	625 (3.4)	881 (2.9)	262 (0.3)	5,580 (86.0)
None (0 MET-hr/week)	48,212 (30.5)	5,741 (30.9)	9,274 (30.7)	32,892 (32.2)	350 (4.7)
Low (0–6.8 MET-hr/week)	30,650 (19.5)	3,934 (21.2)	6,137 (20.3)	20,406 (19.9)	173 (2.7)
Intermediate (6.8–16.7 MET-hr/week)	39,577 (25.1)	4,621 (24.9)	7,868 (26.0)	26,854 (26.3)	234 (3.6)
High (16.7+ MET-hr/week)	31,796 (20.2)	3,631 (19.6)	6,092 (20.1)	21,880 (21.3)	193 (3.0)
Baseline low physical functioning
Missing	3,003 (1.9)	361 (12.0)	526 (17.5)	1,822 (60.7)	294 (9.8)
Yes	45,281 (28.7)	7,098 (15.7)	9,301 (20.5)	27,185 (60.0)	1,697 (3.8)
No	109,299 (69.4)	11,093 (10.1)	20,425 (18.7)	73,287 (67.1)	4,494 (4.1)
Urban-rural residence
Missing	18,480 (11.7)	2,184 (11.8)	3,567 (19.3)	12,396 (67.1)	333 (1.8)
Urban	129,177 (82.0)	14,967 (11.6)	24,670 (19.1)	83,817 (64.9)	5,723 (4.4)
Large rural city/town	5,570 (3.5)	724 (13.0)	1,166 (20.9)	3,441 (61.8)	239 (4.3)
Small rural town	2,388 (1.5)	344 (14.4)	461 (19.3)	1,491 (62.4)	92 (3.9)
Isolated small rural town	1,968 (1.3)	333 (16.9)	388 (19.7)	1,149 (58.4)	98 (5.0)

Table 2 describes the longitudinal relationships between NSES, walkability, and green space and falling incidence. NSES was significantly related to falling before adjustment (intermediate vs low OR=1.00, 95% CI: 1.00–1.01; high vs low OR=1.01, 95% CI: 1.00–1.01) (model 1a; Table 2). NSES did not have a significant relationship with falling after adjustment for covariates, including age, urban-rural residence, race and ethnicity, marital status, climate region, and household income (model 1b; Table 2). Adjusted models showed similar estimates when using a multinomial instead of binary outcome (Appendix A). Of the interaction terms tested, study arm participation (hormone therapy CT p<0.0001), race and ethnicity (p<.0001), age (p<0.0001), baseline household income (p=0.0022), fall history (p<0.0001), low physical functioning (p=0.0033), and climate region (p<0.0001) were significant. Therefore, stratified results were presented according to these modifiers in Table 3.

Table 2.

Generalized estimating equations regression analysis of neighborhood and demographic characteristics on fall incidence

	Number of Falls (0 vs. 1+ times)
	OR	95% CI	p-value
Model - NSES
Model 1a (Unadjusted)1
Intermediate NSES	1.00	(1.00, 1.01)	0.2870
High NSES	1.01	(1.00, 1.01)	0.0292
Model 1b (model 1a + confounders)1,2
Intermediate SES	1.00	(1.00, 1.01)	0.4515
High SES	1.00	(1.00, 1.01)	0.0919
Model - Walkability
Model 2a (Unadjusted)3
Intermediate walkability	1.00	(1.00, 1.01)	0.5678
High walkability	0.99	(0.99, 1.00)	0.0866
Model 2b (model 2a + confounders)2,3
Intermediate walkability	1.00	(1.00, 1.01)	0.2047
High walkability	0.99	(0.98, 0.99)	0.0062
Model - Green space
Model 3a (Unadjusted)4
Intermediate green space	1.00	(1.00, 1.01)	0.8069
High green space	1.00	(1.00, 1.01)	0.3685
Model 3b (model 3a+ confounders)2,4
Intermediate green space	1.00	(0.99, 1.00)	0.9790
High green space	1.00	(0.99, 1.01)	0.6155

OR, odds ratio; NSES, neighborhood socioeconomic status

Bolded values represent significance with p<0.05, and p<.005 for interaction terms (to account for Bonferroni correction).

P-values for interaction terms represent the joint test of association. The joint test for an effect is a test that all of the parameters associated with the that effect are zero.

¹Referent category: Low SES

²Confounders: age, urban-rural residence, race and ethnicity, climate region, baseline marital status, and baseline household income.

³Referent category: Low walkability

⁴Referent category: Low green space

Table 3.

Stratified generalized estimating equations regression analysis of neighborhood and demographic characteristics on falling incidence

