Walkability, Overweight, and Obesity in Adults: A Systematic Review of Observational Studies

We conducted a systematic review to describe and summarize possible associations between the walkability index, overweight, and obesity. Systematic searches using seven electronic databases and reference lists were conducted to identify papers published until December 2017. Observational studies, describing associations using regression-based statistical methods, published in English and Portuguese, reporting markers of overweight and obesity, and involving adults (≥18 years) were included. Of the 2469 references initially retrieved, ten were used for the descriptive synthesis. Seven studies showed significant inverse associations between walkability and overweight and obesity, however, all were cross-sectional studies. High risk of bias scores were observed in “selection bias” and “withdrawals and dropouts”. All studies were published in high-income countries with sample sizes ranging among 75 to 649,513 participants. Weight and height as measures for determining BMI tended to be self-reported. Indicators of walkability, such as land-use mix, street connectivity and residential density were used as components of the indices. Based on this review, more studies should be conducted in low, middle, and middle-high income countries, using longitudinal designs that control neighborhood self-selection; other indicators of the neighborhood environment, such as food access, physical activity facilities, sidewalks, and safety and crime prevention should be considered.


Introduction
Overweight and obesity have risen dramatically over the last 40 years and they now constitute serious public health threats that are contributing to increasing rates of non-communicable disease such as type 2 diabetes and cardiovascular disease [1][2][3]. A recent publication gathering data from 239 prospective studies involving over 10 million individuals in four continents showed that the risk of mortality increases linearly with overweight (body mass index-BMI ≥ 25 kg/m 2 ) and mainly obesity (BMI ≥ 30 kg/m 2 ), regardless of age or sex [4]. A study of trends observed across 19 million participants shows a significant increase in BMI occurring over the last four decades, particularly in low-and middle-income countries. Between 1975 and 2014, the prevalence of age-standardized obesity increased 3.2% in 1975 to 10.8% in 2014 in men, and from 6.4 to 14.9% in women [5].

Methods
The protocol of this review was registered on PROSPERO database (CRD42017071830). Its report is based on PRISMA checklist items [23].
The following inclusion criteria were adopted: (1) Observational studies (e.g., cross-sectional, cohort, time-series or case-control); (2) describing associations using regression-based statistical methods; (3) published in English and Portuguese; (4) reporting on overweight and obesity markers (5) that used walkability indices; and (6) using samples of adults (≥18 years). Articles not addressing the concept of walkability or that used separate indicators, that did not use markers of overweight or obesity, studies with children or adolescents, and qualitative and protocol studies were excluded.
To retrieve potential references, seven electronic databases were used: Pubmed, Scielo, Lilacs, Web of Science, Scopus, Physical Education Index, and SportDiscus. Systematic searches were conducted in accordance with the strategy (Supplementary Materials S1: Search string) applied by Pubmed: ( ). Using Lilacs and Scielo, searches were made in Portuguese using other terms such "caminhabilidade" and "obesidade", "sobrepeso", "índice de massa corporal". Searches covered references available to December 2017. To avoid potential losses, manual searches through reference lists of the included studies were also performed.
Three researchers selected studies independently. Based on the systematic search results, titles and abstracts were screened, and then eligibility through a full text assessment was determined by all three researchers. Doubts and disagreements were resolved through a consensus meeting with a senior reviewer.
Data extraction involved the use of relevant information: Author (year), country/city/province/state (year of data collection), sampling, (n) sample, percentage of females, age (mean or range in years), and study type. Assessment of walkability variables: Geographic scale (e.g., census sectors, home addresses, post codes or geocoding buffers); geocoding of facilities; how walkability indicators were created and calculated. Assessments of overweight and obesity: How the markers were used (as categorical or dichotomous continuous variables); and the regression model used for analyses. Variables related to walkability, overweight, and obesity were extracted from included studies as significant or not significant associations at the 95% confidence interval or "p" values to better describe the odds ratio, beta coefficient, prevalence ratio, relative risks or other values with variables adjusted.
Risk of bias was evaluated with an adapted 12-item version, of the Effective Public Health Practice Project (EPHPP), proposed by Thomas et al. [24]. Other details of the assessment tool are given in Supplementary Materials S2. Original articles were assessed across seven methodological domains: (a) Selection bias (characteristics of the sample); (b) study design (information on the representativeness of the study, sampling methods); (c) confounders (control of relevant confounding factors of the analysis); (d) blinding (of assessors, outcomes and participants); (e) information on tools used to assess walkability and overweight/obesity (reports on previous validity and information, allowing for the reproducibility of the assessment of walkability and overweight/obesity); (f) withdrawals and dropouts (reported numbers and/or ratios and percentages of participants that completed the study); and (g) analysis (use of appropriate methods for analyses). The studies included were then classified by two previously trained researchers as presenting low, moderate or high levels of bias.

