The burden of cardiometabolic diseases (e.g. type 2 diabetes, heart disease and stroke) continues to increase globally(Reference Roth, Mensah and Johnson1). Population-wide approaches are advocated as effective and sustainable strategies to combat the rise in cardiometabolic diseases and their risk factors(2). Public health initiatives focusing on the built environment are important in this regard, as the features of neighbourhood environments are known to be associated with residents’ health-related behaviours, such as physical activity(Reference Giles-Corti, Vernez-Moudon and Reis3).
There is a body of evidence supporting the relationships of neighbourhood walkability (a measure consisting typically of residential density, street connectivity and land use diversity) with walking for transport(Reference Smith, Hosking and Woodward4) and physical activity(Reference Sallis, Cerin and Kerr5). Given the well-established health benefits of physical activity(Reference Colberg, Sigal and Yardley6,Reference Lear, Hu and Rangarajan7) , it can be expected that residents living in high-walkability neighbourhoods could have better cardiometabolic health profiles than those in lower walkability areas. However, there is inconsistent evidence for the relationships between walkability and cardiometabolic health indicators. Although a recent review of longitudinal studies reported evidence for the relationship between higher walkability and lower cardiometabolic risk(Reference Chandrabose, Rachele and Gunn8), an earlier review of mostly cross-sectional studies reported mixed findings on walkability and obesity(Reference Mackenbach, Rutter and Compernolle9). More recent studies also identified null associations(Reference Barbosa, Guerra and Santos10), associations in an unexpected direction(Reference Chandrabose, Cerin and Mavoa11,Reference Braun, Rodriguez and Evenson12) and non-linear associations between walkability and cardiometabolic health indicators(Reference Howell, Tu and Moineddin13).
One potential explanation for this lack of consistency is that high-walkability neighbourhoods may have other environmental features that could influence health-related risk behaviours. For instance, high-walkability neighbourhoods, which have more commercial destinations, can have a wider range of food outlets, including those providing unhealthy food(Reference Bastian and Napieralski14,Reference Frank, Saelens and Chapman15) . It has been shown that the perceived availability of unhealthy food is associated with poor dietary behaviours(Reference Zhao, McGowan and Zenk16,Reference Caspi, Sorensen and Subramanian17) . However, evidence on associations between neighbourhood walkability and residents’ dietary attributes is sparse, with only one study reporting that Walk Score® (a web-based measure of neighbourhood walkability) was not associated with the consumption of vegetables and fruits among Taiwanese older adults(Reference Liao, Lin and Lai18).
Area-level socio-economic status (SES) is also known to be related to residents’ dietary attributes(Reference Ball, Lamb and Costa19,Reference Lagstrom, Halonen and Kawachi20) . A recent systematic review of Australian studies found that socio-economically disadvantaged areas tend to have more unhealthy food outlets(Reference Needham, Sacks and Orellana21), suggesting a possibility that high-walkability low-SES neighbourhoods may have greater availability of unhealthy food options. It can be thus postulated that walkability and area-level SES may be jointly associated with residents’ dietary attributes. For instance, residents of high-walkability low-SES neighbourhoods may have poor dietary attributes due to the exposure to unhealthy food outlets.
In this cross-sectional study, we first examined associations of neighbourhood walkability with residents’ dietary attributes, then examined whether such associations may be moderated by area-level SES.
Methods
Data source and study participants
We used data from the third wave of the Australian Diabetes, Obesity and Lifestyle Study (AusDiab3), which collected data in 2011–2012. Details of this study have been described previously(Reference Dunstan, Zimmet and Welborn22). Briefly, at baseline (1999–2000), participants were recruited from forty-two study sites located in the six Australian states and the Northern Territory (six sites in each), using a two-stage stratified cluster sampling method. Eligible participants were non-institutionalised adults aged over 25 years, without any physical/intellectual disabilities, and residing in private dwellings for at least six months prior to the data collection. At baseline, 11 247 participants (response rate: 55·3 %) provided data. Of these, 4614 (follow-up rate: 44·6 %) provided data in AusDiab3. The analytical sample size was 3590, after excluding seven participants who reported being pregnant during data collection (since pregnancy-related factors may influence dietary attributes), 133 whose residential locations could not be geocoded and 884 with missing data for outcome and potential confounding variables used in the study. Nearly half of AusDiab3 participants had relocated their residence after the baseline data collection, resulting in scattered residential locations across multiple regions in Australia.