	Number of Falls (0 vs. 1+ times)
	OR	95% CI	p-value
Model - Neighborhood NSES
Stratified by study arm participation
Observational study arm1,2
Intermediate NSES	1.00	(0.99, 1.01)	0.8677
High NSES	1.00	(1.00, 1.01)	0.3744
Hormone therapy clinical trial arm – Estrogen alone intervention1,2
Intermediate NSES	0.98	(0.96, 1.01)	0.1536
High NSES	0.99	(0.97, 1.01)	0.3487
Hormone therapy clinical trial arm – Estrogen alone control1,2
Intermediate NSES	1.01	(0.96, 1.05)	0.3923
High NSES	1.00	(0.98. 1.02)	0.9736
Hormone therapy clinical trial arm – Estrogen+ Progesterone intervention1,2
Intermediate NSES	1.00	(0.98, 1.02)	0.8149
High NSES	1.01	(0.99, 1.03)	0.3861
Hormone therapy clinical trial arm – Estrogen+ Progesterone control1,2
Intermediate NSES	1.01	(0.99, 1.03)	0.3426
High NSES	1.00	(0.98, 1.02)	0.8192
Dietary clinical trial arm - intervention1,2
Intermediate NSES	1.00	(0.99, 1.02)	0.5401
High NSES	1.00	(0.99, 1.02)	0.4283
Dietary clinical trial arm - control1,2
Intermediate NSES	1.00	(0.99, 1.01)	0.4866
High NSES	1.00	(0.99, 1.01)	0.7891
Calcium/Vitamin D clinical arm – intervention1,2
Intermediate NSES	1.01	(1.00, 1.01)	0.1792
High NSES	1.01	(1.00, 1.02)	0.0589
Calcium/Vitamin D clinical arm – control1,2
Intermediate NSES	1.00	(0.99, 1.01)	0.9813
High NSES	1.00	(1.00, 1.01)	0.5934
Stratified by race and ethnicity
Non-Hispanic White 1,2
Intermediate NSES	1.00	(0.99, 1.00)	0.3579
High NSES	1.00	(0.99, 1.00)	0.9892
Black or African American1,2
Intermediate NSES	1.02	(1.00, 1.04)	0.0819
High NSES	1.00	(0.98, 1.02)	0.9247
Hispanic/Latino 1,2
Intermediate NSES	1.01	(0.99, 1.04)	0.3386
High NSES	1.01	(0.98, 1.04)	0.5403
Asian or Pacific Islander 1,2
Intermediate NSES	0.98	(0.94, 1.02)	0.3508
High NSES	0.99	(0.95, 1.03)	0.6558
American Indian or Alaskan Native 1,2
Intermediate NSES	0.99	(0.93, 1.05)	0.6704
High NSES	0.94	(0.86, 1.02)	0.1518
Other1,2
Intermediate NSES	1.01	(0.96, 1.06)	0.7365
High NSES	0.99	(0.95, 1.04)	0.8071
Stratified by baseline household income
<$34,9991,2
Intermediate NSES	1.00	(1.00, 1.01)	0.4591
High NSES	1.01	(1.00, 1.02)	0.0191
$35,000 – $74,9991,2
Intermediate NSES	1.01	(1.00, 1.01)	0.1273
High NSES	1.01	(1.00, 1.01)	0.0779
>$75,0001,2
Intermediate NSES	1.00	(0.98, 1.01)	0.8007
High NSES	1.01	(0.99, 1.02)	0.3318
Stratified by age at enrollment
50–59 years1,2
Intermediate NSES	1.00	(0.99, 1.01)	0.7581
High NSES	1.01	(1.00, 1.02)	0.0126
60–69 years1,2
Intermediate NSES	1.01	(1.00, 1.01)	0.0980
High NSES	1.00	(0.99, 1.01)	0.7187
70–79+ years1,2
Intermediate NSES	1.00	(0.99, 1.01)	0.7516
High NSES	1.00	(0.99, 1.01)	0.6957
Stratified by baseline low physical functioning
Low physical functioning1,2
Intermediate NSES	1.01	(1.00, 1.01)	0.1785
High NSES	1.00	(0.99, 1.01)	0.4737
Normal physical functioning1,2
Intermediate NSES	1.00	(1.00, 1.01)	0.5831
High NSES	1.01	(1.00, 1.01)	0.0303
Stratified by baseline fall history
None1,2
Intermediate NSES	1.00	(0.99, 1.00)	0.3252
High NSES	1.00	(1.00, 1.01)	0.6642
One time1,2
Intermediate NSES	1.01	(1.00, 1.02)	0.2776
High NSES	1.01	(1.00, 1.02)	0.1073
Two times1,2
Intermediate NSES	1.01	(1.00, 1.02)	0.1601
High NSES	1.00	(0.99, 1.01)	0.9003
Three or more times1,2
Intermediate NSES	1.01	(0.99, 1.03)	0.2997
High NSES	0.99	(0.98, 1.01)	0.6017
Stratified by climate region
Northeast1,2
Intermediate NSES	1.00	(0.99, 1.01)	0.8371
High NSES	1.00	(0.99, 1.00)	0.3242
South – lower1,2
Intermediate NSES	1.00	(0.99, 1.02)	0.4745
High NSES	1.00	(0.99, 1.01)	0.5630
South – upper1,2
Intermediate NSES	1.01	(0.99, 1.03)	0.2517
High NSES	1.03	(1.02, 1.05)	<.0001
Midwest – lower1,2
Intermediate NSES	1.01	(0.99, 1.03)	0.3964
High NSES	1.02	(1.00, 1.04)	0.0668
Midwest – upper1,2
Intermediate NSES	1.00	(0.99, 1.01)	0.6609
High NSES	1.01	(1.00, 1.02)	0.1587
West – lower1,2
Intermediate NSES	1.00	(0.99, 1.02)	0.5693
High NSES	0.99	(0.98, 1.01)	0.2990
West – middle1,2
Intermediate NSES	0.99	(0.98, 1.01)	0.2437