Results
Systematic searches retrieved 2469 potential articles. After the identification and removal of duplicate references (n = 116), 2353 references were screened by their titles and abstracts. Of these references, 69 were assessed from the full text. In view of the 59 rejected articles (for not addressing the main focuses of this review (n = 55) and for not reflecting qualitative and protocol studies (n = 4), 10 articles were used for the descriptive synthesis [15,[25][26][27][28][29][30][31][32][33] (Figure 1).
Of the three studies employing a longitudinal design, none showed significant results [25][26][27]. Most of the studies that found an inverse association between walkability and overweight and obesity were cross-sectional (n = 6) [15,[29][30][31][32][33]. The study that used time-series analysis presented significant results for the walkability index and a lower prevalence of obesity [28].
Of the three studies employing a longitudinal design, none showed significant results [25][26][27]. Most of the studies that found an inverse association between walkability and overweight and obesity were cross-sectional (n = 6) [15,[29][30][31][32][33]. The study that used time-series analysis presented significant results for the walkability index and a lower prevalence of obesity [28].

Identification
References retrieved by data base searches Different buffer areas around residents' homes were employed in two studies, e.g., a 1-km network buffer measured on the street network of each individual's geocoded residence [15,29], and one study employed network buffers of 800 and 1600 m [31].
Variables used to measure walkability included residential and population density (n = 9; 91.6%), street connectivity (n = 8; 83.3%), and land-use mix (n = 5; 58.3%). Five studies defined walkability as the sum of the scores of main indicators with the index divided into quintiles [28] or quartiles [15] or expressed as a continuous score [30,32].
Variables used to measure walkability included residential and population density (n = 9; 91.6%), street connectivity (n = 8; 83.3%), and land-use mix (n = 5; 58.3%). Five studies defined walkability as the sum of the scores of main indicators with the index divided into quintiles [28] or quartiles [15] or expressed as a continuous score [30,32].
Regarding other variables concerning walkability, only one study showed an inverse association between the proportion of inhabitants who walk to work and housing age (length of residence) to both men and women according to indicators of overweight and obesity. According to this study, the estimate of beta values was found to range from −6829 (p < 0.001) to −0.015 according to the BMI [33].
Three studies did not find associations between walkability, overweight, and obesity [25][26][27]. Two were longitudinal studies [25,27] while the other presented the results of cross-sectional and longitudinal analyses [26]. Despite not having found significant associations, it is important to note that these studies evaluated the changes in variables related to walkability as well as those of overweight and obesity. In addition, they referred to and discussed residential and neighborhood self-selection.   Neighborhood walkability derived from a validated index with standardized scores of 0 to 100 and with higher scores denoting more walkability. Neighborhoods were ranked and classified into quintiles from lowest (quintile 1) to highest (quintile 5) walkability. Sum of z-score land-use mix, net residential density and intersection density and divided into quartiles (lowest quartile, second quartile, third quartile and highest quartile)