Exposure variable
The exposure variable was the neighbourhood walkability index, which was a composite measure of residential density, intersection density and destination density. The rationale, methods and spatial data sources used to calculate this walkability index are described in Supplemental Material 1. Briefly, it was calculated for each participant within a 1-km street-network buffer (sausage type) around the residential location. A 1-km distance was chosen to represent the local neighbourhood, as it has been shown to be a typical distance within which most home-based walking activities can take place(Reference Sugiyama, Kubota and Sugiyama23). Residential density was calculated as the total count of dwellings within the buffer divided by its area. Intersection density was calculated as the number of 4-way (or more) intersections within the buffer divided by its area. We used the density of 4-way (or more) intersections because it was observed as a better measure of street connectivity in the Australian context than the density of typically used 3-way (or more) intersections (see online supplementary material, Supplemental Fig. S1). Destination density was calculated as the total count of destinations to which residents would travel on a regular basis (supermarkets, convenience stores and public transport stops) within the buffer divided by its area. The unit for all density measures was counts/km2. The walkability index was expressed as a standardised score of the summed z-scores of each density measure.
Outcome variables
The dietary attributes examined as the outcome variables were three dietary patterns and total dietary energy intake. Participants completed a validated self-administered semi-quantitative FFQ, which assessed the daily intake of seventy-four food items (on a ten-point frequency scale) over the past 12 months, with additional questions on usual eating habits and portion size(Reference Hodge, Patterson and Brown24). Dietary patterns were identified by applying factor analysis to the estimated daily intake of these food items. The methods for identifying dietary patterns have been reported previously(Reference Cao, Dunstan and Sethi25). A brief overview of these methods is provided in Supplemental Material 2. Three dietary patterns were identified: Western diet, prudent diet and mixed diet. Western diet was characterised by high consumption of take-away foods, snacks, processed meat and red meat. Prudent diet was characterised by high consumption of vegetables (root, leafy and stalk) and fruits. Mixed diet was characterised by high consumption of fish, cereals, pasta, rice and poultry. For ease of interpretation, the three dietary factor scores (mean = 0, sd = 1) were transformed to have a mean of 100 with an sd of 25 (i.e. transformed value = original value × 25 + 100). The average daily intake of each food item (in grams) was converted into total dietary energy intake in kJ/d.
Potential effect modifier
We examined area-level SES as a potential effect modifier of the relationships between walkability and dietary attributes. We used the Index of Relative Socio-economic Disadvantage (IRSD), which is a census-based composite index indicating the level of disadvantage of an area(26). The Australian Bureau of Statistics calculates IRSD scores by applying principal component analysis to relevant area-level variables such as proportions of lower income households, unemployed people, households without cars, people with lower education and overcrowded private dwellings. Lower IRSD scores indicate more socio-economically disadvantaged areas. In the present study, we used the 2011 IRSD scores corresponding to Local Government Areas (LGA)(27). LGA is an administrative division below State and Territory governments; there were 568 LGA in Australia in 2011. Each LGA is administered by a local council that is responsible for providing a range of services and infrastructure for the community. Participants in the current study resided in 196 LGA, which had a median population of 72 972 (first quartile: 36 506, third quartile: 152 389).
Statistical analyses
Descriptive statistics for all variables were calculated for the whole sample and subgroups stratified by IRSD tertiles (low, medium and high SES areas).
For estimating the main effects (i.e. associations of neighbourhood walkability with dietary attributes), we used two-level random intercept linear regression models that accounted for potential area-level clustering (participants at level 1 and LGA at level 2)(Reference Hox28). In the primary analyses, regression models adjusted for potential confounders: individual socio-demographic characteristics (gender, age, education, marital status, employment status, household income and children in the household) and area-level SES (IRSD).