High NSES	0.99	(0.98, 1.00)	0.2062
West – upper1,2
Intermediate NSES	0.99	(0.97, 1.01)	0.4516
High NSES	1.00	(0.98, 1.03)	0.6579
Model - Walkability
Stratified by study arm participation
Observational study arm2,3
Intermediate walkability	1.01	(1.00, 1.01)	0.1144
High walkability	1.00	(1.00, 1.01)	0.5000
Hormone therapy clinical trial arm – Estrogen alone intervention2,3
Intermediate walkability	1.00	(0.97, 1.02)	0.7226
High walkability	0.98	(0.96, 1.01)	0.2230
Hormone therapy clinical trial arm – Estrogen alone control2,3
Intermediate walkability	0.98	(0.96, 1.01)	0.1552
High walkability	0.99	(0.97, 1.02)	0.6416
Hormone therapy clinical trial arm – Estrogen+ Progesterone intervention2,3
Intermediate walkability	1.01	(0.99, 1.03)	0.2965
High walkability	1.00	(0.98, 1.02)	0.9386
Hormone therapy clinical trial arm – Estrogen+ Progesterone control2,3
Intermediate walkability	0.99	(0.97, 1.01)	0.2121
High walkability	0.99	(0.97, 1.01)	0.3355
Dietary clinical trial arm - intervention3,4
Intermediate walkability	1.01	(0.99, 1.02)	0.2441
High walkability	0.99	(0.98, 1.01)	0.3581
Dietary clinical trial arm - control2,3
Intermediate walkability	1.00	(0.99, 1.01)	0.3353
High walkability	0.99	(0.98, 1.00)	0.0445
Calcium/Vitamin D clinical arm – intervention2,3
Intermediate walkability	0.99	(0.98, 1.00)	0.1240
High walkability	1.01	(0.98, 1.00)	0.0591
Calcium/Vitamin D clinical arm – control2,3
Intermediate walkability	1.01	(1.00, 1.01)	0.0115
High walkability	1.00	(0.99, 1.00)	0.2057
Stratified by race and ethnicity
Non-Hispanic White2,3
Intermediate walkability	1.00	(1.00, 1.01)	0.1473
High walkability	0.99	(0.99, 1.00)	0.0083
Black or African American2,3
Intermediate walkability	0.99	(0.98, 1.01)	0.4052
High walkability	1.00	(0.98, 1.02)	0.9596
Hispanic/Latino2,3
Intermediate walkability	1.00	(0.98, 1.03)	0.7508
High walkability	0.99	(0.97, 1.02)	0.6851
Asian or Pacific Islander2,3
Intermediate walkability	1.00	(0.96, 1.03)	0.8248
High walkability	0.97	(0.93, 1.00)	0.0657
American Indian or Alaskan Native2,3
Intermediate walkability	1.01	(0.94, 1.08)	0.8378
High walkability	0.98	(0.90, 1.06)	0.6289
Other2,3
Intermediate walkability	0.98	(0.93, 1.03)	0.3724
High walkability	1.00	(0.95, 1.06)	0.9154
Stratified by age at enrollment
50–59 years2,3
Intermediate walkability	1.00	(0.99, 1.01)	0.9928
High walkability	0.99	(0.98, 1.00)	0.0191
60–69 years2,3
Intermediate walkability	1.01	(0.99, 1.01)	0.1213
High walkability	1.00	(0.99, 1.01)	0.5020
70–79+ years2,3
Intermediate walkability	1.00	(0.99, 1.01)	0.3687
High walkability	1.00	(0.99, 1.01)	0.4788
Stratified by baseline fall history
None2,3
Intermediate walkability	1.00	(0.99, 1.00)	0.3044
High walkability	0.99	(0.98, 1.00)	0.0024
One time2,3
Intermediate walkability	1.00	(0.99, 1.01)	0.3642
High walkability	1.00	(0.99, 1.01)	0.9984
Two times2,3
Intermediate walkability	1.01	(1.00, 1.02)	0.1803
High walkability	0.99	(0.98, 1.01)	0.3773
Three or more times2,3
Intermediate walkability	1.02	(1.00, 1.04)	0.0597
High walkability	1.01	(0.98, 1.03)	0.6354
Stratified by climate region
Northeast2,3
Intermediate walkability	1.00	(0.99, 1.01)	0.5044
High walkability	0.99	(0.96, 1.00)	0.0947
South – lower2,3
Intermediate walkability	1.00	(0.99, 1.01)	0.7244
High walkability	1.00	(0.99, 1.01)	0.7002
South – upper2,3
Intermediate walkability	1.01	(0.99, 1.02)	0.5351
High walkability	1.00	(0.98, 1.01)	0.5679
Midwest – lower2,3
Intermediate walkability	1.01	(0.98, 1.03)	0.6617
High walkability	1.02	(0.99, 1.04)	0.2365
Midwest – upper2,3
Intermediate walkability	1.01	(1.00, 1.02)	0..2284
High walkability	0.99	(0.98, 1.00)	0.1604
West – lower2,3
Intermediate walkability	1.01	(1.00, 1.03)	0.0324
High walkability	1.00	(0.98, 1.01)	0.8630
West – middle2,3
Intermediate walkability	1.00	(0.99, 1.01)	0.9665
High walkability	0.99	(0.97, 1.00)	0.0713
West – upper2,3
