Neighbourhoods play a key role in health^{(Reference Yen, Michael and Perdue1–Reference Roux3)}. Living within socio-economically advantaged neighbourhoods is linked with greater self-rated health^{(Reference Wen, Browning and Cagney4)}, decreased rates of obesity^{(Reference King, Kavanagh and Jolley5,Reference Sundquist, Malmström and Johansson6)} and lower overall mortality^{(Reference Bosma, Dike van de Mheen and Borsboom7)}. In contrast, people living within socio-economically disadvantaged neighbourhoods, such as those with high rates of poverty and lower educational attainment, are more likely to experience negative health outcomes^{(Reference Diez-Roux, Nieto and Muntaner8–Reference Burdette and Hill10)}, even after accounting for individual socio-economic characteristics (e.g. income and education)^{(Reference Ross and Mirowsky11)}. Although neighbourhood disadvantage has persistent negative effects on health, the structures by which neighbourhood disadvantage operates to affect health outcomes are less clear^{(Reference Wight, Cummings and Miller-Martinez12)}.

The food environment may be one avenue through which neighbourhood disadvantage impacts individual and community health^{(Reference Li, Harmer and Cardinal2,Reference Morland and Evenson13)} . Neighbourhood food environments have been repeatedly linked with fruit and vegetable consumption as well as consumption of unhealthy foods (e.g. foods containing excessive fat and sodium)^{(Reference Caspi, Sorensen and Subramanian14)}. Additionally, childhood exposures may influence food quality and eating behaviours throughout the life course^{(Reference Devine15)}. People who consume inferior quality foods have higher rates of obesity^{(Reference Wolongevicz, Zhu and Pencina16)}, hypertension^{(Reference Forman, Stampfer and Curhan17)}, heart disease^{(Reference Schulze and Hu18)}, diabetes^{(Reference Salas-Salvadó, Bulló and Estruch19)} and cancer^{(Reference Snowdon, Phillips and Choi20)}. Further, certain populations are at greater risk for diet-related conditions than others; disparities by socio-economic status are well documented across the USA^{(Reference Claire Wang, Gortmaker and Taveras21–23)}. The combination of limited access to supermarkets, increased reliance on convenience stores, and the difference in price between healthy and unhealthy foods has the potential to perpetuate health disparities. Ultimately, unhealthy food environments may exacerbate health inequities for socially disadvantaged groups^{(Reference Ford and Dzewaltowski22,Reference Burgoine, Mackenbach and Lakerveld24)} .

Although density of healthy and unhealthy food establishments can impact food choice^{(Reference Devine15,Reference Moore, Diez Roux and Nettleton25)} , research has produced mixed findings on the relationship between neighbourhood disadvantage and the quality of available foods^{(Reference Richardson, Boone-Heinonen and Popkin26,Reference Richardson, Meyer and Howard27)} . Some research suggests that disadvantaged neighbourhoods may have fewer high-quality supermarkets^{(Reference Moore and Diez Roux28)}, whilst other studies found no association between area deprivation and supermarket access^{(Reference Maguire, Burgoine and Monsivais29)}. In one study, while high-density urban-disadvantaged neighbourhoods had greater access to all food outlets, low-density urban-disadvantaged neighbourhoods had greater access to only fast-food establishments^{(Reference Richardson, Boone-Heinonen and Popkin26)}. In another study, low-income neighbourhoods had fewer chain supermarkets, which typically stock healthier foods^{(Reference Sallis, Nader and Rupp30,Reference Horowitz, Colson and Hebert31)} , and more non-chain supermarkets and grocery stores, which stock lower quality foods^{(Reference Powell, Slater and Mirtcheva32)}.

Considerable research has examined food deserts, areas that lack access to affordable healthy foods^{(Reference Walker, Keane and Burke33,Reference Beaulac, Kristjansson and Cummins34)} , whereas other research has studied food swamps, areas with a greater density of food outlets selling unhealthy foods^{(Reference Sushil, Vandevijvere and Exeter35)}. However, few studies have addressed both food deserts and swamps^{(Reference Cooksey-Stowers, Schwartz and Brownell36–Reference Hager, Cockerham and O’Reilly38)} or utilised a composite measure to characterise the relative density of unhealthy food to healthier food options^{(Reference Sadler, Gilliland and Arku39–Reference Clary, Ramos and Shareck41)}. In contrast, many studies have used simple counts to enumerate the number of distinct types of establishments or have confined such measures to a given administrative neighbourhood, thereby ignoring proximal food outlets outside this boundary^{(Reference Cooksey-Stowers, Schwartz and Brownell36,Reference Bridle-Fitzpatrick37)} .

