Development of a neighborhood obesogenic built environment characteristics index for the Netherlands

Abstract

Objective

Environmental factors that drive obesity are often studied individually, whereas obesogenic environments are likely to consist of multiple factors from food and physical activity (PA) environments. This study aimed to compose and describe a comprehensive, theory‐based, expert‐informed index to quantify obesogenicity for all neighborhoods in the Netherlands.

Methods

The Obesogenic Built Environment CharacterisTics (OBCT) index consists of 17 components. The index was calculated as an average of componential scores across both food and PA environments and was scaled from 0 to 100. The index was visualized and summarized with sensitivity analysis for weighting methods.

Results

The OBCT index for all 12,821 neighborhoods was right‐skewed, with a median of 44.6 (IQR = 10.1). Obesogenicity was lower in more urbanized neighborhoods except for the extremely urbanized neighborhoods (>2500 addresses/km²), where obesogenicity was highest. The overall OBCT index score was moderately correlated with the food environment (Spearman ρ = 0.55, p <0.05) and with the PA environment (ρ = 0.38, p <0.05). Hierarchical weighting increased index correlations with the PA environment but decreased correlations with the food environment.

Conclusions

The novel OBCT index and its comprehensive environmental scores are potentially useful tools to quantify obesogenicity of neighborhoods.

Study Importance.

What is already known?

• Built environmental characteristics related to obesity are often studied individually, whereas real‐life environments combine a wide range of obesogenic exposures.

• Considering environmental characteristics simultaneously, in an index, could improve the quantification of the obesogenic environment, as shown earlier by walkability studies.

What does this study add?

• The Neighborhood Obesogenic Built Environment CharacterisTics (OBCT) index is conceptualized in terms of both food (density of five food outlet types) and physical activity environments (walkability, drivability, bikeability, and sports‐facility density).

• Neighborhoods in the Netherlands contrast considerably in terms of obesogenicity.

How might these results change the direction of research?

• The ability to characterize the obesogenicity of neighborhoods through the OBCT index can be used in further epidemiological research and has potential applications in monitoring and benchmarking for policy making.

INTRODUCTION

Obesity is a chronic health condition resulting from a prolonged imbalance of energy intake and energy expenditure, causing metabolic, hormonal, and inflammatory reactions in various critical organs [1]. Overweight and obesity have been associated with several chronic comorbidities such as insulin resistance, type 2 diabetes, hypertension, and dyslipidemia [2, 3] and have been ranked as the fifth most common cause of mortality worldwide [4]. In the Netherlands, about half of the adult population has overweight, a third of which has obesity [5].

Obesity has multiple causal pathways in which genetic heritability and the environment play equally important roles [6]. Whereas genetic factors are unchangeable throughout the lifetime, the modifiable nature of the environment provides leverage points for population prevention approaches. The built environment encompasses a person's immediate activity space and therefore it may influence obesity‐related behaviors, including physical activity (PA) and sedentary behavior through, e.g., passive transport options and dietary behavior through, e.g., the availability of unhealthy food [7, 8].

To date, research has shown that most built environment components are largely inconsistently associated with weight, but some components are consistently associated with increased weight, such as a higher fast‐food‐outlet density, lower urbanization degree, lesser land‐use mix, more urban sprawl, and lower green space density [9, 10]. However, most studies thus far have examined only one exposure at a time. A combined exposure approach is preferable because it would potentially yield higher explanatory power and a more realistic estimate of one's exposure to the surrounding environment as well as account for interactions between these exposures [11, 12, 13].

Currently, exposures in studies of the built environment have been aggregated only to a certain extent. A notable example of an aggregated measure is walkability, a composite measure of how conducive neighborhoods are to walking, which typically combines land‐use mix, population density, street connectivity, and green space. Despite inherent heterogeneity in study designs, higher neighborhood walkability has been associated with more walking [14] and lower obesity in both adults and children [11, 15]. Even so, more extensive integration of environmental components is desirable to better capture the complex variety of exposures. In the context of obesity, despite the interdependency between dietary intake and PA [16, 17], studies that combine the food and PA environments have so far been scarce [18]. To our knowledge, Kaczynski and colleagues [19] were the first, and so far only, authors who attempted to construct an obesogenic index on the national level. Their childhood obesogenic environment index (COEI) was developed to quantify county‐level (n = 3142) obesogenicity for children in the United States and was composed of 10 components of the food (density of grocery stores, farmers markets, fast‐food restaurants, full‐service restaurants, and convenience stores; percentage of birth at baby‐friendly facilities) and PA environments (percentage of population living close to exercise facilities, percentage of population living close to a school, walkability, violent crime percentage) [19]. Whereas the COEI was reasonably comprehensive, it suffered from an important shortcoming: the spatial resolution was very coarse (county level), hampering epidemiological analysis at the personal level. Furthermore, the COEI was specifically developed for children; therefore, such an index is still missing for the adult population, whose environmental context requires different components.

