Background
The global obesity epidemic poses one of the greatest challenges in modern public health [
1]. Contributors to obesity are mutlifaceted and complex, with behavioural (nutrition, physical activity), metabolic, and environmental factors implicated [
2,
3]. Increased understanding of factors associated with obesity at the population level is fundamental to efficacious intervention development and implementation.
Escalating interest in the role of built environments (
BEs) as being obesogenic, or obesity promoting, has been demonstrated in the last decade [
1,
4,
5]. Emerging evidence suggests relationships between obesity and various BE attributes in adults [
2,
4,
6‐
11]; however, to what degree the BE is associated with obesity, and the mechanisms by which this occurs, are yet to be fully determined [
2]. Theories suggest direct pathways between the BE and dietary and physical activity behaviours, with subsequent impacts on body size (concurrently acknowledging individual and social environment factors that may influence these relationships) [
4,
12,
13]. Stronger and more consistent relationships have been found between BE characteristics and physical activity than for food environments [
14,
15]. For example, cross-sectional (predominantly US) research has shown clear relationships between BE features and physical activity behaviours (particularly walking) in adults [
8,
16,
17].
The magnitude of the associations between environmental features and physical activity can be considerable - earlier work in the Understanding the Relationship Between Activity and Neighbourhoods (
URBAN) study in New Zealand demonstrated moderate population-wide effects of BE features with self-reported and objectively-assessed physical activity in adults [
18]. A one standard deviation (
SD) higher destination access, street connectivity, and dwelling density was associated with increased odds of any self-reported physical activity via active transport, walking, or leisure-time activities (ranges in effect sizes from 21 to 44 %), and 7 % higher objectively-assessed (via accelerometer) physical activity levels.
Investigating these relationships poses a number of specific challenges, including inconsistencies in conceptual underpinnings, delimitation of geographical boundaries, operationalization of neighbourhoods (e.g., residential and non-residential spaces), and measures of BE features [
2,
4,
10] across studies. Within-study homogeneity in demography and BE attributes has further limited our understanding of these relationships [
4,
19]. It is also not clear how much of a contribution preferences for living in more walkable neighbourhoods and ‘self-selection’ into preferred neighbourhoods may have in explaining the observed association of BE with physical activity and body size; the extent of such confounding bias is not well quantified or understood in existing research [
7,
9,
13,
20‐
22]. The likely collinearity of BE features also makes it difficult to isolate the effects of individual BE attributes on outcomes of interest. With regard to the behaviours and outcomes, a majority of studies have relied on participants’ self-reported physical activity [
23] and height and weight (to calculate the dependent variable, BMI) [
19,
23,
24] respectively; measures known to be limited by social desirability bias [
25,
26]. BMI has been consistently employed as the dependent outcome. However, this measure is only a proxy for body fatness and differential health risks may be found across individuals of differing ethnicities. Furthermore, alternative proxy measures – WC in particular - have been shown to have stronger associations than BMI with health status and chronic conditions in adults [
27].
It is highly likely that physical activity mediates any association between BE features and body size, but few studies have considered this relationship [
7,
23,
28‐
32]. Such examinations are important to test the theoretical coherence of the association between the BE and body size; that is, if physical activity does not contribute somewhat to this relationship, it is possible that systematic error, most notably confounding, may exist. Furthermore, little is known of the potential mediating effect of sedentary behaviour on this relationship. A clear pathway exists between sedentary time (especially prolonged sitting) and body size [
33] and evidence suggests a link between the BE and sedentary time (particularly via sitting during motorised transport) [
34]. Thus it is possible that any observed relationship between BE features and body size may be mediated by sedentary time.
In this paper we extend on earlier research through analysis of objective measures of body size (BMI, WC) and physical activity and sedentary behaviour (accelerometry) collected from a random selection of participants living within urban neighbourhoods in New Zealand, purposively selected to maximise area-level variability in walkability. Building on our earlier investigations in the URBAN Study [
18], we examine the effect of objectively assessed physical activity and sedentary behaviours on the association between commonly-assessed BE features conceptualised as being associated with physical activity and body size. Hypotheses were that body size would be negatively associated with individual BE features under investigation, and that objectively-assessed physical activity (by decreasing body size) and sedentary behaviours (by increasing body size) would mediate this relationship.