Discussion
The aim of this study was to describe and summarize the evidence on associations between walkability and overweight and obesity. Seven out of ten included studies show significant inverse associations between walkability and overweight and obesity. Most studies show that less walkable neighborhoods are related to body weight outcomes in adult populations. Indicators measuring walkability index mainly include residential and population density, street connectivity and land-use mix. Some studies employed different buffer sizes around residents' homes of 800 to 1600 m measured along street networks and network buffers. No studies involving longitudinal design had significant results.
Largest associations were found by cross-sectional studies, which generally support the incapacity to establish causality. In addition, all studies were conducted in high-income countries, which differ from low-and middle-income countries in their application of policies, higher levels of urbanization, broader employment opportunities, and greater availability and quality of public services (e.g., public transportation) [35,36].
The main indicators used to calculate the walkability index are commonly used in studies involving health areas and active transportation [37]. Other indicators include different aspects of urban design (e.g., block group-level measures), residence types and displacement patterns (e.g., traditional core, high density and non-auto commuting) [32]. Most walkability indices have been created by z scores of the different indicators and divided by the number of indicators as residential density, land-use mix and street connectivity. Some walkability indices were categorized in quintiles or quartiles or expressed as continuous scores and for this review we used the studies when the walkability index was calculated. Interestingly, from our synthesis, only one study used specific indicators of food environments and physical activity as indicators of the walkability index [27]. While these indicators may be related to land-use-mix variables, the use of such indicators helps strengthen the index given their known association with overweight and obesity [38][39][40][41].
Another important issue to discuss concerns the sizes of neighborhood buffers used. Buffers of 800 to 1600 m around residences were used in certain works [15,29,31]. Muller-Riemenschneider et al. investigated 800 and 1600 m buffers based "street design" in association with obesity levels and found individuals living in high compared with less walkable areas were less likely to be obese (1600 m OR: 0.84, 95% CI: 0.7 to 1; 800 m OR: 0.75, 95% CI: 0.62 to 0.9) [31]. Frank et al. used 1 km network buffers [15,29] around households based on street segments and found inverse associations between the walkability index and obesity. Using 1 km network buffers, one study of adults (n = 10,878) living in Atlanta, Georgia, USA from 2000 to 2002 shows that each quartile increase in land-use mix is associated with a 12.2% reduction in the likelihood of obesity across sex and race [42]. However, walkability was not used because the authors used main indicators separately and these were interpreted as measures of walkability.
Thus, studies examining the relationship between the walkability index and overweight and obesity appear to exhibit no consensus regarding buffer sizes. While some use distances that individuals can travel by walking for 10 or 15 minutes, which usually vary from 500 to 1600 m [43], other studies use buffers of 400 m to 8 km [8], denoting the difficulty of establishing a common parameter. The studies included in this review use GIS tools to study household participants and census tracts [25][26][27][28]30,32,33]. This poses a challenge to walkability research because some features such as the quality and aesthetics of spaces and facilities cannot be measured with secondary or remote data [33].
Based on assessments of bias risk, the results of the synthesis expose important methodological issues related to such studies, such as in the item "withdrawals and dropouts" for two studies [27,31]. This property can be considered a limitation mainly of longitudinal studies [27], as it is necessary to have a percentage of individuals remaining in the study at the final data collection period, and some longitudinal studies have been able to reassess at least 70% of the individuals with intervals between two or three years [17,44]. This issue should be considered in future studies.
The assessment of certain predictors such as socioeconomic levels, demographic characteristics such as sex and other behavioral characteristics were used in the majority of studies examined in this review. Regarding socioeconomic levels, living in areas of higher socioeconomic levels may play a protective role against obesity [45], and this is interesting because socioeconomic status was used as an adjustment variable in different models and stratified the samples of some studies considered in this review [25,26,28]. Regarding sex, this predictor related to walkability is associated with overweight and obesity in both men and women and particularly among variables of destination diversity [33] (e.g., the land use diversity of a given neighborhood).
Additionally, neighborhood-based changes in walkability can shape other behavioral characteristics, such as leisure, commuting, and physical activity measures [29,46]. Decisions regarding land-use and transport planning can influence, for instance, the safety of walking and cycling as modes of transportation and the convenience of recreational physical activity [35]. This is important given that urban planning that develops neighborhoods with better indicators for walkable neighborhoods promote walking as a mode of transport [35]. Thus, creating smart cities that facilitate physical activity as part of everyday activity can promote health and prevent overweight and obesity in the global population [35]. We found four studies [26,27,31,32] that put physical activity as adjustment variable in different models. It is important to note that the mediating effect of physical activity in such relationship involving walkability and obesity may not be found and some results still remain inconclusive, and this is confirmed in one recent systematic review [20].
Fruit and vegetable consumption is an important predictor with some studies revealing a relationship between diet and certain urban food environment land-use characteristics [34,[47][48][49]. The availability [47] and variety [34,48] of healthy food is associated with diet, and supermarket density is related to higher levels of fruit and vegetable consumption [48] and to a reduced prevalence of obesity [50]. However, according to studies focused on the walkability index covered in this review, fruit and vegetable consumption significantly predicts BMI scores in a cross-sectional analysis model but not in a longitudinal analysis model [26]. One study using only longitudinal data and not using fruit and vegetable consumption as a predictor found that increased levels of obesity observed in lower-income neighborhoods are associated with issues of food accessibility [25]. Therefore, land use policies that protect and support access to healthful foods in urban areas are critical to mitigating differences in terms of access to local food.

