The associations between social, built and geophysical environment and age-specic dementia mortality in a high-density Asian city

Background Although socio-environmental factors which may affect dementia have widely been studied, the mortality of dementia and environmental relationships have seldom been discussed. Method: A register-based study was conducted to evaluate the relationships of four individual-level social measures, two community-level social measures, six short-term (temporal-varying) environmental measures, and four long-term (spatial-varying) environmental measures with dementia mortality in a high-density Asian city (Hong Kong), for the following decedents: 1) all ages, 2) “old-old”: age >= 85, 3) “mid-old”: aged 75 - 84, and 4) “young-old”: aged 65 – 74. Results This study identied 5438 deaths (3771 old-old; 1439 mid-old; 228 young-old) from dementia out of 228,600 all-cause deaths among older adults in Hong Kong between 2007 and 2014. Generally, regional air pollution, being unmarried or female, older age, and daily O 3 were associated with higher dementia mortality, while more urban compactness and greenness were linked to lower dementia mortality. Specically, being unmarried and age effect was linked to higher dementia mortality among the “old-old”, “mid-old” and “young-old”. Regional air pollution was linked to increased dementia mortality, while urban compactness and greenness were associated with lower dementia mortality among the “old-old” and “mid-old”. Higher daily O 3 had higher dementia mortality, while districts with a greater percentage of residents whose native language is not Cantonese was linked to lower dementia mortality among the “old-old”. Economic inactivity was associated with increased dementia mortality among the “young-old”. Gender effect varied by age. The differing strengths of association of various factors with dementia mortality among different age groups implies for health In particular, strategies for usage of greenspace and social and activity to


Introduction
Dementia is a common disorder among the ageing population, particularly in developed countries (Asada, 2012; Currais et al., 2015). Previous studies have indicated an association between demographic structure and dementia (Fisher et al., 2011;Tom et al., 2015; van der Linde et al., 2010). For example, older people are a major vulnerable population with high prevalence of dementia, especially those with an unhealthy diet, low mobility and a low level of physical activity (Beydoun & Kivimaki, 2020;Najar et al., 2019). As activities of daily living as well as mobility could affect both dementia and quality of life (Beydoun & Kivimaki, 2020), numerous studies have also investigated the linkage between dementia and quality of life (Banerjee et al., 2009;Bowling et al., 2015;Hurt et al., 2008), speci cally in the local context of how physical activities improve the wellbeing of older people with this disease (Algar et al., 2016;Scarmeas et al. 2009). In addition, human behaviors, activities of daily living and spatial mobility of older adults could also be affected by the neighboring environment of their residences (Cerin et al., 2013), therefore, this has also led to a research trend of investigating in uences of built and social environments on the prevalence of dementia (Wu et al., 2015). For example, recent studies have found that green space as a built environmental factor reduces dementia and its associated diseases ( Despite the research trend above, the fatal effects of dementia and its relationships with the social and geophysical environment have rarely been discussed, especially in the context of a high-density Asian city. Although mortality associated with dementia may not be as high as other chronic diseases such as respiratory diseases and cancers, there are still remarkable death counts due to dementia in high-density Asian cities. For example, approximately 2.1% of male deaths and 4.0% of female deaths were from dementia in 2016, based on Public Health Information and Statistics of Hong Kong (https://www.healthyhk.gov.hk). Furthermore, dementia was one of the two brain-related diseases recognized among the "ten leading causes of death" for both males and females in the same database.
This indicates that dementia can be a signi cant factor causing fatal effects to those who are aged over 30 and whose dementia is not due to genetic issues. While it is still important to evaluate the survival population with dementia, it is also essential to study dementia mortality in order to better develop protocols for health surveillance and community planning.
Furthermore, as Hong Kong is a city with high-density living and a compact environment (Ho et al., 2019b;Peng et al., 2017), it is expected that "population stress" has been occurring due to the local population growing faster than the available land (Chi & Ho, 2018). Based on the perspective of human ecology, this population growth can also in uence the change of urban form and natural environment (Clement et al., 2015). Therefore, it is important to enhance the understanding of the human-environment nexus across a high-density city, in order to create a "healthy city" that can protect all ages and all vulnerable populations Woo et al., 2017), including people with dementia at different ages.
We hereby develop a territory-wide register-based study to evaluate the potential in uences of social and environmental burdens on dementia mortality in a high-density Asian city (Hong Kong). The objectives of this study include evaluating the links between age-speci c dementia mortality in Hong Kong and 1) individual-level socioeconomic characteristics, 2) community-level socioeconomic characteristics, 3) short-term environmental changes, and 4) long-term environmental deprivation.

