This study of reported built environment attributes from representative samples of participants in 11 countries found that a combination of seven neighborhood environment attributes resulted in five meaningful neighborhood patterns across countries. These patterns of co-existing neighborhood attributes varied in their conceptual supportiveness of physical activity with one revealing an overall pattern of unsupportive built environment attributes, others revealing differing patterns balanced towards between either more walkable or more recreational attributes, and one pattern revealing an “overall activity supportive” pattern. Interestingly, the proportions of participants grouped in each pattern varied by country, suggesting that representation of these neighborhood types differed across countries. Only two patterns (i.e., “overall activity supportive” and “high walkable and unsafe with few recreation facilities”) were significantly associated with meeting the physical activity guideline either by walking only or total physical activities. Previous analyses of the current dataset found bivariate associations between specific supportive neighborhood attributes and physical activity levels, and a summary score suggested that the greater number of neighborhood attributes supporting physical activity the higher odds of meeting the physical activity guideline . The current study adds to previous findings by answering the question “which patterns, if any, are associated with sufficient walking or total physical activity, and which patterns of amenities occur most frequently across countries?”
Combinations of neighborhood attributes revealed specific neighborhood types. Based on the hypothesis derived from ecological models,  it was expected that meaningful combinations of neighborhood attributes would emerge and combinations with more attributes supportive of physical activity would be associated with physical activity. A pattern defined by low probabilities of residential density, shops, sidewalks, transit stops, bicycling and recreation facilities but generally safe from crime at night was identified. This class was used as a reference category because conceptually it represented the most unsupportive pattern for physical activity (i.e., safe but activity unsupportive (Class 5)). This class had the lowest prevalence (5.3%) of participants overall, but participants from the U.S. (14.1%) were 2 times more likely to be classified into this pattern. Canada was also overrepresented, but not to the same extent as the U.S. Both countries were disproportionally overrepresented compared to other countries such as Hong Kong, Colombia, Sweden, Norway (less than 2%). The result is consistent with the history of the last 70 years in North America where many neighborhoods have been designed from a perspective that favored automobile-dependent and suburban type developments with low residential densities and segregated land uses, patterns generally associated with lower PA, higher obesity rates, and other deleterious health and environmental exposure outcomes [41, 42].
The pattern observed in Class 1 (overall activity supportive) showed the most favorable neighborhood design attributes with positive probabilities for all built environment attributes (crime safety was about equal). Based on a hypothesis from an ecological perspective, Class 1 was expected to have the strongest association with physical activity of all the classes. Relative to Class 5, Class 1 neighborhood pattern was strongly associated with meeting guideline for both total physical activity (OR=1.61) and walking (OR=1.90). The association for meeting the guideline in Class 1 for walking appeared stronger than for total activity, but the overlap in confidence intervals suggests that these may not be statistically different. Interestingly, the majority of participants across 11 countries were classified into this overall supportive pattern (52% prevalence). A higher prevalence of participants from many European countries, with cities designed for walking, were included in this class such as Sweden, Norway, and Belgium. Also, proportionally many participants from Hong Kong, New Zealand, and Canada were overrepresented by this class compared to the overall prevalence (52%). Participants from Brazil, the U.S., Japan, and Colombia were underrepresented compared to the overall prevalence. It was surprising that some countries were so well represented by Class 1, such as the U.S. with a prevalence of 44% when objective data for walkability suggest that such neighborhoods are uncommon in many parts of North America . A representative sampling approach was undertaken in each country, so this result is difficult to explain but may reflect the difference between objective and self-reported environmental features of neighborhoods [44, 45]. In general, the strong association between Class 1 and PA supports the hypothesis that a pattern formed by the combined influence of all the activity supportive attributes is associated with PA across countries. The stronger association for meeting the guideline by walking may have resulted from the greater influence and specificity of walkability design features (e.g. residential density, retail destinations, transit stops, sidewalks).
The class labeled high walkable and unsafe with few recreation facilities (Class 2) revealed the strongest association with meeting the guideline for total activity (OR = 1.62) and walking (OR = 2.26) relative to the most activity unsupportive (i.e., Class 5). However, large confidence intervals for both Classes 1 and 2 for total physical activity or walking suggest they may not differ from each other. Approximately, 16% of the sample across 11 countries was classified into this pattern with Colombia, Lithuania and Hong Kong overrepresented. In Class 2, multiple family homes and sidewalks were almost certain to exist as were strong probabilities for shops and transit stops within walking distance. Facilities supporting cycling and recreational physical activity were less likely to be reported within walking distance, and the sense of being unsafe at night was the greatest of all the classes. This combination of attributes suggests a typical urban walkable environment, which may explain the lower perceived safety from crime when population density is higher. However, high-density urban environments have been consistently associated with greater utilitarian or active transportation and less automobile dependence [9–12]. High residential densities provide the critical mass of people needed to support shops, services, and public transit. For example, it is well known that access to transit facilities vary by ridership, which is affected by residential density . Additionally, sidewalks make up an important attribute of “complete streets” and provide a designated area for pedestrian travel. However, it was unexpected that this class, with greatest concerns about crime safety, would be associated with the greatest likelihood of meeting the physical activity guideline.
The relation of neighborhood crime to physical activity has been equivocal . Across patterns, the results from this study found that perceived safe conditions in one’s neighborhood was not sufficient for higher levels of physical activity to occur in the absence of supportive built environment attributes for physical activity (i.e. Class 5). Yet, the greatest odds of meeting the guideline by walking occurred in the neighborhood typology with the highest perceived concerns of unsafe conditions, but any negative effect on physical activity appears to be mitigated by activity supportive features indicating a highly walkable urban design (i.e. Class 2). The neighborhood type with perceived presence of all seven built environment attributes, but equal likelihood of safety and unsafe conditions (i.e. Class 1), also was related to meeting the physical activity guideline. These results regarding personal safety were surprising and further add to the equivocal findings [18, 46]. It seems intuitive that safe neighborhoods would be necessary for physical activity to occur. However, studies of lower income and minority populations in the US, which tend to live in poorer, dense, and more unsafe areas, suggest that prevalence of meeting the physical activity guideline does not differ by subgroup when physical activity is measured objectively , suggesting that transportation and occupational activities account for greater activity in these populations because active transportation occurred of necessity. Perhaps, even in the context of unsafe conditions, people adapt by learning behavioral responses (e.g. walking with groups, avoid eye contact), contextual cues (e.g. walking only during the day) or do not have a choice (e.g. need to go to work) that moderate the relations between unsafe conditions and activity-supportive environments and physical activity. Additionally, personal safety and neighborhoods type might be more salient for youth or older adult populations who are limited in their response to unsafe conditions. Research designed specifically to disentangle these complex effects is needed.
Class 3 (safe with active transport facilities) and Class 4 (transit and shops dense with few activity amenities) were not associated with meeting guideline, except that Class 4 was significantly and negatively associated with meeting the guideline for total activity. Class 3 appears to have attributes supportive of recreational physical activity, such as access to bicycle and low-cost recreational facilities, along with sidewalks, but because the IPS did not measure domain-specific physical activity, it unknown whether recreational activities would be associated with this class. It is unclear why Class 4 (transit and shops dense with few amenities) was negatively related to meeting the guideline for total physical activity. Perhaps a lack of recreation areas combined with safety concerns and lower residential densities, even if shops and transit are available, make this an aversive environment for recreational activities, which would tend to lower estimates of total physical activity.
Strengths of the current study included an international coordinated study with the use of samples of adults from representative urban areas in 11 countries; a validated and culturally adapted physical activity measure designed for international surveillance; a brief environment survey also designed for surveillance to assess key neighborhood attributes known to be associated with physical activity; and state-of-the science latent class analysis. Limitations included those inherent in a cross-sectional design such as the inability to disentangle cause-effect relationships and same source bias that can result from use of self-reports to assess both environment and physical activity. The IPAQ short form is global measure that may underestimate labor and domestic physical activities in developing countries. These reporting biases could be mitigated in the future by an international study collecting objective measures of built environment attributes and physical activity, such as the International Physical Activity and Neighborhood Environment (IPEN) Study , which will provide objective measures of both environment and activity. Additionally, seven single-item questions were used in the present study to assess several built environment constructs. These measures have shown fair-to-substantial agreement in a previous study  but may not capture the full range (e.g. intersection density) or depth of these environmental constructs. For example, measures of safety do not assess the type or severity of the dangers (e.g. property vs. violent personal crimes). While comprehensive self-report surveys of neighborhood design attributes are practical and affordable alternatives , numerous multiple-item scales were impractical for this international surveillance study. The PANES has shown acceptable reliability and validity compared to comprehensive multi-scale surveys of the built environment and physical activity [30, 32]. Because educational attainment was unavailable for all countries, it is also possible that the generalized linear regression model is underspecified. Finally, cultural differences with respect to physical activity domains were not considered in this study.
Neighborhood design attributes supportive of physical activity vary both within and between countries. Previous analyses of the IPS environment data  and the current study suggested that individuals living in neighborhoods likely to have at least four attributes specific to walkability (i.e., access to shops and services, high residential densities, sidewalks, and transit stops) or that are “overall activity supportive” have a greater likelihood of meeting the PA guideline by walking or total activities. The WHO recommendation [1, 5] for population-based strategies to increase physically active through improved urban design, public transportation, and recreation attributes was supported in this study of diverse countries.