Therefore, our aim was to comprehensively quantify the obesogenicity of all neighborhoods at high resolution in the Netherlands using a composite indicator approach. The resulting index, henceforth termed as the Obesogenic Built CharacterisTics (OBCT) index, could improve our understanding of obesogenic neighborhoods and their relationship with downstream energy balance‐related behaviors and health outcomes in Dutch adults. From a policy point of view, this index would enable contrasting, monitoring, and benchmarking across neighborhoods and it can highlight areas to prioritize in health planning at both the regional and national level.

METHODS

Development of neighborhood obesogenic environment index

We followed the 10‐step procedure recommended in the Handbook on Constructing Composite Indicators [20] to create the OBCT index. The 10 steps are listed as follows:

Step 1: Theoretical framework	Step 6: Weighting and aggregation
Step 2: Variable selection	Step 7: Sensitivity analysis
Step 3: Imputation of missing data	Step 8: Back to the details
Step 4: Multivariate analysis	Step 9: Links to other indicators
Step 5: Normalization of data	Step 10: Visualization of the results

We began with a framework to structure our understanding of the built environment pertaining to overweight and obesity (Step 1: Theoretical framework). Using a priori knowledge from socioecological models of health behavior by Swinburn et al. [21] and Frank et al. [7], we conducted an umbrella review of systematic reviews focusing on the associations between the built environment components and obesity. From this systematic search, we yielded the following eligible constructs: fast food, rural urban disparity, urban sprawl, and walkability [10]. Additionally, we consulted national experts in the fields of spatial epidemiology, obesity, lifestyle behaviors, and health geography for inclusion of potentially relevant constructs for the OBCT index, such as obesogenic food outlets [22], drivability [23], and sports‐facility density [24]. Even though neighborhood socioeconomic status (SES) is an established determinant of obesity, we decided to exclude this variable from the framework and focus only on the physical built environmental characteristics. So far, literature has arguably treated neighborhood SES as a component of the social environment together with social capital, collective efficacy, and crime rate [25] and has mostly used this variable as confounder or effect modifier in the relationship between the built environment and obesity [10].

In Figure 1, we outline the framework, with constructs (in yellow) theoretically linking the neighborhood built environment to overweight and obesity. Based on this framework, we deciphered the 17 components making up these different constructs from literature and based on data availability (Step 2: Variable selection). Therefore, each construct could include one or more components. Table 1 outlines the components and their brief descriptions. For each of these components, geographical data from various sources were gathered and operationalized for the components by the Geoscience and hEalth Cohort COnsortium (GECCO) [26, 27], and data are further detailed in online Supporting Information. All components were assessed based on administrative neighborhood borderlines of the year 2016, and data on components were also collected in or close to 2016, with a maximum mismatch of 4 years.

FIGURE 1.

Theoretical framework for associations between built environment components and weight status. Under each environment are constructs (in yellow) that comprise one or more components. The red blocks indicate negative associations: the higher the value, the less obesogenic. The green blocks indicate positive associations: the higher the value, the more obesogenic [Color figure can be viewed at wileyonlinelibrary.com]

TABLE 1.