Intermediate walkability	1.00	(0.98, 1.02)	0.9856
High walkability	0.99	(0.97, 1.02)	0.5795
Model - Green space
Stratified by study arm participation
Observational study arm2,4
Intermediate green space	1.00	(0.99, 1.01)	0.7281
High green space	1.00	(1.00, 1.01)	0.5187
Hormone therapy clinical trial arm – Estrogen alone intervention2,4
Intermediate green space	1.00	(0.97, 1.02)	0.8911
High green space	0.99	(0.96, 1.01)	0.2581
Hormone therapy clinical trial arm – Estrogen alone control2,4
Intermediate green space	1.01	(0.99, 1.03)	0.3687
High green space	1.00	(0.98, 1.02)	0.9352
Hormone therapy clinical trial arm – Estrogen + Progesterone intervention2,4
Intermediate green space	0.99	(0.97, 1.01)	0.4457
High green space	0.99	(0.97, 1.01)	0.3431
Hormone therapy clinical trial arm – Estrogen + Progesterone control2,4
Intermediate green space	0.98	(0.96, 1.00)	0.1179
High green space	1.00	(0.98, 1.02)	0.8337
Dietary clinical trial arm - intervention2,4
Intermediate green space	1.00	(0.98, 1.01)	0.5173
High green space	1.00	(0.98, 1.01)	0.4253
Dietary clinical trial arm - control2,4
Intermediate green space	0.99	(0.98, 1.00)	0.2319
High green space	1.00	(0.99, 1.01)	0.4273
Calcium/Vitamin D clinical arm – intervention2,4
Intermediate green space	1.00	(0.99, 1.01)	0.7143
High green space	0.99	(0.99, 1.00)	0.1548
Calcium/Vitamin D clinical arm – control2,4
Intermediate green space	1.00	(0.99, 1.00)	0.7118
High green space	1.00	(1.00, 1.01)	0.2633
Stratified by race and ethnicity
Non-Hispanic White2,4
Intermediate green space	1.00	(0.99, 1.00)	0.6441
High green space	1.00	(0.99, 1.00)	0.9893
Black or African American2,4
Intermediate green space	1.00	(0.99, 1.02)	0.7864
High green space	1.00	(0.98, 1.02)	0.9441
Hispanic/Latino2,4
Intermediate green space	1.01	(0.99, 1.04)	0.2897
High green space	0.99	(0.97, 1.02)	0.6831
Asian or Pacific Islander2,4
Intermediate green space	0.97	(0.93, 1.01)	0.1624
High green space	0.95	(0.90, 1.01)	0.1144
American Indian or Alaskan Native2,4
Intermediate green space	0.94	(0.87, 1.01)	0.0791
High green space	1.02	(0.96, 1.09)	0.5527
Other2,4
Intermediate green space	1.02	(0.97, 1.07)	0.4263
High green space	1.00	(0.95, 1.06)	0.8674
Stratified by age at enrollment
50–59 years2,4
Intermediate green space	1.01	(1.00, 1.01)	0.2196
High green space	1.00	(1.00, 1.01)	0.4048
60–69 years2,4
Intermediate green space	1.00	(0.99, 1.00)	0.1644
High green space	1.00	(0.99, 1.01)	0.9965
70–79+ years2,4
Intermediate green space	1.00	(1.00, 1.01)	0.3163
High green space	1.01	(1.00, 1.02)	0.0862
Stratified by baseline fall history
None2,4
Intermediate green space	1.00	(0.99, 1.00)	0.4461
High green space	1.00	(1.00, 1.01)	0.6694
One time2,4
Intermediate green space	1.00	(0.99, 1.01)	0.9018
High green space	1.00	(0.99, 1.01)	0.4156
Two times2,4
Intermediate green space	1.01	(0.99, 1.02)	0.4877
High green space	1.00	(0.99, 1.02)	0.6313
Three or more times2,4
Intermediate green space	1.00	(0.98, 1.02)	0.8902
High green space	1.00	(0.98, 1.02)	0.9180
Stratified by climate region
Northeast2,4
Intermediate green space	1.00	(0.99, 1.01)	0.5946
High green space	1.01	(1.00, 1.02)	0.0677
South – lower2,4
Intermediate green space	1.00	(0.99, 1.02)	0.4910
High green space	1.00	(0.98, 1.01)	0.4610
South – upper2,4
Intermediate green space	1.01	(0.99, 1.02)	0.4934
High green space	1.01	(1.00, 1.03)	0.0864
Midwest – lower2,4
Intermediate green space	1.02	(0.99, 1.04)	0.1546
High green space	1.02	(1.00, 1.04)	0.0661
Midwest – upper2,4
Intermediate green space	1.00	(0.99, 1.01)	0.7621
High green space	1.00	(0.99, 1.01)	0.8182
West – lower2,4
Intermediate green space	1.00	(0.99, 1.02)	0.6694
High green space	1.00	(0.99, 1.02)	0.6325
West – middle2,4
Intermediate green space	0.99	(0.98, 1.01)	0.2656
High green space	0.99	(0.98, 1.00)	0.0804
West – upper2,4
Intermediate green space	1.00	(0.98, 1.02)	0.9785
High green space	1.00	(0.98, 1.02)	0.7073