Results
In total, 2033 adults participated (provided data from non-faulty accelerometers) in the study, reflecting a 44 % response rate. All data provided by the 2033 participants were included in the analysis to reduce bias introduced when decisions are made to include or exclude data [
55]. Participants contributed a total of 16,869,826 counts per hour, averaging to 9,177 counts per hour across 1–7 days of data. Descriptive information for participants, outcome measures, and neighbourhood characteristics are provided in Table
3. The population sampled were predominantly female (57 %) and had at least some access to a car (93 %), consistent with the overall New Zealand population (51 % female [
38], 92 % access to a motor vehicle [
56]). Compared with the New Zealand adult population, a greater proportion were employed (86 versus 66 %) [
57] and married (64 versus 45 %) [
38]. A correlation matrix of BE features and neighbourhood-level deprivation is provided in Table
1. Statistically significant positive correlations were observed between neighbourhood-level deprivation and dwelling density, street connectivity, and destination accessibility (
r = 0.36, 0.42, and 0.38, respectively, all
p < 0.01). All BE features were significantly correlated with each other (
r = 0.09 to 0.89,
p < 0.01), with the exception of dwelling density and mixed land use.
Table 3
Participant and neighbourhood characteristics
Age | 2020 | | |
15-29 | | 450 | 22.3 |
30-44 | | 782 | 38.7 |
45-54 | | 462 | 22.9 |
55-65 | | 326 | 16.1 |
Ethnicity | 2033 | | |
Māori | | 241 | 11.9 |
Non Māori | | 1792 | 88.1 |
Sex | 2032 | | |
Male | | 853 | 42.0 |
Female | | 1179 | 58.0 |
Qualification | 2025 | | |
No Qualification | | 546 | 27.0 |
School | | 234 | 11.6 |
Post School | | 474 | 23.4 |
Tertiary | | 771 | 38.1 |
Marital status | 2028 | | |
Never married | | 447 | 22.0 |
Married | | 1302 | 64.2 |
Previous married | | 279 | 13.8 |
Income ($NZ) | 1821 | | |
= <$ 40,000 | | 421 | 23.1 |
$ 40.001-60,000 | | 316 | 17.3 |
$ 60,001-80,000 | | 271 | 14.8 |
$ 80,001-100,00 | | 276 | 15.1 |
> $ 100,000 | | 541 | 29.6 |
Employment | 2030 | | |
Full time | | 1182 | 58.2 |
Part time/Non-standard work | | 542 | 26.7 |
Unpaid | | 306 | 15.1 |
Car access | 2032 | | |
Un restricted | | 1633 | 80.4 |
Restricted | | 241 | 11.9 |
No car access | | 158 | 7.8 |
Neighbourhood preference | 2006 | | |
Strongly prefer walkable | | 700 | 34.9 |
Moderately prefer walkable | | 333 | 16.6 |
Neutral | | 286 | 14.3 |
Moderately prefer less walkable | | 207 | 10.3 |
Strongly prefer less walkable | | 480 | 23.9 |
Neighbourhood deprivation (NZDep06) | 2033 | | |
Least deprived quintile | | 419 | 20.6 |
| | 380 | 18.7 |
| | 419 | 20.6 |
| | 435 | 21.4 |
Most deprived quintile | | 380 | 18.7 |
Exposure and outcome variables | Total | Mean | SD |
Body size | | | |
BMI | 2007 | 27.04 | 5.68 |
Waist circumference | 1994 | 88.60 | 15.01 |
Accelerometer-derived measures | 1838 | | |
Physical activity (accelerometer counts per hourb) | | 9177 | 4803 |
Sedentariness (percentage of time sedentary) | | 57.49 | 11.89 |
Neighbourhood exposuresc | 2033 | | |
Dwelling density | | 5.87 | 2.74 |
Street connectivity | | 5.42 | 2.39 |
Mixed land use | | 5.69 | 2.08 |
NDAI | | 11.74 | 4.93 |
Streetscape | | 87.57 | 11.26 |
Values for correlations between outcome measures and neighbourhood-level deprivation are provided in Table
2. Correlations were all in the expected direction, with both body size measures being significantly correlated with each other (
r = 0.84,
p < 0.01), while a negative correlation was observed between physical activity and sedentary behaviour (
r = −0.72,
p < 0.01). Small but significant correlations were also observed between body size variables and sedentary behaviour (positive), neighbourhood-level deprivation (positive), and physical activity (negative).
