Our findings suggest that exposure to urban residential greenness reduces mortality from all neurodegenerative diseases, and specifically from Alzheimer’s disease (AD) and dementia. We found no significant associations with Parkinson’s disease (PD) mortality. The observed beneficial effects were potentially mediated by a reduction in air pollution concentrations. Moreover, we found that this protective effect was generally stronger in individuals with lower education. We also found the strongest beneficial associations for overall neurodegenerative diseases and unspecified dementia mortality in individuals residing in the most deprived neighbourhoods. In contrast, for AD mortality, the strongest beneficial association was observed in wealthier neighbourhoods.
Our results regarding reduced premature mortality from neurodegenerative diseases with increased surrounding greenness are comparable to those reported by Klompmaker et al. (2021) in a study over 10 million adults (aged ≥30 years). Using the same classification of neurodegenerative diseases, the authors observed a 2% reduction (per 0.14 increase) in neurodegenerative disease mortality with surrounding greenness. Stratified by age, results were similar in the elderly (≥65 years) . An ecological study in Greece also found a significant inverse association (HR 0.91) between greenness and mortality from diseases of the nervous system (ICD-10 codes G00-G99) . Two longitudinal studies found a protective but non-significant association with neurodegenerative disease mortality [22, 23]. James et al. (2016) included 108,630 elderly female nurses followed between 2000-2008, yielding an HR of 0.93 per 0.1 increase in surrounding greenness. Klompmaker et al. (2020) used survey data of 339,633 individuals linked with mortality data (2003-2007), and showed a similar effect estimate as Klompmaker et al. (2021), but with wider confidence intervals. Potentially these studies did not find significant associations because of an insufficient statistical power due to smaller study populations combined with shorter mortality follow-up periods.
Our study findings were consistent for different dementia subtypes. Two of the abovementioned studies additionally assessed dementia mortality (all dementia types, ICD-10 codes: F00-F03). Klompmaker et al. (2021) showed a significant 4% reduction in the risk of dementia mortality with increased surrounding greenness. Klompmaker et al. (2020) reported a similar but non-significant effect. Other longitudinal studies using health administrative databases found a beneficial association between green spaces and an outcome including all dementia types and AD (ICD-10 codes: F00-F03, G30, respectively) [16, 17]. However, studies evaluating the effect of green spaces on AD alone have shown contradictory findings [18–21]. Moreover, we are not aware of prior studies evaluating the association between green spaces and vascular dementia. Thus, further research is needed to confirm our results.
Suggested mechanisms underlying the direct beneficial effect of exposure to green spaces on neurodegenerative disease mortality include inducing psychological restoration and reducing stress , potentially preventing depression, a risk factor for dementia . Additionally, there is suggestive evidence that green spaces could increase physical activity in the elderly , reducing the risk of dementia and AD . Social isolation is moreover associated with an increased risk of cognitive decline and dementia , and greener neighbourhoods could enhance social cohesion and mitigate feelings of loneliness in older adults [54, 55]. Finally, green spaces could contribute to the mitigation of environmental hazards, including air pollution . Our findings suggest that the associations between residential greenness and neurodegenerative disease mortality are partly mediated by a reduction in air pollution. We only reported results with PM2.5. In alternative models with NO2 (results not shown), we observed similar findings. So far, only one longitudinal study on green spaces and cognitive function explored potential mediation by air pollution, but no evidence of mediation effects was found . Current evidence establishes a strong link between air pollution and AD and dementia . The filtering effect of green spaces removing pollutants from the atmosphere has been proven to be generally small [57, 58]. However, green spaces could decrease temperature in cities , indirectly improving air quality by reducing the generation, transportation and toxicity of pollutants . Likewise, fewer sources of air pollution are found in greener areas . Noise pollution or proximity to roads have also been associated with an increased risk of dementia independently from air pollution levels [3, 4], but we unfortunately lacked such information.
As part of the sensitivity analyses, we used perceived neighbourhood greenness as alternative exposure indicator, where we found no association with AD nor vascular dementia mortality and a stronger association with unspecified dementia compared to the results of our main models using surrounding greenness. We suspect that perceived neighbourhood greenness may partly capture certain aspects of neighbourhood socioeconomic position (SEP). As such, this model may be overadjusted for SEP. Our findings are probably explained by individuals with high SEP having a higher likelihood of getting a record of dementia aetiology in death certificates , and, moreover, residing more often in areas where a higher proportion of individuals report very good neighbourhood greenspace provision.