Summary of components included in the OBCT index: operationalization, source, and year of data collection

	Measure	Source and year
Fast food	Kernel density, 1 km search radius	Locatus, 2016
Local food stores
Restaurants
Supermarkets
Food deliveries	Point density, 4 km circular radius
Sports facilities	Point density, 1 km radius	Mulier Institute, 2017
Parking pressure	Ratio between parking capacity and household car ownership	BGT 2019, TOP10NL 2019, CBS
Highways	Distance to the nearest A‐/N‐highway entrance/exit	CBS, 2016
Job density	Time travel to the nearest 10,000 jobs	SPINlab, LISA, 2018
Land‐use mix	Entropy index of four main land‐use classes: 1) industrial, commercial, public, military, and private; 2) residential; 3) urban green; and 4) sports and leisure	TOP10NL, 2015
Population density	Number of inhabitants/km2	CBS, 2015
Short public transport	Point density of short‐range transports (buses, trams, metros, ferries)	NDOV, Geodienst Rijksuniversiteit Groningen, 2018
Long public transport	Distance to the nearest train station
Sidewalk density	Density of sidewalks, pedestrian stairs, and residential areas suitable for walking	BGT, 2019
Green space	Density of parks, forests, and graveyards	TOP10NL, 2015
Intersection density	Ratio of true intersections (>3 legs) and area size	TOP10NL, 2019
Bicycle pathway	Percentage coverage of bicycle paths per neighborhood	TOP10NL, 2019

Note: OBCT is Obesogenic Built Environment CharacterisTics, Locatus is a commercial retail information provider, and BGT is the key register for large‐scale topography of the Netherlands, which includes TOP10NL, the digital national topographic map. CBS is Statistics Netherlands, SPINlab is the Free University Amsterdam's Spatial Information Laboratory, LISA is National Information System of Workplaces, NDOV is the national service for Dutch transit information by Foundation OpenGeo, and Geodienst Rijksuniversiteit Groningen is the spatial expertise center of the University of Groningen.

Step 3 (Imputation of missing data) was omitted because the components were selected on the basis of completeness. In Step 4 (Multivariate analysis), components with negative association with obesogenicity were reverse scored so that a higher score corresponded to higher obesogenicity, as theorized in Figure 1. Pairwise Spearman correlations between components were assessed and presented in a correlation matrix form. In Step 5 (Normalization of data), all components were z‐standardized to have a median of zero and mean absolute deviation of one to achieve a homogeneous scale. We decided to use median and mean absolute deviation instead of mean and standard deviation because most components were not normally distributed, according to recommendations from Leys et al. [28]. Fifth and ninety‐fifth percentile Winsorization was applied to all components to reduce the influence of outliers on the index score. Winsorization refers to the practice of either 1) assigning caps for extreme values, such that any observations below or above the i ^th percentile would receive the i ^th ‐percentile value (commonly capping values are 5th and 95th or 1st and 99th percentiles of the distribution) or 2) discounting or assigning lower weights to outliers [29].

In Step 6 (Weighting and aggregation), we composed the environmental scores by averaging the z scores of all the components in respective environments (see the following formula). We calculated the main OBCT index (denoted as OBCT _m ) by linear aggregation, taking the arithmetic mean of the food and PA environment scores. The index was then minimum‐maximum normalized so that neighborhoods would range from 0 to 100, with a higher score indicating higher levels of obesogenicity. We reported descriptive statistics for this main index, including mean, median, and interquartile range (IQR).

$$\text{Food environment}=(Z_{\text{fast food}}+Z_{\text{restaurants}}+Z_{\text{food deliveries}}+Z_{\text{supermarkets}}+Z_{\text{local food stores}})/5$$

$$\text{Physical activity environment}=\left(\begin{array}{c}{z}_{\text{sport facilities density}}+z_{\text{parking pressure}}+z_{\text{distance highway}}+z_{\mathrm{job}\text{density}}\\ +z_{\text{public transport}}+z_{\text{distance to train station}}+z_{\text{sidewalk density}}+z_{\text{green space density}}\\ +z_{\text{street connectivity}}+z_{\text{bike path density}}+z_{\text{population density}}+z_{\text{land}\mathrm{use}\mathrm{mix}}\end{array}\right)/12$$

$${\text{OBCT}}_m=\frac{\left(\text{Food environment}+\text{Physical activity environment}\right)}{2}$$

In Step 7 (Sensitivity analysis), we tested whether the index was sensitive to another weighting method. In this “hierarchical” weight set, we applied equal weighting to the components under a construct and equal weighting between constructs (obesogenic food, sports‐facility density, walkability, drivability, and bikeability). The resulting equation for this hierarchical index (denoted as OBCT _h ) is shown in the following formula. Sensitivity of ranking was assessed by how OBCT scores using these different methods were distributed and correlated using Spearman correlation.