The estimates for all associations with BMI and waist circumference for Models 1 to 3 are presented in Tables
4 and
5, respectively. In the fully adjusted model (accounting for individual characteristics and neighbourhood-level deprivation; Model 1), street connectivity, NDAI, and streetscape were significant predictors of BMI (a 1 SD increase in each was associated with between 1.27 and 1.41 % reductions in BMI,
p < 0.05). For the same model, 1 SD increases in dwelling density, street connectivity, and NDAI were associated with reductions in waist circumference values of 1.97, 1.76 and 2.29 %, respectively (
p < 0.05).
Table 4
Change in log-BMI for one SD increase in neighbourhood exposures (n = 1813)
1 Adjusted for demographics, individual-level socioeconomic factors, neighbourhood-level deprivation, and neighbourhood preference | Dwelling density | −0.011036 | 0.005767 | | 0.06 | |
Street connectivity | −0.014172 | 0.005562 | | 0.01* | |
Land use mix | −0.009173 | 0.005355 | | 0.09 | |
NDAI | −0.012825 | 0.005395 | | 0.02* | |
Streetscape (SPACES) | −0.013417 | 0.004697 | | 0.007* | |
2 Plus including physical activity (log of accelerometer counts per hour) | Dwelling density | −0.009125 | 0.005768 | −0.001924 | <0.001* | n/aa |
Street connectivity | −0.012243 | 0.005598 | −0.001889 | <0.001* | 13.2 |
Land use mix | −0.008458 | 0.005297 | −0.000741 | 0.17 | 6.9 |
NDAI | −0.010952 | 0.005419 | −0.00191 | <0.001* | 14.6 |
Streetscape (SPACES) | −0.012626 | 0.004670 | −0.000809 | 0.11 | 5.7 |
3 Model 1 plus including percentage time spent sedentary | Dwelling density | −0.0101833 | 0.0057573 | −0.000873 | 0.18 | 7.3 |
Street connectivity | −0.0131472 | 0.0055790 | −0.001039 | 0.09 | 6.9 |
Land use mix | −0.0089294 | 0.0053261 | −0.00025 | 0.69 | 2.4 |
NDAI | −0.0118039 | 0.0054092 | −0.00109 | 0.06 | 8.1 |
Streetscape (SPACES) | −0.0128561 | 0.0046980 | −0.000570 | 0.31 | 4.1 |
Table 5
Change in log- waist circumference for one SD increase in neighbourhood exposures (n = 1801)
1 Adjusted for demographics, individual-level socioeconomic factors, neighbourhood-level deprivation, and neighbourhood preference | Dwelling density | −0.019921 | 0.006521 | | 0.004* | |
Street connectivity | −0.0177428 | 0.0066846 | | 0.01* | |
Land use mix | −0.0020776 | 0.0067994 | | 0.76 | |
NDAI | −0.0231200 | 0.0059381 | | <0.001* | |
Streetscape (SPACES) | −0.0108457 | 0.0060160 | | 0.08 | |
2 Plus including physical activity (log of accelerometer counts per hour) | Dwelling density | −0.017381 | 0.006388 | −0.00248 | <0.001* | 12.3 |
Street connectivity | −0.0151079 | 0.0065353 | −0.00254 | <0.001* | 14.3 |
Land use mix | −0.001062 | 0.006512 | −0.000962 | 0.18 | 6.6 |
NDAI | −0.0206553 | 0.0058506 | −0.00242 | <0.001* | 10.4 |
Streetscape (SPACES) | −0.0096708 | 0.0057933 | −0.001126 | 0.08 | 9.6 |
3 Model 1 plus including percentage time spent sedentary | Dwelling density | −0.0191994 | 0.0063932 | −0.000735 | 0.23 | 3.6 |
Street connectivity | −0.0169189 | 0.0065567 | −0.000822 | 0.17 | 4.5 |
Land use mix | −0.0018313 | 0.0066369 | −0.000214 | 0.71 | 1.9 |
NDAI | −0.0223459 | 0.0058315 | −0.000824 | 0.16 | 3.4 |
Streetscape (SPACES) | −0.0102434 | 0.0058882 | −0.000600 | 0.27 | 5.0 |
Adjusting for the potential mediation by physical activity and sedentary behaviour (Models 2 and 3, respectively) each compared to our ‘best’ total estimate of the BE-BMI association in Model 1 saw 13.2 % (street connectivity adjusting for time spent sedentary) to 14.6 % (NDAI adjusting for physical activity) of the BE-BMI association explained, and 10.4 % (NDAI adjusting for physical activity) to 14.3 % (street connectivity adjusting for physical activity) of the BE-WC association explained. An inconsistent mediation effect was observed for physical activity in the BMI-dwelling density relationship, that is, a non-significant association was found for the total effect, yet a significant mediating effect was found for physical activity. Therefore, no interpretation of results for this model have been provided [
58].