Healthy food outlets are those that increase availability of and access to fresh fruits and vegetables, whilst unhealthy food outlets include those that carry a disproportionate number of items with low nutritional value^{(Reference Lytle and Sokol40)}. Little research has applied advanced spatial methods, such as kernel density, to understand the distributions of healthy and unhealthy food outlets, and none has applied these methods to understand the relationship between the food environment and socio-economic disadvantage (SED)^{(Reference Moore, Roux and Brines42,Reference Rundle, Neckerman and Freeman43)} . Kernel density methods can be used to quantify the impact of a particular food outlet over a given distance, thereby operationalising the assumption that its influence will be stronger on more proximal areas with decaying influence as distance increases^{(Reference Moore, Roux and Brines42)}. This method can be used to study how neighbourhood disadvantage relates to the relative densities of healthy and less healthy food outlets across a geographic landscape to understand their composite relationship with SED^{(Reference Kloog, Haim and Portnov44)}. A comprehensive look at socio-economic status and area demographics is important for gaining an understanding of a neighbourhood; however, when labelling neighbourhoods as disadvantaged or advantaged, often only one neighbourhood indicator has been assessed, such as the percentage of the population below the poverty line^{(Reference Richardson, Boone-Heinonen and Popkin26)}. Few studies have created an index for both SED and the neighbourhood food environment^{(Reference Maguire, Burgoine and Monsivais29,Reference Hughey, Walsemann and Child45)} .

Given these gaps in the literature, the purposes of this study were to examine (1) the separate relationships between SED and the density of multiple types of food outlets, and (2) relationships between SED and composite unhealthy and healthy food environment indices. It was hypothesised that higher levels of block group disadvantage would be associated with a higher density of unhealthy food establishments, a lower density of grocery stores and a more negative overall food environment.

Methods

Study setting

This study occurred in a large Southeastern US county with an area of 795 square miles (2060 km²). As seen in Table 1, in 2013, the county had a total population of 474 266, of which 77·1 % was non-Hispanic white, 18·5 % was African American and 8·5 % was Hispanic or Latino. The median household income was $48 886 and approximately 15·0 % of residents lived below the federal poverty line⁽⁴⁶⁾. The county contained 255 census block groups which served as the units of analysis for this study. Block groups are permanent subdivisions of a county, are the second-smallest geographical unit recognised by the US Census Bureau (usually 600–3000 people with varying area depending on population density) and have been used to approximate neighbourhoods in previous studies^{(Reference Gordon-Larsen, Nelson and Page47–Reference Frank, Saelens and Chapman49)}.

Table 1 Food environment categories and definitions

Food outlet enumeration and categorisation

As described elsewhere^{(Reference Hughey, Walsemann and Child45)}, food outlet data were obtained, enumerated and classified from two secondary sources, the State Department of Health and Environmental Control and Info USA, both of which have been used frequently for research^{(Reference Glanz, Sallis and Saelens50,Reference Saelens, Glanz and Sallis51)} . Duplicate entries were removed, and all food outlets were geocoded at the point address level using ArcGIS and a US street network data set. As outlined and defined in Table 1, outlets were classified as grocery stores/supermarkets (n 80), convenience stores (n 248), discount or drug stores (n 67), fast-food restaurants (n 368) or fast casual restaurants (n 349). Grocery stores and supermarkets usually contain diverse options for fruits and vegetables and other healthy options and have been positively associated with healthy eating and healthy weight status in youth^{(Reference Caspi, Sorensen and Subramanian14,Reference Powell, Slater and Mirtcheva32,Reference Liu, Wilson and Qi52)} . Convenience and discount or drug stores are usually characterised as stocking high-fat, sugary, take-away or snack food, and other unhealthy food options and are open longer hours compared with other food outlets^{(Reference Xin, Zhao and Wu53)}. Fast food and fast casual outlets are characterised as supplying energy-dense and nutrient-poor food options that are cheap and highly convenient^{(Reference Burgoine, Sarkar and Webster54,Reference Schoffman, Davidson and Hales55)} .

Measures

Block group characteristics

All block groups in the study setting were categorised as either rural, mixed or urban. To accomplish this, using GIS, we overlaid all study block groups with US Census Bureau data for urbanised areas (50 000 or more people) and urban clusters (2500–50 000 people)⁽⁵⁶⁾. Urban block groups were those located entirely within areas of 50 000 or more people. Mixed block groups were those that were entirely within areas of 2500–50 000 people or overlapping areas of both 2500–50 000 people and 50 000 or more people. Finally, rural block groups were all others not classified as urban or mixed^{(Reference Hughey, Kaczynski and Porter57,Reference Stowe, Hughey and Hallum58)} .

To characterise the racial/ethnic composition of each block group, the percentage of population that was non-White was calculated from American Community Survey 5-year estimates (2009–2013).