$$\text{Driveability}=\raisebox{1ex}{$\left(\begin{array}{c}{z}_{\text{parking pressure}}+z_{\text{distance highway}}+z_{\mathrm{job}\text{density}}+z_{\text{land}\mathrm{use}\mathrm{mix}}\\ +z_{\text{population density}}+z_{\text{public transport}}+z_{\text{distance to train station}}\end{array}\right)$}\!\left/ \!\raisebox{-1ex}{$7$}\right.$$

$$\text{Walkability}=\raisebox{1ex}{$\left(\begin{array}{c}{z}_{\text{land}\mathrm{use}\mathrm{mix}}+z_{\text{population density}}+z_{\text{public transport}}+z_{\text{distance to train station}}\\ +z_{\text{sidewalk density}}+z_{\text{green space density}}+z_{\text{street connectivity}}\end{array}\right)$}\!\left/ \!\raisebox{-1ex}{$7$}\right.$$

$$\text{Bikeability}=\raisebox{1ex}{$\left(z_{\text{street connectivity}}+z_{\text{bike path density}}+z_{\text{green space}}\right)$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$$

$$\text{Sport facilities density}=z_{\text{sport facilities density}}$$

$$\text{Obesogenic food}=\begin{array}{l}(Z_{\text{fast food}}+Z_{\text{restaurants}}+Z_{\text{food deliveries}}+Z_{\text{supermarkets}}\\ +Z_{\text{local food stores}})/5\end{array}$$

$${\text{OBCT}}_h=\left(\frac{\text{Obesogenic food}+\text{Sport}+\text{Driveability}+\text{Walkability}+\text{Bikeability}}{5}\right)$$

In Step 8 (Back to the details), we stratified the main index score by urbanization degrees, Dutch provinces, and neighborhood SES. Nonparametric tests (Mann–Whitney U and Kruskal‐Wallis [K‐W] H) were carried out to compare the scores between urbanization and SES strata. We also checked the correlations between the environmental scores and the resulting indices to identify whether any component was underrepresented in each index score.

Because there was no earlier version of an obesogenic index, we substituted Step 9 (Links to other indicators) with an examination of the food and PA environment scores separately to identify any remarkable patterns in distribution. In Step 10 (Visualization of the results), the resulting OBCT index was visualized using heat maps depicting overall neighborhood obesogenicity, as well as in terms of food and PA environment scores. We also compared these obesogenic environment maps with overweight and obesity prevalence maps of the Netherlands.

All OBCT components were processed in ArcMap version 10.8.1 (Esri, Redlands, California), and the OBCT index was calculated, analyzed, and drawn in RStudio 1.3.959 (RStudio Team, Boston, Massachusetts).

RESULTS

The OBCT index was calculated for all 12,821 administrative neighborhoods in the Netherlands. For the main (OBCT _m ) index score, the most (score = 100) obesogenic neighborhood was located in Leiden (Pesthuiswijk), and the least (score = 0) obesogenic neighborhood was located in Harderwijk (Stadsdennen‐Zuidwest). The index was slightly right‐skewed (Supporting Information Figure S2), and the median index value across the country was 44.6 (IQR = 10.1; Table 2).

TABLE 2.

Neighborhood obesogenic index stratified by urbanization degrees (N = 12,766) and neighborhood socioeconomic status score (N = 12,817) in 2016

Variable	Stratum values	N	Mean score	Median score	Range	IQR
Urbanization degree	>2500 addresses/km2	1541	49.8	51.4	4.3‐99.9	23.9
	1500‐2500 addresses/km2	2088	37.7	36.9	0.0‐100	14.7
	1000‐1500 addresses/km2	1581	37.9	37.1	12.9‐76.9	11.2
	500‐1000 addresses/km2	1877	41.4	41.3	6.1‐72.8	9.2
	<500 addresses/km2	5679	46.0	46.4	27.2‐70.6	4.3
	Missing	55	46.9	46.6	28.6‐84.8	4.8
Neighborhoood socioeconomic status	Low	3204	42.1	43.1	0.0‐100	12.6
	Medium	3204	43.4	44.9	0.5‐86.8	9.0
	High	3204	43.4	44.7	7.1‐85.7	8.8
	Very high	3205	44.9	45.5	1.4‐99.9	9.8
	Missing	4	44.9	46.6	38.4‐48.2	4.6
All neighborhoods		12,821	43.4	44.6	0.0‐100	10.1

.