Discussion
Significant negative associations between objectively-assessed BE characteristics and body size were observed in the current study. After adjusting for individual characteristics, neighbourhood preference, and neighbourhood-level deprivation, street connectivity, and destination accessibility were associated with reduced BMI and waist circumference (ranging from −1.27 to −2.29 % for a 1 SD change in each BE variable). Higher quality streetscape, as assessed by the NZ-SPACES tool, was also associated with reduced BMI, but this relationship did not hold true for waist circumference. Dwelling density was associated with reduced WC (−1.97 %, p = 0.004), and this relationship neared significance for BMI (−1.10 %, p = 0.061).
When considering the results for the fully adjusted model, the differences found in body size may appear small. However, if we compare settings at either end of the BE spectrum, the consequences of living in these neighbourhoods on body size outcomes could be important. For example, when comparing the 5
th and 95
th percentiles of dwelling density in the current study (data reported in Witten et al. [
18]), a SD of 3.14 is found. If the current study findings are applied to compare these two environments, a 3.14 SD in dwelling density would equate to a difference of approximately 1 BMI unit in an adult, or roughly 3–4 kg, dependent on height. It is important to note that observed associations are not independent of each other and that the approach taken to analysis in the current study does not enable conclusions to be drawn about the cumulative and combined effect of individual BE features on outcomes. For instance, if we consider street connectivity and destination accessibility, a number of scenarios may exist – one could be an indicator of the other, the two factors may interact, or they could have completely independent effects on body size; however it is not possible to determine which is the case from this study. With the exception of streetscape and neighbourhood-level deprivation, and mixed land use and dwelling density, significant correlations between all BE features were observed (
p < 0.01), therefore collinearity was a concern.
The associations found between the BE and body size in the current study are consistent with earlier US research; when comparing the 90
th to the 10
th percentile for BE features in over 13,000 adults in New York City, Rundle et al. [
59] observed reductions in BMI units of 0.41, 0.33, 0.34, and 2.86 kg/m
2 with increases in mixed land use, bus stop density, subway stop density, and population density, respectively. Likewise, in a study of over 16,000 Texas adults, Hoehner et al. [
30] determined that men and women living in neighbourhoods 1 SD above the mean for older homes (proxy measure of more walkable neighbourhoods) and shorter commute times had BMI values of 0.77 and 0.84 kg/m2 lower than those living in neighbourhoods 1 SD below the mean.
Advancing on this earlier work, the mediating effects of objectively assessed physical activity and sedentary behaviour on this relationship were considered. Consistent with earlier research [
28,
32], no mediating effect was found for sedentary behaviour. While clear associations exist between sedentary time (particularly prolonged sitting) and body size in adults [
60,
61], studies of associations between the BE and sedentary time are scarce [
62]. It may be hypothesised from this small but consistent evidence base that a substantial relationship between the BE and objectively assessed sedentary time is unlikely. Conversely, it is also plausible that a clear understanding of this relationship has been limited by methodological issues, and lack of sensitivity and specificity in quantifying outcomes, particularly sedentary behaviours. For example, while improving on self-reported behaviours, the use of accelerometry to measure time spent sedentary is still imprecise; postural transitions cannot be assessed, and consensus is lacking on best practice for measurement protocols, data processing, cleaning, and aggregation [
63]. We recommend employing additional methods in future to explore possible relationships between sedentary time and the BE in more detail (e.g., inclinometers, life-logging cameras, global positioning systems combined with accelerometry).