No beneficial association between greenness and PD mortality was found, probably a result of the differing aetiology of PD compared to that of the other neurodegenerative diseases under study . Additionally, in sensitivity analysis the associations with PD changed direction and became beneficial when commuting zone residents were excluded. Morphologically, these areas are characterized by an extensive land use in both housing and commercial activities . Thus, we may speculate that increased risk of PD mortality could partly be explained by exposure to agricultural land, potentially encompassing exposure to pesticides, although this has been mainly explored for agricultural workers , being the evidence available for residential exposures currently limited . Furthermore, we were not able to further explore this hypothesis since we lacked data on different types of green such as agricultural land.
Comparison between different population subgroups in the stratified analyses should be done with caution, given statistical restrictions in such interpretations. We observed strongest beneficial effects of living near greener areas in the lower educated for all neurodegenerative diseases and AD mortality. No clear patterns were found for other studied causes. Similarly, the longitudinal study of de Keijzer et al. (2018) did not find consistent evidence for differences between different education groups in the association with residential greenness and cognitive decline. Regarding neighbourhood SEP, we observed a general trend for all neurodegenerative disease and dementia mortality, where strongest associations with residential greenness were found in more socially deprived areas. The aforementioned study by de Keijzer et al. (2018) observed similar patterns with cognitive decline. Lower SEP has been linked to both poorer living conditions and limited access to resources which may be related to increased risk of cognitive decline in later life . Additionally, availability of residential green has been associated to reduced risk of mortality, where health benefits seemed to be largest among most deprived population groups . Such gradient was partly confirmed by our study results, although for AD mortality we found strongest beneficial associations in individuals residing in wealthier areas. This contradicts the findings from Brown et al. (2018), in which a trend by neighbourhood SEP in the association between greenness and AD prevalence was found, showing the strongest estimate in low-income neighbourhoods.
Our study comes with several limitations. The major limitation of our study is that we were not able to control for lifestyle factors, e.g., body mass index (BMI), smoking status and alcohol consumption, which are well-known risk factors for AD and dementia . Prior studies that were able to account for these factors observed a small attenuation in the association [22, 23]. However, comparability of findings after such adjustment is challenging given several important differences in study design, population characteristics and exposure assessment. Hence, the direction of potential bias in our effect estimates remains unclear. Furthermore, our study did not include time-varying variables of exposure throughout the follow-up period. We only had one measure of surrounding greenness for the year 2006, close to the middle of the follow-up period, which is another limitation of our study. We assumed that, although the quantity of green spaces may vary across time, their spatial distribution remains relatively stable. However, no other exposure information was available for other years to test this. Moreover, exposure assessment was based on the geocoded residential address at baseline (2001), and we lacked information on residential mobility during follow-up (2001-2014). Still, we were able to limit the analyses to a group of residents who did not move in the last 10 years prior to baseline, which did not invalidate our main results. Surrounding greenness captured all types and sizes of green spaces, independently of these being private or public. Also, limiting greenspace exposure assessment to the residence may result in exposure misclassification, as it potentially ignores exposure in other life spheres, e.g., working place. Similarly, episodes of nature interaction were not measured in quality (e.g., accessibility, type of use) nor in time (e.g., frequency, duration). We also relied on baseline information of sociodemographic characteristics, but these may not vary considerably among the elderly population. Applied missing values techniques present limitations given that the missing values may not be completely at random, which potentially affects generalizability of our findings. However, by using two approaches to handle missing data (i.e., multiple imputation and listwise deletion), we may assume that main conclusions are fairly robust. Lastly, we excluded the institutionalized population (i.e., care homes) to minimise selection bias, given that almost half of individuals diagnosed with neurodegenerative diseases live in institutions .
Notwithstanding the limitations described, our study counts with an important number of strengths. We analysed the association between surrounding greenness and neurodegenerative diseases over a longer follow-up period (13.25 years) than prior studies [22–24]. We also relied on a high resolution environmental database, linked to the geocoded residential address of each individual officially residing in the five largest Belgian urban areas at baseline . Our study was also the first to conduct mediation analyses in the studied associations by air pollution concentrations using individually linked exposure data. Finally, using large administrative data allowed us to study effect modification by gender and socioeconomic characteristics through stratification for representative subgroups.