Stratified analysis showed significant difference in obesogenicity between all urbanization degrees (K‐W H was 2612.9, df = 4, p <0.05). The most rural neighborhoods (less than 500 addresses/km²) had a median of 46.4 (IQR = 4.26), and the most urbanized (more than 2500 addresses/km²) had a median of 51.4 (IQR = 23.9; Table 2); the distributions in the two groups differed significantly (Mann–Whitney U = 52,113,960; n ₁ = 5679, n ₂ = 1541, p <0.05). There was significant difference in obesogenicity between all neighborhood SES quartiles (K‐W H was 139.75, df = 3, p <0.05); however, these differences were small (Table 2). When examined across the 12 Dutch provinces, obesogenicity was comparable, but the distributions of obesogenicity differed significantly between the provinces (K‐W H was 385.98, df = 11, p <0.05). However, there was a large disparity in obesogenicity between major cities of each province. The densely populated Randstad, a metropolitan conglomeration of four major Dutch cities, had markedly higher obesogenicity than less densely populated cities in the north or east, such as Assen, Arnhem, or Nijmegen (Table 3, Figure 4).

TABLE 3.

Neighborhood obesogenic index value by Dutch provinces (bolded) and their capital city ^a and most populated city ^b (N = 12,821)

Province	N	Mean score	Median score	Range	IQR
Friesland	808	45.6	46.9	15.3‐66.0	5.1
Leeuwarden a , b	96	47.1	48.5	15.3‐66.0	11.9
Noord‐Holland	1795	45.1	45.2	3.9‐86.8	16.4
Haarlem a	111	45.8	44.0	18.9‐78.3	15.4
Amsterdam b	478	55.7	60.6	14.6‐86.8	18.3
Drenthe	663	44.7	46.5	18.1‐64.5	5.6
Assen a	109	40.8	41.9	18.7‐64.5	13.9
Emmen b	83	45.3	46.8	25.4‐54.5	5.2
Zeeland	395	44.3	45.8	18.5‐73.5	6.6
Middelburg a	24	42.2	43.4	31.3‐58.4	10.3
Terneuzen b	41	44.5	47.3	23.5‐71.5	10.7
Groningen	595	45.7	46.0	17.9‐99.9	6.3
Groningen a , b	105	54.3	53.5	25.8‐99.9	14.3
Limburg	901	43.8	44.6	17.7‐82.3	6.8
Maastricht a , b	44	48.1	49.2	24.6‐74.1	11.4
Noord‐Brabant	1649	43.1	44.1	9.7‐83.6	8.33
Den Bosch a	107	45.9	46.4	13.8‐75.3	12.5
Eindhoven b	116	48.3	45.5	26.3‐83.6	15.3
Overijssel	1057	42.2	43.4	9.08‐87.9	9.3
Zwolle a	78	47.0	46.4	9.08‐87.9	13.8
Enschede b	70	48.2	48.2	29.0‐81.4	12.1
Gelderland	1608	41.1	42.9	0.0‐79.4	9.8
Arnhem a	83	41.9	39.1	18.2‐79.4	17.1
Nijmegen b	44	39.7	37.5	24.9‐64.6	15.1
Utrecht	857	43.0	43.4	12.8‐87.1	12.2
Utrecht a , b	111	50.7	51.1	24.1‐70.8	20.2
Flevoland	318	41.7	42.7	16.6‐71.4	11.2
Lelystad a	139	42.4	42.0	20.9‐71.4	11.0
Almere b	66	42.6	43.4	27.7‐56.1	9.0
Zuid‐Holland	2175	42.9	44.0	0.5‐100	14.0
Den Haag a	114	50.7	51.3	8.6‐83.5	23.1
Rotterdam b	92	54.3	53.9	31.2‐82.8	15.8
All neighborhoods	12,821	43.4	44.6	0.0‐100	10.1

^a Capital city.

^b Most populated city.

FIGURE 4.