To examine the potential effect modification by area-level SES on the association between walkability and dietary attributes, a two-way interaction term of walkability and IRSD was added to the regression model of each outcome and observed for an interaction effect. If an interaction was found, associations between walkability and dietary attributes were estimated for each subgroup stratified by IRSD tertiles.
All statistical analyses were conducted in R 4.0.5 (R Core Team). The package ‘lme4’ version 1.1.14 was used to model linear mixed models. The package ‘ggmap’ version 2·7·9 was used to geocode study participants’ residential addresses. For all other geographic information systems analyses, we used ArcGIS Pro version 2·3·3 (ESRI Inc).
Results
Table 1 shows the key characteristics of study participants. Their mean age was 59 years (range: 34–86 years) and 54 % were women. The medium SES areas had a relatively higher mean walkability value compared with the low and high SES areas. The average total dietary energy intake was 7150 kJ/d.
Characteristics of the analytical sample
SES, socio-economic status; IRSD, Index of Relative Socioeconomic Disadvantage corresponding to the Local Government Area where participant resided, a score of 1000 equals the national average, with lower scores indicating more socio-economically disadvantaged areas
Table 2 shows the results of the regression models fitted to estimate the main and interaction effects. As shown in the main effect models, the walkability index was not associated with any of the dietary attributes in the whole sample. However, we found interaction effects of walkability and IRSD on Western diet scores (P < 0·001) and total energy intake (P = 0·012), indicating that the associations of walkability with these two dietary attributes may vary by area-level SES. Such interaction effects of walkability and IRSD were not found for prudent and mixed diet scores.
Associations of neighbourhood walkability with dietary attributes, AusDiab3 study, 2011–2012 (n 3590)
IRSD, Index of Relative Socioeconomic Disadvantage
* Regression coefficients correspond to one sd increment in walkability index. These models adjusted for age, gender, education, work status, household income, marital status, household children status, area-level socio-economic status (IRSD as a continuous variable) and corrected for area-level clustering. The interaction effect models additionally included the product of walkability and IRSD scores.
As shown in Fig. 1, in low SES areas, walkability was positively associated with Western diet scores (P = 0·062) and total energy intake (P = 0·066). On the other hand, in high SES areas, walkability was inversely associated with Western diet scores (P = 0·021) and total energy intake (P = 0·058).
Regression coefficients for the Western diet score (a) and total dietary energy intake (b) corresponding to one sd increment in walkability index, stratified by area-level socio-economic status (SES), AusDiab3 study, 2011–2012 (n 3590). Each model was adjusted for age, gender, education, work status, household income, marital status, household children status and corrected for area-level clustering
Discussion
In our examination of the associations of neighbourhood walkability with dietary attributes using a cohort of Australian adults, we did not observe associations between walkability and any of the dietary attributes in the whole sample. However, we found interaction effects of walkability and area-level SES on Western dietary pattern and total energy intake. In the analyses stratified by area-level SES subgroups, higher walkability was associated with higher Western diet scores (i.e. higher consumption of take-away foods, snacks, red meat and processed meat) and higher total energy intake in low SES areas. In contrast, in high SES areas, higher walkability was associated with lower Western diet scores and lower total energy intake.
There is little evidence available on the associations between walkability and dietary attributes. Consistent with our main effect results, one Taiwanese study also did not find an association between neighbourhood walkability and consumption of vegetables and fruits among older adults(Reference Liao, Lin and Lai18). However, this Taiwanese study is not directly comparable to our study because it was conducted in a non-Western setting, investigated older adults, used different types of exposure and outcome measures and did not examine the effect modification by area-level SES. Our findings contribute to advancing the understanding of the relationship between walkability and dietary attributes, by showing its dependence on the area-level SES.
