This population-based study included a large-scale and representative sample of elderly adults in Taiwan. We found that the risk of dementia significantly decreased by 12% among older adults living in areas with a high density of playgrounds and sport venues and the result did not change in any adjusted model. Also, lower odds of dementia (8%) were found in areas with a medium density of community centers, but the association was not significant after further controlling for individual-level factors. Although higher odds of dementia (14%) were found in areas with high median annual family income, such a significant association did not exist after further adjustment for physical environmental features.
Previous studies on the effects of recreational resources on dementia outcome were limited and mixed. The only two studies on this topic were cross-sectional designs from the UK [7, 8] and one of them found that mixed land use areas (inclusion of residential, commercial and recreational facilities, services and resources) showed no significant effect on prevalent dementia in later life [7]. Despite another UK study showing older adults living in higher mixed land use areas were significantly associated with an approximately 60% decreased odds of dementia [8], it is difficult to interpret the causal-relationship from these findings due to its cross-sectional design.
Our study used a population-based case-control design and found a high density of playgrounds and sport venues (as a proxy for availability of recreational resources) was associated with a 12% decreased odds of dementia in older adults even after controlling for individual factors, health care resources, and urbanization level. The mechanisms by which factors may affect this association remains unclear. However, it has been suggested that the availability of neighborhood spaces for recreational activities significantly promotes adults’ willingness to participate [34], which in turn can lead to older adults spending more time in recreational environments. These conditions are helpful for older adults to increase physical activity [34], social interactions [15] and cognitive stimulations [4], in turn improving mental health [35, 36] and cardiovascular health [37, 38] or enhancing cognitive reserve [19], resulting in a reduced risk of dementia [3, 19].
In addition, animal studies have shown that animals exposed to richer environmental stimulation contribute to neurogenesis via potential ways such as promoting proliferation, astrocyte, and inhibiting cell death [19]. Thus, the availability of playgrounds and sport venues may negatively influence dementia through mechanisms related to participation in recreational environments that could affect dementia directly [19] or lead people to engage in positive health behaviors [15, 19, 34,35,36,37,38].
This study also showed cases of dementia were more likely than controls to live in townships with medium compared with low community center availability. However, this finding is not in agreement with the only published longitudinal cohort study showing that older adults living in neighborhoods with more community resources had slower rates of cognition decline over an 18-year observation period even after adjustment of individual factors [6]. In our study, we did not find any significantly positive association between high density of community centers and dementia. Additionally, the effect of a medium density of community centers on the risk of dementia was no longer significant after adjustment of individual-level characteristics. The reason for this may be much explained by the individual characteristics of the older adults living in these areas, instead of the effects of community centers in local areas. For example, we further found that older adults living in townships with medium community center density have a higher proportion of white- and blue- collar workers (66.9% vs. 41.4%), lower dependence (27.1% vs. 43.6%), and less comorbidities (% with more than two diseases: 24.8% vs. 27.5%) (data not shown). This may explain why these people living in areas with a medium community center density have a lower risk of developing dementia than those living in areas with low community center density.
The effect of green environments was the other environmental effect which has been reported in association with cognition function in older adults [8, 9]. This environmental feature could be helpful in reducing cognitive loading, restoring attention, and thus benefit cognition [4, 9]. However, another study reported that more green environments were significantly associated with increased odds of dementia in later life, possibly because such environments may be associated with isolation, which limits access to local services, resulting in a lack of cognitive stimulation [8]. Our study found no significant association between parks, greeneries, and squares (as a proxy for availability of green environments) and the risk of dementia. This suggests that land use related to parks, greeneries, and squares at the township level in Taiwan doesn’t lead to any visible influences on the risk of dementia among older adults.
Only a few studies investigating the effects of area-based socioeconomic status in relation to the risk of developing dementia have been conducted [20, 21]. A UK cohort study recruiting 6220 nationally representative participants aged 65 years and over with a 12-year follow-up period, investigated the association between an index of multiple deprivation (i.e., a summarized index based on income, employment, education and training, health and disability, barriers to housing and services, living environment, and crime) and dementia incidence. The results showed older adults living in areas in the second-highest quintile of multiple deprivation was associated with an increased risk of dementia (HR = 1.62, 95%CI 1.06–2.46) compared to those living in areas in the least deprived quintile after adjustment of individual factors [20]. Another French study (three-city cohort) including 70 l6 individuals aged 65 years and over with a 12-year observation period also found the risk of developing dementia was positively related to a neighborhood deprivation score in women (HR = 1.29, 95% CI 1.00–1.67) but not in men [21].
The mechanisms by which factors affect the association between area-based socioeconomic status and dementia risk have not been fully identified, but it has been suggested that people living in better socio-economic environments are more exposed to higher densities of recreational resources (e.g., recreation centers, healthy food stores), and social and cultural resources (e.g., libraries, community centers) and thus lead people to engage in positive health behaviors and cognitively stimulating activities [4, 5]. These features could be helpful in reducing the risk of developing dementia. In contrast, people living in deprived areas could be related to poor conditions, lower densities of recreational, social, and cultural resources (less cognitive stimulation) [4], and greater presence of environmental stressors [5]. These latter factors may contribute to an increased risk of dementia.
Contrary to previous studies [20, 21], we found that cases were more likely than controls to live in townships with higher median annual family income, but the significant effect did not exist when physical and social environmental factors were controlled simultaneously. This may suggest that much of the effect of median annual family income is due to the influence of physical and social environments on each other, rather than the features of social environments themselves. It is possible that people with poor cognition may choose to live in areas with better socio-economic environments because they can obtain more health promoting services and resources in the affluent neighborhoods [39, 40]. It is also likely that high median annual family income influences dementia through individual factors because the association between high median annual family income and increased odds of dementia attenuated after controlling for individual factors.
One explanation for this condition could be that there were fewer white-collar workers (13.6% vs. 14.8%) among the dementia cases living in areas with high median annual family income, and they were more likely to have lower insurance premiums (percentage with median and higher insurance premiums: 25.9% vs. 27.8%), and more comorbidities (percentage with more than two diseases: 31.6% vs. 24.3%) than the control group (data not shown). These individual socioeconomic disadvantages may reduce physical activity [41], impair psychological function [42], and increase the risk of chronic disease [43], which contributes to an increased risk of dementia. Alternatively, we speculate that people with poor cognition might choose to live in areas with higher socioeconomic status because there are more medical facilities that can be utilized [44]. Therefore, after further controlling for factors of hospitals and clinics and urbanization level, the effect of high median annual family income on the risk of dementia was weakened; and the significant effect of medium median annual family income on the risk of dementia can be overlooked.
Regarding the effect of area-based education, we found that the adjusted OR of dementia for the exposure of living in townships with a higher percentage of illiterate people aged≧65 did not significantly increase in any adjusted models. Although a previous study reported most domains of cognition function were independently related to social cohesion [10], our study showed no significantly elevated adjusted OR of dementia for the exposure of living in areas with a higher density of elderly living alone. This suggests that the “percentage of illiterate people aged≧65 “and “density of elderly living alone” at the township level in Taiwan are inadequate in detecting associations with the risk of developing dementia among older adults.
As for the effects of salary-based insurance premium at individual level, we observed that salary-based insurance premium at individual level may change the associations of specific features of environments on the risk of dementia. Although no significant difference in the environmental features-specific odds of dementia was observed among subjects with dependents or salary-based insurance premiums lower than the median, the adjusted OR of dementia were significantly negative associated with a medium density of parks, greeneries, and square area (adjusted OR = 0.91, 95% CI 0.84–0.99) and a high density of playgrounds and sport venues (adjusted OR = 0.85, 95% CI 0.77–0.93). The reasons for the results are not clear.
We suspect that increased income status provides better affordable or accessible environmental resources (such as “parks, greeneries, and square area” and “playgrounds and sport venues”), potentially encouraging residents to utilize their environmental resources and resulting in increased cognitive stimulation and social participation. This condition may contribute to the significant effect of “parks, greeneries, and square area” and “playgrounds and sport venues” on the risk of dementia in subjects with salary-based insurance premiums greater than the median rather than dependents or subjects with salary-based insurance premiums lower than the median. More interestingly, we observed that the significantly positive effect of “parks, greeneries, and square area” on the risk of dementia only appeared in the medium density group (adjusted OR = 0.91, 95% CI 0.84–0.99) and not in the high density group, which may have resulted from the different distances to residential parks, greeneries, and square area [32]. A previous study in Taiwan categorized the measures of parks and green spaces into Quartiles and found parks and green spaces in the highest group were farther away from residential areas which may make it hard for the older adults to access. In contrast, the parks and green spaces in the Q3 group were close to residential areas [32].
This study has several strengths. First, we included a large numbers of study subjects by using the Taiwanese NHIRD, which makes the study population highly representative and leaves little room for selection bias. In Taiwan, the NHIRD covers around 99% of the entire population [45]. The NHIA performs expert reviews quarterly on a random sample of 50–100 outpatient and inpatient claims to ensure the accuracy of the claim files [45]. Thus, information obtained from the NHIRD is considered to be complete and accurate. Second, we used a case-control design to collect exposure information more efficient. All exposure information in this study was collected before the first diagnosis of dementia, which is helpful for explaining a causal relationship of the results with fitting temporal inference. Third, the likelihood of prevalence-incidence bias was also largely reduced by using initially diagnosed cases with dementia rather than prevalent cases.
Despite these advantages, this study also has some limitations. First, some environmental factors, such as neighborhood psychosocial disorders (e.g., crime), public transport availability, pollutants, and some individual factors, such as social engagement, smoking status, educational level, physical function and genes were not included in the analysis due to the lack of available data. Although we used occupational status as a proxy for individual educational level and used COPD (i.e., one of the comorbidities) as a proxy for smoking, residual confounding bias is still possible.
Second, the factors potentially associated with dementia analysed in our study included both individual-level and ecological-level variables. Using ecological-level factors to indicate an individual’s exposure status may incur exposure misclassification, which is likely to be non-differential and may result in underestimation of the associations between ecological-level factors and risk of dementia.
Third, we relied on the physician-recorded diagnosis in the medical claims to select dementia cases, which might result in disease misclassification. The medical claims only included medical information for people who sought care for dementia in hospitals or clinics. Therefore, it may have been mixed up with new onset or undiagnosed dementia in the control group. To address this concern, we included solely dementia cases that had at least 3 ambulatory visits with dementia-related diagnosis and the first and last outpatient visits at least 90 days apart to reduce the likelihood of disease misclassification.
Fourth, given the diagnostic procedures of dementia can be different across medical resource and medical care, we adjusted the hospitals and clinics in each township and the level of urbanization to reduce the differences in medical resources and medical care resulting in unequal opportunity to be diagnosed as having dementia among dementia cases.
Fifth, previous studies suggested the pathology often starts at least 10 years before the onset of symptoms of Alzheimer’s disease [46, 47]. Owing to the lack of available nationwide land use data prior to 2006, the retrospective period in our study was limited to 4 years or less. This may not have been adequate for finding longitudinal associations. We only used each subject’s exposure to physical and social environments beginning from the year of 2006, which might have led to some degree of environmental exposure misclassification. However, such exposure misclassification was also likely to be non-differential in the dementia and control groups.
Sixth, our ability to examine the biological gradient effect of environmental features on the risk of dementia was limited by the measures available. We have left this area (such as biological gradient effect) for further investigations.
Finally, information on residential mobility in our study showed about 40% of subjects living in different townships over the 5 years, which may have resulted in relocation bias. To avoid this, we further excluded these subjects in the logistic multilevel regression analysis and found the association between the features of physical and social environments and the risk of dementia was little changed. Thus, relocation bias in our study may be small.