OBCT index score for all neighborhoods in the Netherlands (left) and the municipality of Amsterdam (right) in 2016. Darker red indicated higher obesogenicity. The score is graphically presented in deciles. OBCT, Obesogenic Built Environment CharacterisTics [Color figure can be viewed at wileyonlinelibrary.com]

In terms of environments, the most obesogenic food environment seemed to be concentrated around the cities within Randstad (Figure 2, Table 3). On the contrary, obesogenic PA environments were more present in less populated regions such as the agricultural, less populated areas in the north and south (Figure 3). This was evident in the negative, albeit weak, correlation between the two environments (ρ = −0.37, p <0.05). In terms of distribution, the food environment was right‐skewed (Supporting Information Figure S3), whereas the PA environment was lightly left‐skewed (Supporting Information Figure S4). The OBCT _m index score was more correlated with the food environment (ρ = 0.55, p <0.05) than with the PA environment (ρ = 0.38, p <0.05, Figure 5). Visually, the obesogenic PA environment score (Figure 3) seemed to resemble overweight prevalence (Supporting Information Figure S1A) and obesity prevalence (Supporting Information Figure S1B) more than the food environment.

FIGURE 2.

Obesogenic food environment score for all neighborhoods in the Netherlands (left) and the municipality of Amsterdam (right) in 2016. The higher the food environment score (and darker red), the more obesogenic the neighborhood. The score is graphically presented in deciles. OBCT, Obesogenic Built Environment CharacterisTics [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 3.

Obesogenic PA environment score for all neighborhoods in the Netherlands (left) and the municipality of Amsterdam (right) in 2016. The higher the PA environment score (and darker red), the more obesogenic the neighborhood. The score is graphically presented in deciles. OBCT, Obesogenic Built Environment CharacterisTics; PA, physical activity [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 5.

Spearman correlation matrix of OBCT index components. Empty cells denote nonsignificant correlation (p > 0.05). OBCT, Obesogenic Built Environment CharacterisTics index; OBCTm, main index; OBCTh, hierarchical index, PA physical activity [Color figure can be viewed at wileyonlinelibrary.com]

In terms of component correlations, components of the PA environment often clustered together and they were moderately‐to‐strongly correlated among each other, especially among job density, density of sports facilities, short‐distance public transport, bicycle pathway density, parking pressures, street connectivity, sidewalk density, and population density (Spearman ρ range from +/− 0.55 to +/− 0.86). The most strongly correlated pair was sidewalk density and population density (ρ = 0.86). Green space and land‐use mix were least correlated with other components (highest value of ρ = 0.31, Figure 5). The food environment measures were also moderately‐to‐strongly correlated with each other (ρ range from 0.45 to 0.85), especially between fast‐food outlets, supermarkets, and local food stores. Sensitivity analysis showed that hierarchical weighting (OBCT _h ) increased the index's correlation with the PA environment (ρ = 0.92, p <0.05) at the expense of the food environment (ρ = −0.16, p <0.05). Correlation between OBCT _m and OBCT _h was 0.57 (p <0.05, Figure 5).

DISCUSSION

In this study, we used an evidence‐based, expert‐informed theoretical framework to compose a comprehensive, high‐resolution index that quantified the obesogenicity of neighborhoods in the Netherlands. The resulting OBCT index score for all 12,821 administrative neighborhoods in the Netherlands in 2016 showed considerable variation in obesogenicity of neighborhoods. It also highlighted that the scores were unevenly distributed over the provinces and urbanization degrees.

The characterization of neighborhoods in terms of their obesity‐driving nature was a pioneering attempt. Previous Dutch studies have attempted to assess the obesogenicity of the environment in terms of exercise‐friendliness [30] combined with air pollution in the form of a Healthy Living Environment Index [31]; in terms of density and healthiness of the food environment [22]; or focusing on single regions across Europe, including some in the Netherlands [32]. The exercise‐friendly environments have typically included sports facilities such as sports halls, gyms, outdoor playgrounds, and (sports) parks; infrastructures for active transport such as bicycle, walking, or horse‐riding routes; and nature areas such as forests, heaths, dunes, and beaches, as well as proximity to public facilities [30]. The food environment studies have focused on density and accessibility of food retailers [33, 34, 35], especially fast food [36, 37, 38, 39]. To our knowledge, this study is the first to assess obesogenicity of neighborhoods by combining most of these components using objectively measured high‐resolution geographical data.