Differential associations of walkability with dietary attributes observed between the low and high SES areas may be explained by differences in food environments, differences in individuals’ food preferences, or both. Previous research has demonstrated that high-walkability neighbourhoods have better access to more and diverse retail food outlets(Reference Bastian and Napieralski14,Reference Frank, Saelens and Chapman15) . Further, Australian studies have consistently found that socio-economically disadvantaged areas tend to have a greater number of unhealthy food outlets(Reference Needham, Sacks and Orellana21). In the context of our study, residents of high walkability neighbourhoods in low SES areas might have had greater exposure to unhealthy food outlets, whereas those residing in high walkability neighbourhoods in high SES areas might have better access to healthy food outlets. Given that availability of food types in local areas could be a key determinant of dietary attributes(Reference Janssen, Davies and Richardson29,Reference Murphy, Badland and Koohsari30) , differential characteristics of neighbourhood food environments may be contributing to the observed moderation effect by area-level SES. In addition, differences in residents’ food preferences between low and high SES areas might also be contributing to the current findings(Reference Janssen, Davies and Richardson29,Reference Murphy, Badland and Koohsari30) . Some population groups may prefer to purchase and consume unhealthy food due to affordability, ease of preparation, cultural background or lack of nutritional awareness(Reference Janssen, Davies and Richardson29). It is possible that residents of high-walkability neighbourhoods in socio-economically disadvantaged areas may have chosen unhealthy foods for various reasons, even if healthy food options are available. Future research might consider investigating how these factors (food environments and food choice) are involved in the link between walkability, area-level SES and dietary attributes.
Research on the relationship between neighbourhood walkability and cardiometabolic health indicators (e.g. obesity, hypertension, type 2 diabetes, CVD and their risk markers) typically considers physical activity as the direct behavioural pathway(Reference Chandrabose, Rachele and Gunn8). Our findings suggest that dietary attributes may also be involved in this relationship. Higher consumption of unhealthy foods among residents of high walkability neighbourhoods in socio-economically disadvantaged areas may have partly contributed to the inconsistent findings observed for the associations between walkability and cardiometabolic health indicators(Reference Barbosa, Guerra and Santos10–Reference Howell, Tu and Moineddin13). In addition, high-walkability neighbourhoods may also expose residents to other environmental factors that can increase the risk of cardiometabolic diseases such as high levels of noise/air pollution, low levels of natural features (green spaces, blue spaces and tree cover) and psychological stress due to overcrowding(Reference Frank, Iroz-Elardo and MacLeod31). Further empirical research needs to investigate how these environmental exposures are independently and jointly related to health behaviours and outcomes to obtain a more-comprehensive understanding of the relationship between the built environment and population health.
Our findings are relevant from a perspective of socio-economic inequalities in health. It can be argued based on the findings of this study and previous studies that higher walkability may benefit high SES areas due to greater physical activity and better dietary attributes, while such benefits may be limited in low SES areas. This could suggest that improving walkability (i.e. densifying areas through infill development and high-density redevelopment) may result in widening health disparities, if they are effective in promoting health-enhancing behaviours (more physical activity, healthier diet) only in high SES areas. Inequalities in health are persistent in our society and reducing them is a key challenge in public health(Reference Liburd, Hall and Mpofu32). To address this challenge, research may need to consider developing specific environmental initiatives targeting socio-economically disadvantaged neighbourhoods to reduce inequalities in health. Such focused efforts are warranted to prevent health disparities from worsening.
Our study has several strengths. We investigated a large sample of adults whose residences were in diverse geographical settings across Australia. This provided greater variabilities in walkability and area-level SES measures. We used a nationally consistent objective measure of neighbourhood walkability. In addition to total dietary energy intake, which is an important dietary attribute from an energy balance point of view for obesity prevention(Reference Hill, Wyatt and Peters33), we also examined dietary patterns, which were characterised by not only the quantities of food intake but also by the patterns of specific types of food consumed. Such dietary pattern scores are considered to be better predictors of cardiometabolic disease risk(Reference Hu34). A limitation of this study is the generalisability of the findings to the Australian population. There may have been a selection bias due to the relatively high attrition rate (55 %) at the 12-year follow-up. The average amount of total dietary energy intake in our study sample was slightly lower than the national average at that time(35), which may be due to the attrition bias. Since this is a cross-sectional study, it is not possible to rule out the possibility of reverse causation, i.e. people with specific dietary attributes may have chosen to live in areas with better availability of their preferred food, rather than characteristics of the neighbourhood affecting their dietary attributes. Future quasi-experimental studies may examine whether moving to high walkability neighbourhoods changes dietary attributes and whether such changes are moderated by area-level SES.