Strengths and Limitations
The strength of the present study was a broad review of observational studies specifically on walkability, overweight, and obesity, identifying which qualities contributed to walkability and how they measured or quantified these qualities and the association with overweight and obesity.
Most of the positive evidence has been obtained from cross-sectional studies that support the notion that certain neighborhood characteristics are related to low overweight and obesity prevalence [36]. Therefore, caution should be exercised when extrapolating these results due to neighborhood self-selection bias, as people who are not obese and who live a healthier lifestyle that prevents obesity may choose to live in neighborhoods with better living conditions [17,18]. To obtain stronger causal inferences, further longitudinal and quasi-experimental studies should be conducted in addition to natural experimental studies to further our understanding of how walkability at different urban scales affects risk of obesity [36]. Another limitation refers to the type of sampling of the included studies, four did not mention the type of sampling and two were convenience samples, the rest being studies with randomized samples. And because of this, it is possible that unmeasured confounders contributed to some findings.
Weight and height as measures for determining BMI scores were self-reported in the majority of the selected studies, collected through home-based and telephone surveys, and correctly cited reliable and previously validated information [25,28,29,33,51].
A broader variety of methodologies, including variables such as shade from street trees, the widths of sidewalks, safety and crime prevention, and others should also be employed and assessed in terms of walkability. Environmental determinants of obesity, including a healthy diet and certain food environments, such as supermarkets and restaurant chains, among others, are not addressed in the studies reviewed [38][39][40][41]. Few studies have explored walkability outside of the neighborhood setting (e.g., in areas surrounding workplaces).

Conclusions
Most studies have found that less walkable neighborhoods are related to overweight and/or obesity in adult populations. Positive evidence has been obtained from cross-sectional studies and time-series studies, rather than longitudinal studies, and studies have been conducted in high-income countries. In addition, most studies have used a walkability index. Based on these results, the following recommendations can be made: 1) More studies should be conducted in low-income, middle-income, and middle-high-income countries; 2) more longitudinal studies (cohort and natural experiment) that control neighborhood self-selection need to be conducted; 3) other variables of the walkability index, such as food access, physical activity facilities, sidewalk access, and safety and crime prevention measured should be considered; and 4) better operationalizations of GIS evaluation variables (buffers sizes and census tracts) must be developed. Based on cross-sectional and time series studies, potential implications for clinical practice and policy-making can be reported, city planning and policy-related strategies aimed at improving the connectivity of the street network, mix of land uses and density of housing would enable the necessary supportive environments for health-related behaviors and prevention of chronic diseases. Understanding the factors that contribute to walkability can enable urban planners, designers and healthcare professionals to replicate better walkability conditions, providing more opportunities for active routes to help reduce overweight and obesity.