Similar to the American COEI by Kaczynski et al. [19], we used an a priori approach to select relevant environmental components based on conceptual frameworks, current literature, and data availability. However, the COEI was constructed at the county level, which has an average size of 3130.00 km² compared with average neighborhood size of 2.62 km² in our current study. Nevertheless, the COEI highlighted some geographical patterns, including consistently higher obesogenicity in the Southern US counties compared with Northeastern US counties. Rural American counties were more obesogenic compared with the metropolitan and micropolitan counties, which somewhat reflected our results, except for the extremely urbanized Dutch neighborhoods. However, in the Dutch context, we observed higher median SES scores with increased rurality, and neighborhoods across SES quartiles only differed very slightly in terms of obesogenicity. The underlying association between obesogenicity and area‐level SES is thereby dissimilar between the countries.

A recently developed index similar to the index of the current study is the healthy location index (HLI), which was developed for New Zealand [40]. Even though this index's focus was on general health and not specifically obesogenicity, there was some shared methodology and observations. First, HLI also has high spatial resolution (average mesh‐block size of 5.06 km²), and it is a combination of access to both healthy and unhealthy destinations. However, similar to the COEI, HLI generates a ranking‐based index as opposed to using continuous measures in our OBCT index. Nevertheless, the HLI highlighted the co‐location of both healthy and unhealthy environmental features, such that associations between the index and area‐level deprivation (SES) are also nuanced, similar to what we found in the OBCT index score. Interestingly, a reverse pattern across urbanization was observed: whereas the most urban and rural strata in our analysis had higher obesogenicity compared with the middle strata, neighborhoods in the middle urban strata in New Zealand tended to be less healthy.

Our OBCT index also highlighted some interesting domestic observations. First, major cities of the Netherlands contain both obesogenic and leptogenic neighborhoods. This is not surprising given the mixed functions of neighborhoods in major cities. The most obesogenic neighborhoods a found in the city centers where tourists most frequently visit or where commercial centers are located. For this reason, the streets of these neighborhoods are densely occupied by fast‐food outlets, which result in high obesogenic scores. This raises an interesting research question of how relevant this highly obesogenic food environment is for the people living in or near the city centers, namely whether these “touristy” food outlets are also regularly visited by local residents. Moreover, these neighborhoods with an obesogenic food environment are compensated by greater opportunities for PA such as biking, walking, and sports, as well as discouraged driving. Vice versa, rural areas with less obesogenic food outlets also tend to have fewer PA activities, which render the associations with overweight and obesity in a nuanced way.

Second, the obesogenic PA environment score (Figure 3) seems to visually resemble the overweight or obesity prevalence (Supporting Information Figure S1A‐S1B) better than that of the food environment (Figure 2). However, it should be noted that these comparisons do not take confounders into account, such as sociodemographic composition of neighborhood residents. Nevertheless, this proves the challenging nature of measuring the (obesogenic) food environment, in which heterogenous operationalizations could potentially cause inconsistent associations with health outcomes. To date, the most consistent association between the food environment and cardiometabolic health outcomes in the Netherlands was between presence of fast‐food outlets and diabetes incidence [41]. However, this dichotomous measure was not compatible with other continuously measured components for the OBCT index.

Third, the overall OBCT index is sensitive to different methodological operationalization choices, especially the weighting methods applied to its components and constructs. Spearman correlation between the food environment and main OBCT index was higher than that between the PA environment and OBCT. Slight adjustment to the weight by weighting components within a construct changed the magnitude of associations with both environments and, potentially, with health outcome. Especially with the PA environment score visually resembling overweight prevalence, higher weights could be assigned for PA environment components in association to better quantify obesogenicity of the environment.

Our attempt to quantify obesogenicity of neighborhoods has many merits. We used high‐resolution, objectively measured geographical data for the construction of the index components. Compared with earlier studies, this high resolution allowed us to detect more subtle differences within a region or a city, thereby drawing more nuanced observations about obesogenicity of the immediate environment. Novel constructs such as drivability, bikeability, and sports‐facility density have shown to be of added value, which was reflected visually in the PA environment's mirroring of overweight prevalence map, even though further statistical examination is warranted. Beyond walkability, these novel constructs might play an important yet understudied role in the study of obesogenicity of the environment. For example, whereas walkability highlights the attractiveness of urban areas for walking [14], the OBCT index integrates the food environment in the same areas and therefore provides a more balanced picture of the built environment. In terms of components, we composed a measure for the food environment that encompassed both healthy and unhealthy outlets that are frequently visited. Finally, we were able to characterize all neighborhoods across the entire country, thereby increasing the spatial heterogeneity observed.