NSES, neighborhood socioeconomic status

Bolded values represent significance with p<.05.

¹Referent category: Low SES

²Confounders: age, urban-rural residence, race and ethnicity, climate region, baseline marital status, and baseline household income

³Referent category: Low walkability

⁴Referent category: Low green space

Stratified analyses by race and ethnicity and fall history showed that there was not a significant relationship between NSES and falling in any subgroups (Table 3). For household income, only those with income <$34,999 showed a significant relationship between NSES and falling incidence (intermediate vs low OR=1.00, 95% CI: 0.99–1.01; high vs low OR=1.01, 95% CI: 1.00–1.02) (Table 3). For age, only those 50 to 59 years of age at enrollment showed a significant relationship between NSES and falling incidence (intermediate vs low OR=1.00, 95% CI: 1.00–1.01; high vs low OR=1.01, 95% CI: 1.00–1.02) (Table 3). For physical functioning, only those not reporting functional limitations showed a significant relationship between NSES and falling incidence (intermediate vs low OR=1.00, 95% CI: 1.00–1.01; high vs low OR=1.01, 95% CI: 1.00–1.01) (Table 3). For climate region, only those living in the upper South showed a significant relationship between NSES and falling incidence (intermediate vs low OR=1.01, 95% CI: 0.99–1.03; high vs low OR=1.03, 95% CI: 1.02–1.05) (Table 3).

Walkability was not significantly related to falling before adjustment (model 2a; Table 2). Walkability had a significant relationship with falling incidence after adjustment for covariates (intermediate vs low OR=1.00, 95% CI: 1.00–1.01; high vs low OR=0.99, 95% CI: 0.98–0.99), including age, urban-rural residence, race and ethnicity, marital status, climate region, and household income status (model 2b; Table 2). Adjusted models showed similar estimates when using a multinomial instead of binary outcome (Appendix A). Of the interaction terms tested, study arm participation (hormone therapy p<0.0001), race and ethnicity (p<0.0001), age (p<0.0001), fall history (p<0.0001), and climate region (p<0.0001) were significant. Stratified results were presented in Table 3.

Stratified analyses by race and ethnicity showed that only participants identifying as Non-Hispanic White had a significant relationship between walkability and falling (intermediate vs low OR=1.00, 95% CI: 1.00–1.01; high vs low OR=0.99, 95% CI: 0.99–1.00) (Table 3). For age, only participants 50–59 years old had a significant relationship between walkability and falling (intermediate vs low OR=1.00, 95% CI: 0.99–1.01; high vs low OR=0.99, 95% CI: 0.98–1.00) (Table 3). For fall history, only participants with no fall history at baseline had a significant relationship between walkability and falling (intermediate vs low OR=1.00, 95% CI: 0.99–1.00; high vs low OR=0.99, 95% CI: 0.98–1.00) (Table 3). For climate region, only participants in the lower Western region had a significant relationship between walkability and falling (intermediate vs low OR=1.01, 95% CI: 1.00–1.03; high vs low OR=1.00, 95% CI: 0.98–1.01) (Table 3).

Green space was not significantly related to falling before adjustment (model 3a; Table 2). Green space also did not have a significant relationship with falling after adjustment for covariates, including age, urban-rural residence, race and ethnicity, marital status, climate region, and household income (model 3b; Table 2). Adjusted models showed similar estimates when using a multinomial instead of binary outcome (Appendix A). Of the interaction terms tested, study arm participation (hormone therapy p<0.0001), race and ethnicity (p<0.0001), age (p<0.0001), fall history (p<0.0001), and climate region (p<0.0001) were significant. Stratified analyses show no subgroups of study arm participation, race and ethnicity, age, fall history, or climate region had significant relationships between green space and falling (Table 3).

DISCUSSION

Summary of Findings

This study assessed outdoor environmental features related to incident falls among postmenopausal women. NSES and green space were not associated with falls in adjusted models. Walkability was significantly associated with falls, but odds ratios were close to null. Study arm, race and ethnicity, household income, age, low physical functioning, fall history, and climate region modified the relationship between NSES and falling. Race and ethnicity, age, fall history, and climate region modified relationships between walkability and green space and falling.

Comparison to Past Findings

Unadjusted results, showing higher NSES is weakly related to increased falling odds, differ from past research. One study showed a strong positive effect of housing-based SES on falls (hazard ratio= 0.58, 95% CI:0.44, 0.76) (Ryu et al., 2017). Additional research found that greater objectively-measured neighborhood disadvantage was related to greater incident falls (Lo et al., 2016). Our study did not show a relationship between NSES and falls after adjustment, meaning that age, urban-rural residence, race and ethnicity, marital status, climate region, and income level accounted for any negative impact NSES had on falling. Stratified analyses showed that there was only a significant relationship between NSES and falling among those with income <$34,999, among those 50–59 years at baseline, among those with no reported functional limitations, and among those in the upper Southern region. These findings are consistent with past research showing that those with lower (vs. higher) income levels are more impacted by NSES (Fuentes et al., 2007; Lo et al., 2016), elderly adults tend to spend less time outdoors and experience a cumulative health impact of low NSES over time (Yen et al., 2009), and those with no physical limitations spend more time outdoors (Jansen et al., 2015). No past research has examined climate region differences for the association between NSES and falling, but poverty is greater in the South (Holt, 2007).

Our study showed that women in high versus low walkability neighborhoods demonstrated decreased falling odds. This adds to a growing body of research investigating walkability and falling among older adults. One study reported that most outdoor falls were precipitated by walkability-related factors, more active individuals had higher risk for outdoor falls, and those with poorer health had greater risk for indoor falls (Li et al., 2006). Our study did not find that baseline walking and physical impairment modified the relationship between walkability and falling, but these measures did not account for recall bias or changes over time. In addition, only Non-Hispanic White participants had a significant relationship between walkability and falling. While research has not examined racial and ethnic differences for the impact of walkability on falling, emerging research shows that minoritized groups may experience intersectional barriers to outdoor engagement that go beyond environmental conditions (Roberts et al., 2019). Younger age and no fall history also increase outdoor engagement among older adults (Yen et al., 2009), which explains why these groups saw reduced odds of falling in high compared to low walkability areas.

Results demonstrated that green space was not associated with falling. Although the relationship between green space and falling risk is understudied, research shows that quality, maintenance, and safety of green space can improve accessibility for older adults (Hong et al., 2018; Tan et al., 2019). Null associations may be explained by the fact that our green space measure does not measure barriers to access.

Strengths and Limitations

An important research limitation is the conceptualization of neighborhood exposures. Walkability was measured using “diversity, density, and design,” a common method (Cervero & Kockelman, 1997). More microscale walkability measures, such as sidewalk quality and crosswalk length (Chippendale & Boltz, 2015; Nyman et al., 2013), were not integrated due to reliability and feasibility barriers. In addition, land use data were unavailable for this sample so diversity of land cover was used to capture exposure to various developed or vegetative land cover types. While land cover does not capture important land uses that may facilitate walking, land cover is also related to outdoor walking among older adults (Besser & Mitsova, 2021). Our green space indicator is also widely used in research (Dalton et al., 2016; de Vries et al., 2003; Maas et al., 2006), but measures that capture recreation and accessibility are understudied (Chong et al., 2019; Hillsdon et al., 2006). Relatedly, we conceptualized “neighborhood” data broadly to include census tract-level NSES, 5-mile buffers for population density, and 500-meter buffers for green space, diversity of land cover, and exposure to high traffic roads. While smaller buffers may better describe activity zones for older adults, our approach allowed us to accommodate a geographically diverse dataset. Body mass index was not accounted for, because it can be a poor indicator of body fat and health, especially among older women (Bhurosy, 2013; Humphreys, 2010; Nuttall, 2015), and is likely an intermediate variable in the relationship between environments and falling (Schisterman et al., 2009). Next, fall events that did not clearly meet the definition for falling or were forgotten may have gone unreported (Centre for Clinical Practice at NICE, 2013; Freiberger & de Vreede, 2011), and participants did not report circumstances or severity. Regardless, reporting falling circumstances often suffers from recall bias and outdoor environments may impact outdoor and indoor falling (Talbot et al., 2005). Lastly, person-level measures were reported at baseline and do not capture time changes.

This research also has many strengths. This study employed objective measures of NSES, walkability, and green space that are reliable across diverse locations, avoid recall bias, and can inform local infrastructure investments. This study also employs WHI data, a sample well-characterized on person-level risk factors, with large national reach, and a large, prospective sample (McGowan & Pottern, 2000). Current longitudinal research on environmental determinants of falling has not focused on large cohorts of community-dwelling older women and inadequately accounts for known person-level risk factors (Lo et al., 2016; Nicklett et al., 2017; Talbot et al., 2005). Next, sensitivity analyses confirmed that 100- and 500-meter buffer sizes for walkability and green space showed similar estimates. Conservative Bonferroni correction was used to determine the significance threshold for interaction testing, resulting in lower risk of a type 1 error (Armstrong, 2014; Emerson, 2020).

IMPLICATIONS FOR PRACTICE AND/OR POLICY

This study has important implications for research on environments that may increase fall risk, and responds to growing efforts to encourage positive health behaviors and reduce injury risk in older adults (Sherrington et al., 2020). Researchers can build on our findings by examining whether indoor and outdoor environments synergistically impact falling. Indoor environments are well studied (Valipoor et al., 2020), and our study provides input on outdoor risk factors, but no current research incorporates both to understand how aging women can avoid falls. Our findings provide a basis for researchers to expand the assessment of whether and under what circumstances indoor and outdoor environments can be modified to prevent falling. Subsequently, researchers can investigate the relationship between walkability and falling to identify both perceived and objective walkability components that are most impactful. Relatedly, research is needed on the relationship between green space and falling that considers known barriers (i.e., safety, private vs. public access) among older adults (Cronin-de-Chavez et al., 2019; Enssle & Kabisch, 2020; Hong et al., 2018). Future research can also assess whether increased walking always increases fall risk, whether environments modify this, and what prevention approaches support walking while reducing injury (Curl et al., 2020; Gillespie et al., 2003; Kumar et al., 2014). Next, researchers may examine the overall impact of neighborhoods by testing the interactive effect of NSES, walkability, and green space on falling incidence. Because this was the first study to examine the longitudinal impact of objectively-measured walkability and green space on falling, establishing individual associations was a priority. Relatedly, research may examine other interactions that were outside the scope of this research, such as the interaction between race and physical functioning (Pebley et al., 2021; Spencer et al., 2009). Lastly, researchers should examine variables that may mediate the relationships between NSES, walkability, and falling (e.g., walking levels, fall history). Understanding the mechanisms through which neighborhood environments impact falling would inform community-based prevention efforts (Curl et al., 2020).

This study has several practice-based implications. First, results can inform local fall prevention efforts creating natural, built, and social environments that promote independent aging (Plouffe & Kalache, 2010). Second, this study can inform initiatives that support older adults at high risk of falling, such as CDC’s STEADI program (Stevens & Phelan, 2013), Liveable Communities Initiative (American Association of Retired Persons, n.d.), or the Four Dimensions of Fall Prevention (World Health Organization, 2007). Programs may incorporate outdoor environment screening, improve environments through policy advocacy for infrastructure changes, or facilitate transportation to more suitable locations. Due to data availability, our study excluded more granular, microscale walkability features such as shade, benches, public transport, or sidewalk hazards that are cost-effective to improve (Alves et al., 2020). Efforts should be made to make granular audit data publicly available (Adams et al., 2022).

CONCLUSIONS

Understanding the effects of neighborhood environments on falling risk is important for promoting aging in place (Iecovich, 2014; Wang et al., 2012). Based on our findings, intervention strategies that improve objectively-measured walkability could slightly reduce fall risk among postmenopausal women. Research is needed that integrates indoor environment audits and walkability measures that more directly relate to outdoor activity and falling among older adults. This study contributes to a growing field of research on outdoor environments and falling, which can inform fall prevention initiatives targeting community-dwelling older women.

Biographies

Author Biography

ME Wende is a postdoctoral research associate in the Department of Public Health at Baylor University. Her research is focused on built, physical, and social environments and their relationship with physical activity and other health behaviors and outcomes.

M Lohman is an Assistant Professor in the Department of Epidemiology and Biostatistics at University of South Carolina. His primary area of research is in the epidemiology of adverse health outcomes, such as falls, hospitalizations and acquired disabilities among older adults.

DB Friedman is a Professor and Department Chair in the Department of Health Promotion, Education, and Behavior at University of South Carolina. Her interdisciplinary and partner-engaged research is focused on health and risk communication with older and diverse populations.

AC McLain is an Associate Professor in the Department of Epidemiology and Biostatistics at University of South Carolina. He has a broad range of statistical interests including modeling complex longitudinal and clustered data, length-biased survival analysis, and multiple testing.

MJ LaMonte is a Research Professor in the Department of Epidemiology and Environmental Health at the University of Buffalo. His research focuses on the measurement of physical activity and functional capacity, and their influence on cardiovascular disease and healthy aging.

EA Whitsel is a Professor in the Department of Epidemiology and Adjunct Professor in the Department of Medicine at the University of North Carolina. His research has focused on topics at the intersection of environmental, cardiovascular, and genetic epidemiology.

AH Shadyab is an Assistant Professor of Epidemiology at University of California, San Diego. His research interests include geroscience (study of biological aging), aging biomarkers, multi-omics (genomics, epigenetics, proteomics), healthy longevity, and Alzheimer’s disease

L Garcia is a Professor in the Department of Public Health Sciences at University of California, Davis. Her research focuses on health disparities and social determinants of health in underresourced and immigrant communities, and cardiovascular outcomes such as obesity and diabetes.

BW Chrisinger is an Associate Professor of Evidence-Based Policy Evaluation at the University of Oxford. He conducts interdisciplinary research on the relationships between place and health, especially health disparities, and the role that place-based policies can improve health equity.

K Pan works in the Department of Medical Oncology and Hematology, Downey Medical Center at Kaiser Permanente. Her specialty is hematology-oncology. Previously, she worked as an Assistant Clinical Professor and Associate Fellowship Program Director at Harbor-UCLA Medical Center.

CE Bird, an adjunct sociologist at the RAND Corporation and professor of policy analysis at the Pardee RAND Graduate School, studies women’s health and determinants of sex/gender differences in health and health care.

GE Sarto is a Professor Emeritus in the Department of Obstetrics and Gynecology at the University of Wisconsin-Madison. She plays a critical role in bringing awareness to women’s health issues, to influence public policy and national research initiatives.

AT Kaczynski is an Associate Professor in the Department of Health Promotion, Education, and Behavior at University of South Carolina. His research focuses on healthy community design and important health outcomes such as physical activity, obesity, and mental health.

Appendix A. Generalized estimating equations regression analysis of neighborhood and demographic characteristics on fall incidence, sensitivity analysis for multinomial outcome

	Number of Falls
	OR	95% CI	p-value
Model - Neighborhood NSES
Original analyses (0 vs. any falls)
Model 1a (Unadjusted)1
Intermediate NSES	1.00	(1.00, 1.01)	0.2870
High NSES	1.01	(1.00, 1.01)	0.0292
Model 1b (model 1a + confounders)1,2
Intermediate SES	1.00	(1.00, 1.01)	0.4515
High SES	1.00	(1.00, 1.01)	0.0919
0 vs. 1 fall
Model 2a (Unadjusted)1
Intermediate NSES	1.00	(0.99, 1.01)	0.6822
High NSES	1.01	(1.00, 1.01)	0.0231
Model 2b (model 2a + confounders)1,2
Intermediate NSES	1.00	(0.99, 1.01)	0.8579
High NSES	1.00	(1.00, 1.01)	0.0635
0 vs. 2+ falls
Model 3a (Unadjusted)1
Intermediate NSES	1.01	(1.00, 1.02)	0.0489
High NSES	1.01	(1.00, 1.01)	0.0488
Model 3b (model 3a + confounders)1,2
Intermediate NSES	1.01	(1.00, 1.01)	0.0872
High NSES	1.01	(1.00 1.01)	0.1026
Model - Walkability
Original analyses (0 vs. any falls)
Model 4a (Unadjusted)3
Intermediate walkability	1.00	(1.00, 1.01)	0.5678
High walkability	0.99	(0.99, 1.00)	0.0866
Model 4b (model 4a+ confounders)2,3
Intermediate walkability	1.00	(1.00, 1.01)	0.2047
High walkability	0.99	(0.98, 0.99)	0.0062
0 vs. 1 fall
Model 5a (Unadjusted)3
Intermediate walkability	1.00	(0.99, 1.01)	0.9823
High walkability	0.99	(0.99, 1.00)	0.0157
Model 5b (model 5a + confounders) 2,3
Intermediate walkability	1.00	(0.99, 1.00)	0.8806
High walkability	0.99	(0.99, 1.00)	0.0173
0 vs. 2+ falls
Model 6a (Unadjusted)3
Intermediate walkability	1.01	(1.00, 1.02)	0.0064
High walkability	0.99	(0.98, 1.00)	0.1153
Model 6b (model 6a + confounders)2,3
Intermediate walkability	1.01	(1.00, 1.02)	0.0128
High walkability	0.99	(0.98, 1.00)	0.0737
Model - Green space
Original analyses (0 vs. any falls)
Model 7a (Unadjusted)4
Intermediate green space	1.00	(1.00, 1.01)	0.8069
High green space	1.00	(1.00, 1.01)	0.3685
Model 7b (model 7a + confounders)2,4
Intermediate green space	1.00	(0.99, 1.00)	0.9790
High green space	1.00	(0.99, 1.01)	0.6155
0 vs. 1 fall
Model 8a (Unadjusted)4
Intermediate green space	1.00	(1.00, 1.01)	0.7577
High green space	1.00	(1.00, 1.01)	0.8260
Model 8b (model 8a+ confounders)2,4
Intermediate green space	1.00	(1.00, 1.01)	0.8795
High green space	1.00	(0.99, 1.00)	0.9121
0 vs. 2+ falls
Model 9a (Unadjusted)4
Intermediate green space	1.00	(0.99, 1.01)	0.7877
High green space	1.00	(1.00, 1.01)	0.1965
Model 9b (model 9a + confounders)2,4
Intermediate green space	1.00	(0.99, 1.01)	0.9329
High green space	1.00	(1.00, 1.01)	0.3042

Odds Ratio (OR)

Neighborhood Socioeconomic Status (NSES)

Bolded values represent significance with p<0.05, and p<0.10 for interaction terms.

P-values for interaction terms represent the joint tests of association. The joint tests for an effect is a test that all of the parameters associated with the that effect are zero.

¹Referent category: Low NSES

²Confounders: age, urban-rural residence, race/ethnicity, climate region, baseline marital status, and baseline household income.

³Referent category: Low walkability

⁴Referent category: Low green space

Appendix B. Generalized estimating equations regression analysis of neighborhood and demographic characteristics on fall incidence, sensitivity analysis for differing buffer sizes

	Number of Falls (0, vs. 1+ times)
	OR	95% CI	p-value
Model - Walkability
Original buffer sizes
Model 1a (Unadjusted)1
Intermediate walkability	1.00	(1.00, 1.01)	0.5678
High walkability	0.99	(0.99, 1.00)	0.0866
Model 1b (model 1a + confounders)1,2
Intermediate walkability	1.00	(1.00, 1.01)	0.2047
High walkability	0.99	(0.98, 0.99)	0.0062
Smaller buffer sizes
Model 2a (Unadjusted)1
Intermediate walkability	1.01	(1.00, 1.01)	0.0010
High walkability	0.98	(0.98, 0.99)	<.0001
Model 2b (model 2a + confounders) 1,2
Intermediate walkability	1.01	(1.00, 1.01)	0.0044
High walkability	0.99	(0.98, 0.99)	<.0001
Model - Green space
Original buffer sizes
Model 3a (Unadjusted)3
Intermediate green space	1.00	(1.00, 1.01)	0.8069
High green space	1.00	(1.00, 1.01)	0.3685
Model 3b (model 3a + confounders)2,3
Intermediate green space	1.00	(0.99, 1.00)	0.9790
High green space	1.00	(0.99, 1.01)	0.6155
Smaller buffer sizes
Model 4a (Unadjusted)3
Intermediate green space	1.00	(0.99, 1.00)	0.8290
High green space	1.00	(0.99, 1.01)	0.7874
Model 4b (model 4a+ confounders)2,3
Intermediate green space	1.00	(0.99, 1.00)	0.9755
High green space	1.00	(0.99, 1.01)	0.8979

Odds Ratio (OR)

Neighborhood Socioeconomic Status (NSES)

Bolded values represent significance with p<0.05, and p<0.10 for interaction terms.

P-values for interaction terms represent the joint tests of association. The joint tests for an effect is a test that all of the parameters associated with the that effect are zero.

¹Referent category: Low walkability

²Confounders: age, urban-rural residence, race/ethnicity, climate region, baseline marital status, and baseline household income.

³Referent category: Low green space