Conclusions
Our findings suggest that the relationships between neighbourhood walkability and residents’ dietary attributes are complex and subject to area-level socio-economic status. Urban planning initiatives being implemented to improve local area walkability, such as the ‘20-minute neighbourhoods’ development plan in Australia(36), may assist residents to be more active in their neighbourhoods. However, such development plans may have differential health impacts on neighbourhoods with different levels of socio-economic status. If they provide a greater health benefit to only high SES areas, they can result in expanding health disparities. Our findings demonstrate the need to consider the combination of physical activity and dietary attributes with the socio-economic context where they take place in order to better understand the relationships between the built environment and health. Evidence from such research can contribute to developing more responsible urban planning policies that can enhance population health through multiple health behaviours without widening socio-economic inequalities in health.
Acknowledgements
Acknowledgements: The AusDiab study acknowledges the support and assistance given by K. Anstey, B. Atkins, B. Balkau, E. Barr, A. Cameron, S. Chadban, M. de Courten, A. Kavanagh, D. Magliano, S. Murray, K. Polkinghorne, J. Shaw, T. Welborn, P. Zimmet and all the study participants. The authors are grateful to Professor Billie Giles-Corti (Director of the Healthy Liveable Cities Lab, RMIT University, Australia) for providing thoughtful comments in writing this manuscript. Financial support: The AusDiab study was co-coordinated by the Baker Heart and Diabetes Institute and sponsored by National Health and Medical Research Council (NHMRC grants: #233200, #1007544); Australian Government Department of Health and Ageing; Abbott Australasia Pty Ltd.; Alphapharm Pty Ltd.; Amgen Australia; AstraZeneca; Bristol-Myers Squibb; City Health Centre-Diabetes Service-Canberra; Department of Health and Community Services, Northern Territory; Department of Health and Human Services, Tasmania; Department of Health, New South Wales; Department of Health, Western Australia; Department of Health, South Australia; Department of Human Services, Victoria; Diabetes Australia; Diabetes Australia Northern Territory; Eli Lilly Australia; Estate of the Late Edward Wilson; GlaxoSmithKline; Jack Brockhoff Foundation; Janssen-Cilag; Kidney Health Australia; Marian & FH Flack Trust; Menzies Research Institute; Merck Sharp & Dohme; Novartis Pharmaceuticals; Novo Nordisk Pharmaceuticals; Pfizer Pty Ltd.; Pratt Foundation; Queensland Health; Roche Diagnostics Australia; Royal Prince Alfred Hospital, Sydney; Sanofi Aventis and sanofi-synthelabo. Spatial data (road network and destinations data) were provided by the Healthy Liveable Cities Lab from the National Liveability Study funded by The Australian Prevention Partnership Centre (NHMRC Grant # 9100003). NO is supported by NHMRC Senior Principal Research Fellowships (grants: #1003960 and #1107672) and by the Victorian Government’s Operational Infrastructure Support Program. DD was supported by an NHMRC Senior Research Fellowship (#1078360) and the Victorian Government’s Operational Infrastructure Support Program. The funders had no role in the design, analysis or writing of this article. Authorship: M.C. and T.S. conceptualised the study and wrote the manuscript. M.C. calculated environmental measures and conducted statistical analysis; T.S. and N.O. guided the execution of the study; YC calculated dietary pattern measures; F.K.S. was involved in spatial data analysis; D.D., N.O. and T.S. were involved in collecting AusDiab data; N.O., Y.C., N.H., C.H. and D.D. contributed to manuscript writing by reviewing and substantially revising the drafts. Ethics of human subject participation: This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving research study participants were approved by the Alfred Hospital Ethics Committee (# 39/11). Written informed consent was obtained from all subjects.
Conflict of interest:
There are no conflicts of interest.
Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S1368980022001197
Open access