In contrast, a significant mediating effect of physical activity was observed on the relationship between body size and NDAI and street connectivity. These findings are consistent with the limited research that has considered the mediating effects of physical activity on the relationship between the BE and BMI [
28‐
32,
64], and confirm the hypothesis that the relationship between BE and body size, mediated by physical activity, is not the result of systematic error. Preliminary analyses (not reported here) showed that the addition of neighbourhood preference to modelling resulted in no changes to relationships and only minor changes in the magnitude of differences. It is worth noting that when selecting their preferred neighbourhood, participants were requested to assume uniformity across neighbourhood types with respect to key issues such as schools, neighbourhood demographics and housing cost, which may have confounded findings somewhat as it is an idealistic measure. In addition, the neighbourhood preference measure employed was relatively broad, offering participants a choice of two neighbourhoods, essentially dichotomised by features that enable or disable use of active and public transport modes; consequently, selection effects may not have been captured completely.
The current study improves on limitations in previous research in this field [
7,
65] by: undertaking robust assessment of the BE using contemporary GIS databases; purposefully selecting heterogeneous neighbourhoods in terms of walkability and region; randomly selecting participants; objectively assessing physical activity and body size (using BMI and waist circumference); using a continuous measure of BMI rather than thresholds (which can bias results [
66]; a consideration especially important given the ethnic diversity in this study); adjusting for neighbourhood preference; and utilising mediation analyses to consider the effect of physical activity on the BE-body size relationship. Debate regarding approaches for accelerometer data reduction approach is ongoing [
67], and to date, no agreed-upon best-practice method exists. In this study we employed a pragmatic accelerometer data reduction protocol that enabled inclusion of data for all complete hours of wear time (as opposed to limits of hours/days of wear), with the aim of reducing bias (for example with less compliant participants, [
55]). Mean accelerometer counts were used to describe physical activity, recognising the issues surrounding the use of accelerometer count thresholds, and the contribution of activity of any intensity to health [
68].
We did not estimate associations with individual BE characteristics with adjustments for all other BE features in the same model (due to collinearity) – this means that associations with each BE feature are not independent of associations with other BE attributes (as they are likely to be confounded by these variables). Caution should be applied when generalising these findings internationally, considering the substantial variability in urban form across countries [
69]. It is also worth noting we did not consider the food environment in this study. In their earlier research, Brown et al. [
29] undertook detailed and robust assessment of average caloric intake (two 24-h recalls); when controlling for this factor, a significant mediating effect of objectively-assessed physical activity still remained for the relationship between the BE and BMI. It is possible that clustering or collinearity exists between BE features that support physical activity and healthy nutritional practices; it is also plausible that the food environment (e.g., density of fast food outlets) may play an additive role with the physical activity environment in predicting body size.
While this study contributes to the cross-sectional evidence base, it does not enable the identification of causal relationships. Moreover, mediation analyses with cross-sectional data (particularly when stratified such as in the current study) can be subject to collider bias and measurement error [
70]. Study findings are limited to New Zealand adults only; it is possible that differential findings may be observed for younger and older population groups, and for other population groups. With a response rate of 44 %, it is possible that self selection bias was present in the current study sample, although it is worth noting this response rate is higher than similar studies (e.g., 12 % in the study of Owen et al. [
71], and 26 % in Frank et al. [
37]).
Examination of relationships between objectively assessed body size, nutrition behaviours, physical activity, and food environments in future population research (particularly including younger and older participants) would be challenging, but an important next step in determining the relative contributions of behaviours and environmental features to obesity risk for people of all ages. Longitudinal, repeated measurements of BE exposures, risk behaviours, and body size outcomes are also necessary. Making use of existing longitudinal health datasets and alignign BE measurement with routine population studies are approaches that may make this feasible.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
HMB, GS, KW, RAK, SM, and EH originally conceived of and designed the URBAN study; MO, TB, and VI further contributed to the study design and coordination. HMB led the data collection and data cleaning. MO, KW, TB, and GS conceptualised and led the current investigation. KP, TB, and PS conceptualised and undertook data treatment and analyses. SM led the built environment measurement. JP and RAK provided geographic expertise and input to the mansucript draft; JP led the development and construction of the NDAI. MO drafted the manuscript. All authors read, contributed to, and approved the manuscript.
Funding
We thank all members of the Understanding the Relationship between Activity and Neighbourhoods (URBAN) team, as well as the study participants, for their contributions to the research. The URBAN Study was funded by the Health Research Council of New Zealand. The authors are independent of the funder, and the funder had no role in the design and conduct of the study or in the preparation of the manuscript. The authors declare they have no actual or potential competing financial interests.