Health measures
This territory-wide register-based study applied the Hong Kong Census and Statistics Department's mortality dataset (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014), with information of all decedents recording the following information: 1) date of death; 2) age; 3) gender; 4) occupation; 5) marital status; 6) cause of death registered based on the 10th version of the International Statistical Classi cation of Diseases and Related Health Problems (ICD-10); and 7) location of residence registered at the level of "tertiary planning unit" (TPU), which is a ne spatial unit for regional planning with vital statistics in Hong Kong. Speci cally, there were 287 TPUs within the total land of Hong Kong (area: 1,106 km 2 ) in 2006.

Individual-level socio-demographic measures
A binary measure of gender with 1 for male and 0 for female was classi ed based on the personal information of each decedent. A binary measure of economically inactive was also classi ed based on the information of occupation, with 1 for decedents who were "economically inactive" and 0 representing other decedents. This measure represented the socioeconomic status of each decedent. A binary measure for marital status was used, with 1 for "unmarried" and 0 for "married" to indicate potential social isolation of each decedent. Furthermore, age was recorded as a continuous variable. These individual-level social factors are related to prevalence of dementia (Cadar et al., 2018).

Community-level socio-demographic measures
Community-level socio-demographic measures were spatial varying variables obtained from the 2006 Hong Kong census data recorded at the TPU level, including 1) percent of low-education population (Low education %), and 2) percent of the population whose native language is not Cantonese (Not Cantonese %). In this study, Low education % was de ned as the percentage of people with a primary school education or less. This measure represents a community with low preparedness for health risk due to the level of education. Since the native dialect of the local population is Cantonese, a high percentage of Not Cantonese % indicates a neighbourhood with potentially mixed cultures (e.g. cultural heritage of migrants, and district with visible minorities), which may suffer more problems associated with community health. These community-level socio-demographic measures have been documented to be associated with dementia ( , were used in this study: 1) urban compactness, and 2) greenness.
Speci cally, urban compactness is calculated from an urban sky view factor (SVF) dataset commonly used in local studies (Shi et al., 2018a). SVF is an indicator of unobstructed sky across the urban area ranging from 0 to 1, in which a lower value can represent an area with more obstructed sky due to a highdensity built environment, and a higher value can indicate an area with less obstructed sky. This study rst obtained the average SVF of each TPU and matched the average values to all decedents based on their locations of residence. A measure of building density representing a "decrease in percentage of sky view" (percent of no sky view) was calculated by the following equation: (1 -average SVF) x 100%.
Greenness was estimated by the average normalized difference vegetation index (NDVI) of each TPU. NDVI is a spectral index derived by the infrared and red bands of a satellite image (Tucker, 1979;Yuen et al., 2019), with a range between -1 and 1. In theory, a low value indicates an area with less greenness and a high value represents an area with more greenness. In this study, the NDVI map was a resampled spatial dataset with 15 m resolution estimated by an IKONOS multispectral image. Since NDVI itself was unitless, we de ned a 0.1 increase of NDVI as the unit for the measure of greenness.
Air pollution is mainly contributed by the following two sources: regional and local (tra c-related). Therefore, this study applied two spatially varying measures of air pollution related to long-term impacts of the geophysical environment: 1) regional air pollution, and 2) tra c-related air pollution. Several studies have found associations between these measures and dementia ( . Additionally, a local study has found that approximately 50% of air pollution was regional while 50% was related to tra c emissions (Bilal et al., 2019). In this study, regional air pollution was represented by the spatial distribution of ne particulate matters. Tra c-related air pollution was represented by the spatial distribution of black carbon. Since air quality maps from land use regression ( . Therefore, this study applied six temporally varying measures (without spatial information) related to short-term environmental impacts, including: 1) in uence of hot weather (Hot day), 2) in uence of cold weather (Cold day), 3) daily variation of respirable suspended particulates (daily PM 10 ), 4) daily variation of nitrogen oxides (daily NO x ), 5) daily variation of tropospheric ozone (daily O 3 ), and 6) daily variation of humidity (daily humidity).
Hot day is a binary measure calculated based on the daily average temperature at the headquarters of HKO, with "1" indicating a day with temperature > = 95th percentile (based on long-term weather records between Jan. 1,1971 and Aug. 31,2015) and "0" indicating other days. Cold day is a binary measure indicating days with temperature < = 5th percentile as "1" and the other days as "0". Daily humidity is calculated based on the daily average of humidity obtained from the weather station located at the headquarters of HKO.
Daily PM 10

