The research data consisted of two categories. The first is the individual-level sociodemographic and health-related data of the elderly, which is collected from the “Fourth Survey on the Living Conditions of the Elderly in China” in 2015. The survey was organized by the China Committee on Aging and adopted a multi-stage and multi-level sampling method proportional to the population scale, targeting older adults aged 60 and above, with samples covering 31 provincial-level administrative regions (excluding Hong Kong, Macao, and Taiwan), with a total sampling ratio about 1‰. Each selected respondent was numbered by sampling. The trained surveyors went directly to the households and conducted face-to-face interviews with the respondents to ensure that the respondents fully understood the survey content and the surveyors filled in the questionnaire according to the respondents' answers. The study selected the representative Chinese megacity, Beijing (the capital), which has a hot and rainy summer and a cold and dry winter, with a total area of 16,410.54 km² and a resident population of 21.886 million at the end of 2021. In this study, 112 neighborhoods in Beijing were selected as samples, covering seven administrative districts: Haidian, Dongcheng, Xicheng, Chaoyang, Fengtai, Fangshan, and Shunyi. Each neighborhood had a sample of 30 older adults, and the total number of samples was 3,360. We removed the invalid samples with information loss, finally preserving 2,778 valid samples. The second category is built environment data at the neighborhood level. The data mainly comes from point of interest (POI) data, the 2014 land-use status map of Beijing, and Baidu Street View images.

Multilevel Regression Model and SEM were applied in this study. Multilevel modeling has been widely used in studies of neighborhood environments and health outcomes (52) because it allows for the simultaneous estimation of the effects of individual-level and group-level factors. The SEM has some advantages in dealing with the quantitative study of the interaction between multiple variables. The data had a clustered structure with older adults nested within the neighborhood, that is, the older adults living in the same neighborhood share the same environmental exposure. Moreover, there are complex interactions between the built environment, health activities, and health outcomes, so MSEM is more appropriate for analysis. Before constructing the model, six dimensions of variables, such as density, diversity, design, destination accessibility, distance to transit, and 3D were constructed based on 13 built environment indicators. Since the built environment indicators are all derived from objective measurement data rather than scale data, the traditional method of constructing latent variables is no longer applicable. Based on the relevant literature, this study used logistic regression to construct latent variables (53). The three steps of the analysis are as follows:

The first step was to construct a multi-level multiple regression model. Based on the five buffer scales of neighborhood administrative boundaries, 300, 500, 800, and 1,000 m buffers were used to measure the built environment characteristics. We then identified the built environment factors affecting the elderly health based on the results and defined at which scale of the geographic unit the relationship between the built environment and elderly health is more appropriate.

The second step was to construct the MSEM. Mediating variables were incorporated to examine the pathways of health activities. In addition, the stability of the model was verified by avoiding residential self-selection. The specific method was to screen out the elderly living in work-unit compounds and affordable housing neighborhoods as subgroups, including 48 neighborhoods with 1,222 elderly people, accounting for 44% of the total sample.

The third step was to construct a heterogeneity analysis model. Based on the stratified analysis, we explored whether the pathways of neighborhood-built environment and elderly health differed by age and gender. For the age heterogeneity model, the study sample was divided into three groups: low-aged group (persons aged 60–69), middle-aged group (persons aged 70–79), and high-aged group (persons aged 80+). For the gender heterogeneity model, the study sample was divided into two groups: elderly males and elderly females. The study combined Fisher's combination test for heterogeneity effect and set the number of sampling times to 1,000 based on bootstrap.

The analysis was conducted using Mplus7.4.

Descriptive statistics for the study population are given in Table 2. Over half of the respondents were female (52.99%), or low-aged elderly (56.30%). The majority was presented as non-living alone, educated at primary level or above (95.21%), had an annual household income of more than 50,000 RMB (74.37%), and had the habit of undergoing routine medical exams (65.87%). Overall, the participants exercised three to five times on average per week and they participated in at least one type of social activity on a regular basis. Less than half (44.67%) of the older adults had a good self-rated health status and the majority (80.63%) suffered from at least one chronic disease. For mental health, most older adults never felt alone (83.3%), and believed they were relatively happy (79.91%).

Descriptive statistics of neighborhood characteristics are given in Table 3. The average land-use mix tended to increase as the buffer scale increased. The density of road and bus stops was similar in all five buffer scales, while the density of intersections (21.72) and subway stations (0.18) was the lowest within the administrative boundaries. The density of commercial, recreational, and medical facilities decreased as the buffer scale increased, but the density of leisure facilities did not show this pattern. Greenery and sky visibility rates were similar in all five buffer scales while building visibility rates were the highest within the administrative boundaries (0.39).

The results of the multilevel regression models of the relationship between the built environment at the five buffer scales and elderly health are summarized in Table 4. We found that the correlation between built environment factors and elderly health varied by geographic scale. The population density was positively correlated with chronic diseases and negatively correlated with subjective wellbeing. Land-use mix was positively correlated with self-rated health, which was found only in the 500 m buffer. Street connectivity had a positive effect on self-rated health and a negative effect on chronic diseases at 300 and 500 m buffers, while it had a positive effect on the subjective wellbeing only within the administrative boundaries. Destination accessibility was negatively correlated with loneliness (all five buffer scales) and positively correlated with subjective wellbeing (except for the 800 m buffer). Distance to transit was positively associated with self-rated health and negatively associated with chronic disease (only in the 1,000 m buffer), while it was positively correlated with subjective wellbeing (only in the 500 m buffer). The 3-D environmental perception had a negative effect on loneliness in the 500 m buffer, and significantly associated with all health outcomes in the 800 m buffer. In general, compared with other buffer scales, there are more built environment elements in the 500 m buffer that were significantly associated with health outcomes, which could better explain the relationship between the built environment and elderly health.

Table 5 shows the results of the MSEM of the full sample. Physical activity and social interaction were found to be key pathways through which the neighborhood-built environment influences the elderly health. The following are the effects of the physical activity path-based built environment on elderly health: population density had a negative effect, while land-use mix, distance to transit, and 3D-environmental perception had a positive effect. Higher population density may reduce self-rated health and subjective wellbeing and increase the risk of chronic diseases and loneliness in older adults by reducing the physical activity. Higher land-use mix can reduce the risk of chronic diseases and promote the subjective wellbeing of the elderly by increasing the physical activity. A shorter distance to transit may promote self-rated health and subjective wellbeing and reduce the risk of chronic diseases and loneliness by increasing physical activity. Moreover, better 3-D environmental perception may reduce chronic diseases and loneliness and promote subjective wellbeing by increasing physical activity.

The effects of social interaction path-based built environment on elderly health are as follows: population density had a negative effect, while street connectivity and distance to transit had a positive effect. Higher population density may have the potential to reduce self-rated health and subjective wellbeing and increase the risk of chronic disease among older adults by decreasing social interaction. Higher street connectivity was associated with higher self-rated health and subjective wellbeing and a lower risk of chronic disease among older adults through increased social interaction. Higher proximity to public transportation may promote self-rated health and subjective wellbeing and reduce the risk of chronic disease through increased social interaction.

Table 6 shows the results of the MSEM of the subgroup sample. This study found that the model results after controlling for residential self-selection did not differ significantly from the full sample, with only a few effect pathways changing. For self-rated health, the mediating effect of population density through physical activity was no longer significant, and the mediating effect of population density through social interaction on self-rated health (−0.018) was stronger than that of the full sample (−0.014). The mediating effects of street connectivity through social interaction and distance to transit through physical activity and social interaction on self-rated health were no longer significant. The mediating effects of the neighborhood-built environment through physical activity on chronic disease and subjective wellbeing were similar to that of the full sample. However, the mediating effects of street connectivity and distance to transit through social interaction on chronic diseases were no longer significant, nor were the mediating effects of street connectivity through social interaction on subjective wellbeing significant. For loneliness, the mediating effect of the built environment affecting loneliness through physical activity was consistent with that of the full sample.

Figure 1 summarizes the differences in the mediating effects of neighborhood-built environment on elderly health among different age groups. This study found that in the low-aged group (persons aged 60–69), population density, street connectivity, distance to transit, and 3D environmental perception affected chronic disease and subjective wellbeing through physical activity. Street connectivity, destination accessibility, and 3D environmental perception affected self-rated health, chronic disease, and subjective wellbeing through social interaction. In the middle-aged group (persons aged 70–79), population density, land-use mix, and distance to transit were critical for chronic disease and subjective wellbeing by affecting physical activity, but for loneliness, only population density and distance to transit showed a significant impact through physical activity. In the social interaction pathway, only population density played a significant role on self-rated health, chronic disease, and subjective wellbeing. In the high-aged group (persons aged 80+), only population density showed a significant correlation with all health outcomes through physical activity. In the social interaction pathway, both population density and distance to transit showed a significant correlation on self-rated health, chronic disease, and subjective wellbeing.

Figure 2 summarizes the differences in the mediating effects of neighborhood-built environment on elderly health in different gender groups. The results showed that in the male group, population density, distance to transit, and 3-D environmental perception affected chronic diseases and subjective wellbeing through physical activity, and land-use mix only had a positive effect on subjective wellbeing. In the path of social interaction, population density, street connectivity, and distance to transit showed a significant correlation between self-rated health, chronic diseases, and subjective wellbeing. In the female group, population density, distance to transit, and 3-D environmental perception were found to have a significant effect on chronic disease, loneliness, and subjective wellbeing by physical activity; while population density and street connectivity were associated with self-rated health, chronic disease, and subjective wellbeing through social interaction. However, for loneliness, only population density had a positive effect on it through social interaction.

This study explored the complex relationship between neighborhood-built environment and elderly health. The results indicate that different characteristics of the built environment played different roles in health outcomes through physical activity and social interaction. There was variability in these pathways among older adults of different age and gender groups. Our main findings are as follows.

The choice of neighborhood area definition may dramatically influence the observed relationships between built environment characteristics and elderly health (9). Our research revealed that built environment elements are most significantly related to health outcomes in the 500 m buffer compared with other areas. One possible explanation is that with the increase of age, the physical function of the elderly decreases, resulting in the restriction of their accessibility. The 500 m buffer was selected as the point beyond which daily activity of the elderly begins to decline dramatically and may better capture older adults' daily activity space (54). In addition, neighborhood environmental attributes within the 500 m buffer also appear to be strong predictors of activity for older adults (55). This is consistent with the buffer scale selected in other previous studies conducted in China on the relationship between neighborhood-built environment and elderly health. It suggests that the 500 m buffer can more accurately reflect the built environment characteristics to which older adults are exposed in their daily lives (56). Therefore, the issue of the uncertain geographic context problem cannot be ignored when considering the impact of the neighborhood-built environment on public health. The actual range of daily activities of different groups needs to be considered and the exposure scale of the built environment needs to be defined by multiple comparisons.

In the path of physical activity, some built environment characteristics could have a positive impact on health through physical activity, which is similar to previous studies (57). Land-use mix, distance to transit, and 3D environmental perception are enablers for outdoor physical activity among older adults. When living in a different land-use area, older adults may have easier access to various facilities and resources, which helps them to achieve higher levels of physical activity (58). Convenient public transport stations can provide various travel routes, which helps encourage the elderly to choose public transportation as a travel mode and increase their physical activity as they walk to transit stops. Previous studies have shown that ~29% of Americans can achieve adequate levels of physical activity only by commuting to and from public transport stations every day (59).

Regarding 3D environmental perception, an increase in the visibility of street greenery implies a more comfortable and pleasant street environment, and older adults are more likely to choose to walk or exercise in such green environments (58). Strong evidence showed that increased physical activity can help promote metabolism, improve body resistance, enhance cardiopulmonary function and flexibility of joints and bones, and reduce the risk of chronic diseases (57). Also, increased interaction with other people during physical activity can be effective in relieving mood and increasing serotonin levels, which can help promote self-rated health and subjective wellbeing and reduce loneliness (60, 61).

It is worth noting, however, that higher population density was negatively associated with health. This is in contrast with many previous studies conducted in Western countries (62, 63). However, the results are in line with many other studies on older adults in Asian cities (64). This is probably due to the fact that older adults are more likely to live in higher-density neighborhoods in Beijing, resulting in a lack of adequate space and places for physical activity. Decreases in physical activity may prolong sedentary time, aggravate the risk of chronic diseases such as obesity, diabetes, and cardiovascular disease, and thus are harmful to health (65).

In the path of social interaction, in line with previous research, our results also suggest that older adults can benefit from high levels of street connectivity and convenient transportation. Higher street connectivity can help encourage older adults to go outdoor for activities and increase their willingness to interact with neighbors (66). Similarly, easy access to public transportation can reduce social exclusion and isolation by increasing physical accessibility (67). As expected, increasing social participation and access to social support for older adults can help generate a positive attitude and a sense of neighborhood attachment and belonging, which is important for maintaining and promoting wellbeing and self-rated health (68). Previous research has also shown that older adults can gain more opportunities to interact with health care providers through social support and discussions with older friends about disease management, which can help them persist in treating chronic diseases (69).

