The first aim of this study was to identify built environmental patterns in a European urban context. To approach the complexity of urban forms in European settings, we used a methodology enabling us to analyze the co-occurrence of several important characteristics of the built environment. Using GIS datasets of urban characteristics and a combination of MCA and cluster analysis, we identified seven built environmental patterns in the region around Paris, France. A pattern characterized by low spatial accessibility to green spaces and proximity facilities and an absence of cycle path was found only in neighborhoods in the outer suburbs, whereas patterns characterized by higher spatial accessibility to green spaces and proximity facilities and the presence of cycle paths were more evenly distributed across the region. In addition, we found an increased likelihood of walking and cycling in subjects residing in neighborhoods characterized by high accessibility to green spaces and proximity facilities compared to those living in neighborhoods with low spatial accessibility to green spaces and proximity facilities and without cycle paths, after adjusting for individual sociodemographic characteristics (age, gender and education level) and median neighborhood income.
One strength of the present study lies in our analysis of built environmental patterns that take into account specificities of the physical environment in urban contexts. Examination of spatial distribution of the seven patterns and analyses of the relation between these patterns and walking and cycling behaviors reveal the complexity of the urban built environment in this study region (Ile-de-France), beyond a simple division between the city center, the inner and the outer suburbs. In the current investigation, we were specifically interested in pedestrian and cycling-related built environments based on three variables: green spaces, cycle paths and proximity to facilities. Previous research suggested the importance of built environment attributes facilitating walking, such as high residential density, street connectivity and land use mix [24, 50–52]. In the present study, information on residential density was not included in the analyses of environmental patterns because of strong correlations observed between population density (using built-up area density) and spatial accessibility to proximity services (Spearman rank correlation = 0.73 p < 0.0001). In addition, information on the presence of sidewalks on both sides of the street (a component of walkability) was not found relevant to the situation in France. We estimated the availability of sidewalks (length of sidewalks in kilometers in each neighborhood in Paris) and the density of road network (excluding large roads and highways) in each neighborhood of the study region; however, the variability of these characteristics was not significant (data not shown). In French suburbs, there are almost always sidewalks alongside the streets, and these may include cycle paths. Indeed, our patterns with high availability of cycle paths were localized both within Paris and in the suburbs (inner and outer).
To categorize neighborhoods into homogeneous patterns according to built environmental characteristics, we used a combination of MCA and hierarchical cluster analysis. This approach has been used in previous studies to define social and demographic neighborhood patterns and to assess individual health outcomes according to these social patterns . As described by Escofié et al. , MCA was used as a preliminary step for classification due to its role as a filter for eliminating non-relevant dimensions that might be assimilated with ‘statistical noise’. In a second step, most significant factorial dimensions obtained by MCA were included in cluster analysis to generate groups of neighborhoods with similar built environmental characteristics. Integration of these two methods enables analyses based on the most significant interrelations between categories of built environmental variables  and might better explain the diversity and natural grouping of characteristics, thereby overcoming limitations due to use of either factor or cluster analysis .
An important feature of our study was that we were able to assess the relationships between built environmental patterns and walking and cycling behaviors in a sample of middle-aged adult residents. We found that patterns 1, 3 and 5 were significantly associated with the likelihood of walking and cycling over the median weekly duration. In order to generate hypotheses concerning these relationships with built environmental patterns, we assessed the geographic distribution of the seven neighborhood patterns. The principal difference between patterns 1 and 3, both localized across the three zones of the region, was in the availability of cycle paths, lower in pattern 3 than in the reference pattern. There is evidence that the presence of cycle paths is important in creating an active living local environment. In New Orleans, an increase in the average number of riders (both adults and children) per day was observed after the first on-street bike lane was painted on the streets of the city in 2007 . However, in our study, increased likelihood of walking and cycling was also found for pattern 5, characterized by an absence of cycle paths. Depending on the urban context, this relationship may be due to other aspects of the built environment that may influence walking and cycling, such as specific equipment in the green spaces [21, 55, 56] or the amount of traffic .
In contrast to positive relationships with walking and cycling of some built environment patterns, no significant association was found with BMI. This is consistent with findings from a recent systematic review  which suggested that some built environmental features were associated with an increased level of physical activity, especially walking, but not with BMI. In addition, in another recent review  the relationship between green space and weight was found inconsistent. As emphasized by the UK Foresight Report , excess weight and obesity represent highly complex systems shaped by multiple interdependent factors acting throughout the lifespan. Therefore, to be able to demonstrate a direct and simple relationship between built environment patterning and weight status would be unexpected. Especially, the cross-sectional design of available studies cannot take into account potential time-lags between exposure to built environment and body weight change , pointing to the importance of longitudinal studies.
In the present study, walking and cycling data were available only for recreational physical activity. It should be noted, however, that more than 50% of study subjects (68.6% in men and 51.1% in women) were retired. Therefore, information on commuting activities may not be relevant. In a recent systematic review focusing on older adults (over 65 years), Van Cauwenberg et al. . pointed out that, while non-significant associations between walkability indices and recreational walking were found in the US and Australia a significant positive association was found in Belgium. In a European urban context, Van Dyck et al. assessed the link between walking and cycling (transport and recreation) and the level of walkability in an urban city (Ghent) in Belgium . That study showed that a high level of walkability was positively associated with a high level of walking both during transport and recreation.
Several limitations of our study need to be pointed out. The design was cross-sectional and thus we could not establish causal relationships between built environmental patterns and walking and cycling behaviors. Nor can we rule out a bias related to residential neighborhood self-selection. Neighborhood selection may be determined by numerous factors based on financial constraints, availability of equipment, transportation infrastructures and lifestyle [60, 61]. In our study, subjects were volunteers in a nutritional intervention study [42, 43]. Although characteristics of the participants of the SU.VI.MAX study were found to be close to those of the national population according to socioeconomic status and to the distribution of major risk factors for cardiovascular disease and cancers [42, 43], these subjects may have had healthier lifestyles.
Another limitation was that measurements of walking and cycling were derived from self-report and, as noted in previous studies [62, 63], this may be a source of potential misclassification. In general, it is known that physical activity (especially information concerning duration and frequency) tends to be overreported and sedentary behavior is underestimated . In addition, although our patterns were based on analyses of objective GIS built environmental data, we lacked information on environmental variables which may promote or limit walking and cycling behaviors, such as road traffic and safety , and specific characteristics of green spaces such as access points and the presence of equipment [21, 55, 56]. Another limitation is related to geographic scale and area arrangement. Results are likely to vary with the size and the arrangement of areas, with larger administrative units being more heterogeneous, which may occur when the same range of data, calculated at various spatial levels, produces different results [65, 66]. To limit a potential bias related to the relationship between size and location, the variable used for measuring spatial accessibility to facilities was defined according to the built-up area of each IRIS.
Finally, our study territory was limited to one European city (Paris and suburbs); further studies are needed to determine whether they can be generalized to other European urban settings.