Finally, data from American Community Survey 5-year estimates (2008–2012) were used to construct a composite SED index for each block group that served as the independent variable⁽⁴⁶⁾. Four key economic and education variables were selected based on prior research, including the percentage of the population that was unemployed, percentage with less than a high school education, percentage of renter occupied housing and percentage of the population under 125 % of the federal poverty threshold^{(Reference Kirby and Kaneda48)}. These variables were subjected to a principal components analysis with oblique rotation and all loaded strongly (>0·50) on a single factor^{(Reference Hughey, Walsemann and Child45)}. Each variable was then standardised to ensure a similar scale and equal weighting prior to summing them together to create the SED index value for each block group^{(Reference Kirby and Kaneda48,Reference Turney and Harknett59)} . Figure 1 provides an illustration of the study area by SED and food environment.

Fig. 1 Food environment by socio-economic disadvantage

Food outlet density surfaces

A kernel density analysis was conducted for convenience stores, discount and drug stores, fast casual restaurants, fast-food restaurants and grocery stores using ArcGIS. Kernel density diffuses the impact of an environmental feature across a specified distance and creates a raster surface representing 100 m × 100 m cells with varying weights^{(Reference Hughey, Kaczynski and Porter57)}. For the four types of less healthy food outlets, a standard 1-mile (1·6 km) buffer was applied around each establishment^{(Reference Lamichhane, Puett and Porter60,Reference Gamba, Schuchter and Rutt61)} , while for grocery stores, a 3-mile (4·8 km) buffer was applied because these outlets have greater population reach and a recent study demonstrated that the average distance to a grocery store for youth in our study area was 2·9 miles^{(Reference Hughey, Kaczynski and Porter57,Reference Lamichhane, Puett and Porter60)} . A separate surface was created for each type of food outlet by aggregating the scores for each cell (e.g. a cell might be impacted by multiple fast food restaurants at varying distances). Each food outlet surface was then normalised such that 0 represented a cell with no access to the food outlet and 1 represented the cell with the greatest access to the food outlet^{(Reference Hughey, Kaczynski and Porter57)}. To match the unit of analysis, a block group score for each type of food outlet was created by calculating the average density score for all cells within the block group.

The four unhealthy food environment layers – convenience stores, discount and drug stores, fast-food restaurants and fast casual restaurants – were aggregated using ArcGIS to create a composite unhealthy food environment index. The healthy food environment was represented by the grocery store layer.

It has been shown that the relative density of food outlet types is a better predictor of fruit and vegetable intake than counts of healthy and less healthy food outlets.^{(Reference Clary, Ramos and Shareck41,Reference Pinho, Mackenbach and Oppert62)} . Thus, a variation of the Modified Retail Food Environment Index, which accounts for relative densities of food establishments, was used to create a total food environment index^{(Reference Ahalya, Polsky Jane and Éric63)}. The Modified Retail Food Environment Index produces a ratio of unhealthy food outlets to the total number of food outlets, per area^{(Reference Ahalya, Polsky Jane and Éric63)}. For this study, the raster calculator was used to divide the kernel density of unhealthy food outlets by the total density of food outlets (kernel density of unhealthy outlets plus the kernel density of grocery stores), all times one hundred. A greater score on the total food environment index indicates that a larger proportion of food outlets are unhealthy. All three indices – healthy, unhealthy and total – were calculated at the block group level by finding the mean value for all cells within the block group using the Zonal Statistics tool in ArcGIS.

Analysis

Multivariate linear regression was used to examine differences in food environments according to level of SED. Separate models were conducted to test the relationship between SED and the density of each type of food outlet. To examine the relationship between SED and the composite food environment, three separate models explored the association between SED and the unhealthy food environment index, healthy food environment index and total food environment index. All analyses were conducted in SPSS 25.0 and adjusted for urbanicity and the percentage of the population considered a racial/ethnic minority.

Results

Block group characteristics are displayed in Table 2. Table 3 reports relationships between SED and the four unhealthy food outlet layers. There was a positive association between neighbourhood SED and all types of unhealthy food outlets, such that higher block group disadvantage was associated with greater convenience store density (R ² = 0·34, β = 0·02, P < 0·01), discount store density (R ² = 0·07, β = 0·01, P = 0·02), density of fast casual restaurants (R ² = 0·05, β = 0·01, P = 0·04) and density of fast food restaurants (R ² = 0·05, β = 0·01, P = 0·02).

Table 2 Block group descriptive statistics (n 255)

Table 3 Relationships between neighbourhood socio-economic disadvantage and food outlet density (n 255)

Table 3 also reports relationships between SED and the composite unhealthy, healthy and total food environment indices. There was a positive association between SED and the density of unhealthy food outlets (R ² = 0·15, β = 0·01, P < 0·01); as block group disadvantage increased, so too did the overall density of unhealthy establishments. Likewise, there was a positive association (R ² = 0·22, β = 0·04, P < 0·01) between SED and the density of healthy food outlets (i.e. grocery stores); as block group disadvantage increased, so too did the density of grocery stores. Finally, there was a negative association between SED and the total food environment (R ² = 0·04, β = –1·74, P = 0·02); as block group SED increased, the total food environment index (as a ratio of unhealthy to total food outlets) decreased.