Our study further found that population density had a significant negative impact on health, which is in line with the existing literature (70). In high-density neighborhoods, residents are often exposed to large numbers of strangers, exacerbating the weakening of interpersonal relationships, and reducing social interaction (71). High-density environments may also bring negative psychological feelings, such as psychological stress, anxiety, loss of control, cognitive overload, and violation of personal space (72), reducing the willingness to participate in social activities. Previous studies have shown that inadequate support for social interaction may lead to sleep problems in older adults, resulting in decreased immunity and increased mortality, impairing self-rated health and subjective wellbeing, and increasing the risk of chronic disease (52, 73).

In addition, China's unique housing policy creates an opportunity to control residential self-selection (74). In the model excluding the self-selection samples, some mediating effects are no longer significant, which reflects that individuals who love physical activities and social life are more likely to choose to live in neighborhoods with higher-street connectivity and convenient transportation. However, the subgroups do not differ significantly from the results of the full-sample model in general, and built environment characteristics still play an independent and robust impact. The result may be explained by the fact that the main subjects of the study are local elderly people who have lived in the neighborhood for a longer period and have less self-selection of living location, reducing the possibility of taking the built environment characteristics as the main consideration for housing selection.

The relationship between neighborhood-built environment, health activities, and elderly health differed significantly by age group. The role of population density in changing health activities intensifies with increasing age, with significant effects on health outcomes concentrated among older adults in the middle and high age groups. The reason may be that the negative effects of environmental pressure, such as noise and insecurity caused by higher population density on physical activities, social interaction, and other behavior habits gradually appear after enduring them for long periods of time (75). While in the low-aged group, the built environment characteristics that had the most significant positive effects on health included street connectivity, destination accessibility, distance to transit, and 3-D environmental perception, which is consistent with previous research (76). Compared to other age groups, low age elderly has more physical activity and social needs, and higher demands on the quality of their environment (37). For the high-aged group, in addition to population density, no significant relationships were found between other built environment elements and health outcomes through physical activity, indicating that older adults have a weaker perception of the environment conducive to physical activity.

It is interesting to note that the effect of distance to transit on health through physical activity was more significant in the low and middle age groups, but the effect through social interaction was only significant in the high-aged group. The potential explanation might be that compared with the high-aged group, lower age elderly have a wider range of activities, and convenient public transportation facilities have a greater impact on their physical activity. For the high-aged group, the time and intensity of physical activity tended to decline with age, and their range of activity was constrained, resulting in a lower need to rely on public transportation for physical activity. However, previous studies have suggested that the high-aged group elderly have a higher level of need for social interaction and social participation (37) and are more dependent on physical space. As the walking range of the high aged elderly is limited, it is more difficult for them to travel by bike or car for social interaction, and they have a higher demand for public transportation (77).

Moreover, the study found that the relationship between neighborhood-built environment, health activities, and elderly health differed significantly by gender group, but the gender heterogeneity was not as obvious as the age heterogeneity. The effect of distance to transit on health through social interaction was only significant in elderly males. Consistent with the findings from previous studies (78), compared to elderly females who preferred to carry out activities near their neighborhoods, elderly males were more active in social interaction and had a larger spatial range, leading to a higher reliance on public transportation for their outdoor activities. The effect of land use mix on subjective wellbeing through physical activity was only significant in elderly males. The reason might be that elderly males had a wider variety in types of daily exercise activities, such as Taichi, sword dancing, gateball, and other activities. However, the need for elderly females was for facilities more focused on home shopping, which is relatively homogeneous. Therefore, for elderly males, higher land-use mix has a greater impact on physical activity by providing diverse places for physical activity to meet their exercise needs, and improve their subjective wellbeing (79, 80).

Interestingly, the effect on loneliness by improving the built environment was only significant in elderly females but not in elderly males. These characteristics include population density, distance to transit, and 3D environmental perception, which is consistent with previous studies (81). The reason might be that elderly females are more sensitive to the environment and emotions, more dependent on social support and companionship through the external environment, to develop interpersonal relationships in various activities, and to alleviate loneliness.

This study has many limitations. First, the cross-sectional research design is not sufficient for explaining the correlation between neighborhood-built environment and elderly health. More longitudinal data are needed to consider a quasi-natural experiment in order to generate high-quality evidence to provide a more accurate basis for new policy decisions. Second, the study has discussed only two pathways, namely, physical activity and social interaction, and other potential pathways are missing. Further studies should investigate more potential mediation, such as stress exposure, pollution exposure, dietary behavior, and traffic safety, to further deepen the understanding of the relationship between the built environment and health. Last but not the least, this study adopted a static residence-based approach to measure the built environment, which is prone to the problem of uncertain geographic context and average neighborhood effect (82). Future studies should measure the built environment exposure based on individual mobility and incorporate wearable environmental monitoring devices to improve accuracy.