Register-based study
This study rst selected persons who died from dementia as "cases" and all-cause deaths as "controls" for comparison. The reason for using all-cause deaths as controls was to evaluate whether dementia could be more fatal in situations with various socio-environmental impacts, in comparison with the "normal scenario" (based on all-cause deaths). Binomial logistic regression was applied to estimate the potential impacts of social, built and geophysical environment on dementia decedents, with the use of the glm package in R software. Odds ratios (OR) and 95% con dence intervals (CI) estimated by independent effects of all measures are reported.
Subgroup analysis was then applied to estimate the associations of the socio-environmental factors with mortality at different ages. Decedents were strati ed as follows: 1) "old-old", aged > = 85, 2) "mid-old", aged between 75 and 84, and 3) "young-old", who were aged between 65 and 74. We re-analyzed the links between mortality and the social, built and geophysical environment for these four subgroups of decedents based on binomial logistic regression with the same model structure.
In order to avoid multicollinearity, preliminary variance in ation factor (VIF) analyses have been applied to the above regressions. Based on preliminary tests, VIFs for all independent variables in each regression were lower than 3, indicating the suitability of applying all variables as co-variates in this study.

Data summary
Listwise deletion for missing death date or missing location of residence was rst performed for retrieving the analytic dataset, as these data are missing completely at random. After exclusion of decedents with a missing death date or missing location of residence, the analytic dataset included 5438 deaths of older adults who died from dementia. There were 3771 old-old deaths from dementia. Among the mid-old decedents, there were 1439 deaths from dementia. There were also 228 young-old decedents who succumbed to dementia. In general, the ratio between dementia and all-cause deaths increased as age increased (Table 1). Based on a summary of socio-environmental characteristics (Table 2), dementia deaths in Hong Kong tended to occur among those living in a district with long-term impact of regional air pollution much higher than the standard of the World Health Organization (WHO). According to the WHO air quality guidelines, the acceptable annual mean of ne particulate matters should be lower than 10 µg/m 3 . However, dementia deaths in Hong Kong were associated with ne particulate matter levels at more than three times the threshold.

Socio-environmental impacts on all decedents
Among all decedents, long-term exposure to regional air pollution was linked to higher dementia mortality (OR: 1.244 [1.132, 1.368]) when compared with all-cause mortality, controlling for other social and environmental factors (Table 3). Female and older age were also factors associated with higher dementia mortality. Table 3 Odds ratio (OR) and the 95% con dence interval (CI) for the evaluation of relative impacts of social, built and geophysical environment on age-speci c dementia mortality. Based on each regression, the OR was used to evaluate the difference between dementia and all-cause mortality.

Socio-environmental impacts on old-old decedents
Among the old-old decedents ( Finally, dementia deaths were less likely to be linked to daily PM 10 compared with all-cause deaths.

Socio-environmental factors and mid-old decedents
The associations of socio-environmental factors with dementia mortality among the mid-old decedents were similar to those among all decedents and old-old decedents (Table 3). Speci cally, long-term exposure to regional air pollution, unmarried status and older age were linked to increased dementia mortality when compared with all-cause mortality,

Socio-environmental factors and young-old decedents
Among the young-old decedents, only several individual-level social factors were linked to a difference in mortality risks between dementia and all-cause deaths (