However, this study also has important limitations. First, some aspects of the food and PA environments that are potentially relevant for weight were not included in the index owing to data unavailability. These components include food and transport affordability (costs) and street aesthetics [42, 43, 44]. Furthermore, two theorized components were excluded owing to the lack of contrast in exposures (retail and service density and percentage of paid parking). Another limitation related to data availability was a mismatch in year. Not all of the components included in this study were available for 2016; therefore, an earlier or later data set was sometimes used. Whereas some components such as the food environment have a relatively high temporal variability [34], others, including walkability, are relatively stable over time [45]. Moreover, the index suffers from limitations inherent in the index‐making process: subjective choices are made in the processes of data selection, cleaning, weighting, and aggregation. The most‐discussed so far in this study was weighting: the equal weighting presented in the main index is simple and commonly used, and it may not entirely capture the relative importance between index components.

To proceed with this index, first, a formal validation is needed. We intend to study the construct validity of the environmental scores and the overall index using behavioral and health outcomes measured in nationwide cohorts and administrative (surveillance) data. In particular, each environment can be associated with intermediate outcomes such as self‐reported dietary behaviors, PA, or transport patterns. Overall, the index's association with weight outcomes such as body mass index or central obesity can also be studied as validation. Nevertheless, this preliminary result shows that quantifying the food environment and its associations with dietary behaviors and weight‐related outcomes remain challenging.

Second, an important aspect for further examination is the role of weighting methods in determining association with weight outcomes. The Organisation for Economic Co‐operation and Development (OECD) Handbook on Constructing Composite Indicators suggests a few different methods of assigning weights: subjective weights using expert opinion, budget allocation, or analytic hierarchy process; or data‐driven weights using outcomes such as body mass index or overweight prevalence [20]. From a behavioral discipline, weights could be assigned by the energy expenditure or intake from activities associated with each component. Nevertheless, these methods of weighting are highly contextual and might reduce the comparability of OBCT indices across different countries.

Once validated, the OBCT index could be used in observational studies to examine associations with further downstream health outcomes such as COVID‐19‐related hospitalization, hypertension, and cardiovascular disease mortality and morbidity. Furthermore, the OBCT index could be used as a tool to measure, monitor, and benchmark neighborhoods in terms of their potential influence on health. It could aid municipal and regional policy makers in directing their preventive efforts in specific neighborhoods in need of infrastructural changes to encourage healthy lifestyles and reduce disease burden. Moving forward, the OBCT index, which measures the physical, tangible components of the food and PA environments, could potentially be combined with other environments such as the economic, social, and policy environments (i.e., with components such as social capital, collective efficacy, and crime) to fully understand the complexity of neighborhood influence on obesity [25, 43].

CONCLUSION

We characterized the obesogenicity of all Dutch neighborhoods based on a wide variety of objectively measured environmental exposures. Upon validation, this OBCT index is a potentially useful tool to quantify obesogenicity of neighborhoods and it warrants further research and policy applications.

FUNDING INFORMATION

The Geoscience and hEalth Cohort COnsortium (GECCO) was financially supported by the Netherlands Organization for Scientific Research (NWO), the Netherlands Organization for Health Research and Development (ZonMw), and Amsterdam University Medical Centers.

CONFLICT OF INTEREST

Thao Minh Lam, Nicolette R. den Braver, and Erik J. Timmermans received funding from the Netherlands Organization for Scientific Research (NWO) under the Exposome‐NL consortium. Jeroen Lakerveld and Alfred J. Wagtendonk received funding from Netherlands Organization for Health Research and Development (ZonMw) to collect and process environmental data used in this study. The other authors declared no conflict of interest.

Supporting information

Appendix S1: Supplementary Information

Appendix S2: Supplementary Information

Lam TM, Wagtendonk AJ, den Braver NR, et al. Development of a neighborhood obesogenic built environment characteristics index for the Netherlands. Obesity (Silver Spring). 2023;31(1):214‐224. doi: 10.1002/oby.23610