Discussion
Implications of territory-wide register-based study This study showed a signi cant difference in associations of social, built and geophysical environment with dementia and all-cause mortality, and this may indicate that community planning to improve social and environmental resilience among populations with dementia should be tackled with a variety of action plans.
The analyses suggested that social isolation was the biggest problem related to dementia. Speci cally, being economically inactive, unmarried and older are known factors directly or indirectly related to social isolation. More people speaking a foreign language within a district could be a cultural barrier; however, this can also increase the social support within a district, and as a result, this enhances community engagement and social cohesion. Furthermore, although urban compactness can be an indicator of lower environmental quality, it may somewhat re ect an environment with better social cohesion, as urban compactness in a high-rise, high-density city can refer to an urban design with high-rise building complexes that can provide better access to various facilities (e.g. community facilities, local clinics) with multifunctional land use to the local population. More speci cally, a local study has found that higher accessibility of local facilities can enhance cognitive functions of older adults in Hong Kong (Guo et al., 2019). Furthermore, a recent study in the United States has also found that areas with higher urbanization have lower dementia mortality (Kramarow, & Tejada-Vera, 2019). Therefore, an alternative interpretation of the results regarding building density can be an indication of higher mortality risk from dementia in remote areas with higher potential of social isolation. Thus, a better community plan to prevent risk from dementia across a compact city (e.g. Hong Kong) should be to design the city with better social cohesion to sustain resilience among citizens. Speci cally, the concept of "livability" has been widely documented, with the goal of improving the social and physical connectivity among urban residents through urban compactness, in order to maintain a lifestyle of the local population with better quality of life. Based on this concept and the results of urban compactness in this study, public facilities in a high-rise, high-density environment that can maintain the social network among the local population (e.g. community centers) could be further improved, in order to reduce the health burden of older people suffering from dementia.
The present results regarding regional air pollution align with ndings from previous studies that longterm air pollution can adversely in uence neurodegenerative diseases (Chen et al., 2017a;Chen et al., 2017b), while this study further shows adverse effects varying for those with dementia, controlling for age. Speci cally, our result, although not allowing a clear causal conclusion, implies that a long-term increase in dust particles may in uence oxidative stress, neuroin ammation, and neurodegeneration, as a result inducing higher dementia mortality, which is consistent with the hypotheses of a previous study (Calderón-Garcidueñas, & Villarreal-Ríos, 2017).
The results of greenness can be interpreted in two ways. First, urban greenness has been documented to be a protective factor against mental health problems, including dementia and related diseases (e.g. depression). Our study further implies the necessity of applying greenness as a natural-based preventive medicine against dementia. Second, vegetation can reduce local air pollution. This may be a modi er regarding the effect of regional air pollution on dementia. least among the local population in Hong Kong. This could be due to a short-term increase in dust particles (e.g. PM 10 ) inhaled into the lungs inducing a stronger risk of having cardiorespiratory illness than dementia. However, the stronger association between daily O 3 and dementia mortality implies that toxic gas (e.g. ozone) may in uence oxidative stress in the short-term, resulting in higher dementia mortality than that of the other diseases. Therefore, the vulnerable population should be well prepared to improve their resilience in order to prevent any short-term and critical changes caused by extremes in environmental quality. In particular, health education to improve knowledge, attitudes, and practices to better cope with the above environmental hazards from short-term changes caused by extremes in environmental quality should be delivered to the vulnerable population.

Limitations
This study has a lack of information regarding the spatial mobility of each decedent. Several studies have suggested that socio-environmental exposure can be a spatiotemporal component among the urban population, while variations in socio-environmental exposure from multiple locations within cities can affect the degree of impact on each citizen. While the census registry, due to its nature, is not useful to represent this spatiotemporal phenomenon, such spatiotemporal bias is not as signi cant as in studies representing the working population and adolescents, since persons suffering with dementia are usually older people with lower mobility. In addition, older people usually prefer to stay in their own neighborhoods for daily living, and this suggests that our approach may still be appropriate. For future study, a GPS tracker could be used to enhance the measurement of each subject, but this approach can only target the surviving population, not the decedents.
Another limitation is that the database lacks information on speci c types of dementia for many cases, and information related to co-morbidities was not provided by the database either. Two diseases -Alzheimer's disease and rst stroke -may have some overlap in the cause of death resulting from vascular dementia, and vascular dementia can be a result of ischemic stroke (Kalaria et al., 2016).
Therefore, more comprehensive pathways regarding burden of diseases will be provided if information on co-morbidities can be included in the analyses.
Furthermore, this study used community-level information to represent residential location for the estimation of the possible in uence of urban characteristics on dementia mortality, due to the limitation of register-based mortality data. However, some studies have suggested that since older adults are less mobile, their movements should be analyzed with a shorter spatial distance (Cerin et al., 2013), such as 500 m from their homes. Therefore, future studies with population-based cohort data and ner spatial information could be developed to estimate the association at various scales (e.g. 500 m radius versus TPU-level) between relevant urban factors and dementia mortality through multilevel modelling (e.g. a linear mixed effect model). However, the above approach is limited by data collection, especially because the sample size is usually not as large as register-based data from government. Therefore, the approach used in the present study is still appropriate.
Finally, this study followed other local studies (Guo et al., 2019;Ho et al., 2017) in assuming that the older adults may have been staying in their locations of residence for a long period. However, some of the residents may have been moved to another location such as long-term health care facilities. As this is a limitation of register-based data, future studies should apply a cohort design with more information related to residential address for a comparison.

Conclusions
This territory-wide register-based study indicated a difference in the association of various factors with dementia mortality among different ages in Hong Kong. Speci cally, social isolation and regional air pollution may have stronger links to dementia mortality. A short-term increase in daily O 3 is also associated with higher dementia mortality within a day. A comprehensive framework for community health planning should be conducted based on the relative impacts of various diseases in order to reduce vulnerability at all ages.

Declarations
Ethics approval and consent to participate Not applicable.

Consent to publish
Not applicable.

Availability of data and materials
All data generated or analyzed during this study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests.