Exposome factors associated with multi-morbidity in older adults: a discovery-based cross-sectional analysis in the Longitudinal Aging Study Amsterdam

Abstract

Ageing populations and longer life expectancies challenge healthcare systems due to rising noncommunicable diseases (NCDs) and multi-morbidity. Designing healthier living environments may reduce population risks of NCD onset, but knowledge is needed on environmental factors related to NCDs in older adults. We therefore examined associations between residential neighbourhood built, physico-chemical, and sociodemographic environmental factors and NCD prevalence in the Netherlands among older adults. Cross-sectional data from 1578 older adults from the Longitudinal Aging Study Amsterdam (2008–09) were matched with environmental data from the Dutch Geoscience and Health Cohort Consortium (GECCO). Multivariable logistic regression analyses were conducted to assess the odds of having a single NCD versus no NCD and multi-morbidity versus no NCD, adjusted for sociodemographic factors. Participants had a mean age of 73.2 years, 55% were female, and 77% reported at least one NCD. Multi-morbidity was more common in women, who were also older and had lower education and income. Higher green space density within 500 m was significantly associated with lower odds of single NCD [odds ratio (OR): 0.52, 95% confidence interval (CI): 0.33–0.83]. A higher number of cars in a household also showed lower odds of single NCD (OR: 0.14, 95% CI: 0.04–0.55). No significant associations were observed for physico-chemical exposures. Results were consistent in sensitivity analyses. The findings underscore the need for urban planning and policies that support healthy ageing while embracing a longevity-ready perspective, accounting for the built, physico-chemical, and sociodemographic environments across the life-course.

Introduction

As populations age, healthcare systems face growing strain due to increased reliance on medical care and long-term support [1]. This challenge is heightened by ‘dual ageing’, a simultaneous rise in the proportion of older adults and life expectancy, leading to higher rates of noncommunicable diseases (NCDs) and multi-morbidity—the latter is here defined as having two or more NCDs [2]. Multi-morbidity in older adults is linked to greater healthcare use, reduced quality of life, and higher mortality, underscoring the unsustainability of relying solely on healthcare systems [3]. Preventive action targeting modifiable upstream determinants, such as environmental exposures, is essential for healthy ageing [4].

The built environment—including structures, parks, and public transport—shapes everyday behaviours like physical activity and diet, which are critical for NCD management and prevention [5]. Since most older adults are active within their neighbourhoods, age-friendly design becomes increasingly important as mobility declines [6]. Meanwhile, physico-chemical exposures such as air pollution, noise, and extreme temperatures are linked to respiratory and cardiovascular diseases [7, 8]. An umbrella review of meta-analyses found fine particulate matter (PM_2.5) to be associated with 29 disease and mortality outcomes, including type 2 diabetes and cardiovascular mortality [7]. Other atmospheric pollutants, such as nitrogen dioxide (NO₂), ozone (O₃), and sulphur dioxide (SO₂), also increase NCD risk [7, 9]. Ambient noise and temperature extremes heighten risks of hypertension, type 2 diabetes, and mortality, especially among older adults [7, 8]. Sociodemographic factors such as socioeconomic status (SES) influence where people live and their access to health-promoting amenities [10].

The exposome concept offers a comprehensive view of exposures—biological, behavioural, and environmental, across the life-course. This framework allows us to consider multiple environmental influences, from individual factors like ageing and health behaviours, to various environmental factors such as green space, air quality, and home values [11]. A previous study showed inconsistent results by sex: where higher air pollution and temperatures, and lower greenness, were associated with increased type 2 diabetes prevalence among men, but not women [12]. Residual confounding of menopause-related metabolic changes and gendered exposure patterns may explain such differences [12, 13]. Likewise, exposures may vary by age, as older adults of more advanced age may spend more time indoors, which can reduce contact with pollutants but limit access to green space and activity [14].

Despite growing attention to environmental influences on health, research focuses heavily on biological risk factors or on individual exposures [15]. Studies also assess specific NCDs, rather than multi-morbidity more broadly [16]. Here, we define a ‘single NCD’ as any one NCD, regardless of type. By adopting a broad outcome perspective, we aim to explore how built, physico-chemical, and sociodemographic environmental factors are related to both single NCDs and multi-morbidity among older adults in the Netherlands. We also examine potential variation in these associations by sex and age group.

Methods

Study design and setting

We conducted a cross-sectional study using data from the ongoing Longitudinal Aging Study Amsterdam (LASA), enriched with environmental exposure data from the Dutch Geoscience and Health Cohort Consortium (GECCO) [17–19]. LASA focuses on physical, emotional, cognitive, and social functioning in adults aged 55+ in the Netherlands [17]. Participants were randomly selected from municipal registries in three regions in the Netherlands, varying in religion and urbanization to ensure a representative sample. Participants provided informed consent, and this study was approved by the VU Medical Center ethics committee.

Participants

The analysis used data from the 2008–09 wave of the LASA, including participants from both the original (1992–93) and second (2002–03) LASA cohorts [17]. Participants’ residential addresses were geo-matched to environmental data from the GECCO dataset. Of the 1818 respondents in the 2008–2009 wave, 1,601 completed the main interview and 217 completed the telephone interview. Participants with a telephone interview were excluded from the analysis due to incomplete coverage of relevant topics. The final analytical sample consisted of 1578 participants who were successfully geo-matched, after excluding 21 with unmatched addresses and 2 residing abroad. No sampling weights were applied, as the LASA cohort was designed to be representative of the Dutch older adult population through sampling across culturally and geographically diverse regions [20]. Previous analyses have shown that mortality rates in the LASA cohort closely align with national statistics, supporting its representativeness [20]. In addition, the distribution of sociodemographic variables used in the analyses reflects national-level statistics. Given the analytical focus on associations rather than population-level estimates, weighting was not performed.

Exposures

The study included 24 environmental exposures from various national sources, resulting in 50 variables, provided by GECCO, grouped into exposures from built, physico-chemical, and sociodemographic environments (Table 1). Exposures were mapped using participants’ residential addresses as centroids of 500 m Euclidean exposure buffers for primary analysis and 1000 m for sensitivity analysis. Other exposures were linked by administrative neighbourhood boundaries or four-digit postal code areas. The Z-scores were calculated across the Netherlands, whereas LASA participants are representative of three specific regions. As a result, their distribution may differ from the national average, leading to deviations from a mean of 0 and an SD of 1 (Supplementary Table S1).

Table 1.

Description of exposure data, sources, and resolution

Exposures	Source	Description	Resolution
Built environment
Density of green space within 0.5 km (Z-score)	CBS	Obtained by aggregating Z-scores of land use data of trees, shrubs, and low vegetation.	25×25m (0.5 km and 1 km buffers)
Land use mix entropy index 0.5 km (Z-score)	CBS	Calculated as the sum of Z-scores of different land use classes: residential, commercial, social-cultural services, offices and public services, green space, and recreation.
Density of public transport stops within 0.5 km (Z-score)	ESRI	Assessed as the sum of the Z-scores of the public transport network in the Netherlands (bus, ferry, metro, taxi, tram).
Density of sports facilities within 0.5 km (Z-score)	The Mulier Institute	Calculated as the sum of Z-scores for sports facilities requiring significant physical effort.
Driving destination accessibility index score (0–100)	GECCO	Ease of reaching different types of destinations by car, based on a weighing system for areas that are more suitable for active transportation or walking. Index values were normalised to a scale of 0 (low drivability) to 100 (high drivability).	100 × 100 m
Accessibility of neighbourhood facilities (km): Distance to general practice Distance to daycare centre Distance to swimming pool Distance to secondary education Distance to library	CBS	Distance (km) to public amenities (nearest medical, recreational, or educational facilities) in the neighbourhood.	Administrative neighbourhood
Physico-chemical environment
Ammonia (NH3) (μg/m3)	RIVM, ALO	Annual average concentrations of air pollutants modelled by Land-Use-Regression models (ESCAPE) and a combination of dispersion model calculations and measurements (data source RIVM).	1 × 1 km
Sulphur dioxide (SO2) (μg/m3)
Nitrogen oxides (NOx) (μg/m3)
Particulate matter 2.5 micrometres (μg/m3)
Monthly temperature (July) (oC)	KNMI	Monthly July temperature (oC) data was interpolated based on 10 automatic monitoring stations.	25 × 25 m
Daily noise levels (dB(A))	PBL	Daily levels of noise (road, rail, and air) were modelled and expressed as Lden (level day-evening-night) in decibels.	25 × 25 m
Sociodemographic environment
Age groups (%): Neighbourhood age 0–14 years Neighbourhood age 15–24 years Neighbourhood age 45–64 years	CBS	Shares of residents in different age groups (%).	Administrative neighbourhood
Marital status (%): Neighbourhood marital status: separated Neighbourhood marital status: widowed		Shares of marital status among residents in the neighbourhood (%).
Immigration status (%): Immigrants from non-western countries Immigrants from western countries		Shares of immigrants from Western and non-Western countries (%).
Liveability score	Dutch Ministry of the Interior and Kingdom Relations	Neighbourhoods categorised into liveability classes: 1 = Very insufficient to 9 = Excellent, and recategorized to ≤5 as Poor, and >6 as Good.	100×100 m
Neighbourhood income: Incomes ≤40th percentile of national distribution (%) Incomes ≥80th percentile of national distribution (%) Passenger cars per household Motorcycles at a neighbourhood level Households below or around social minimum (%) Income per Inhabitant (×1000€) Home values (×1000€)	CBS	Different measures of neighbourhood income: Residents with high or low incomes compared to the national income distribution (%). Count of ownership of passenger cars and motorcycles per household. Households below or around the social minimum income threshold (%), income per inhabitant (×1000€), and home values (×1000€).	Four-digit postal code neighbourhoods

LOCATUS (a commercial retail information provider), CBS (Statistics Netherlands), ESRI (the Environmental Systems Research Institute), The Mulier Institute (scientific sport-research institute in the Netherlands), GECCO (the Geoscience and hEalth Cohort Consortium), RIVM (Institute for Public Health and the Environment), ALO (Living Environment Atlas), KNMI (Royal Netherlands Meteorological Institute), PBL (the Netherlands Environmental Assessment Agency). More comprehensive details on the meta-data are available [18, 32].

Outcomes

The outcomes were single NCD (versus no NCD) and multi-morbidity (versus no NCD), where NCD is defined as a disease in which symptoms and/or treatment had been present for at least 3 months at the time of the interview, self-reported by participants, and multi-morbidity, defined as the presence of two or more NCDs. Participants were categorized into three groups: no NCD, single NCD, and two or more NCDs (multi-morbidity). The reported NCDs were classified as follows: chronic lung diseases, cardiovascular diseases, peripheral arterial disease, stroke, type 2 diabetes, osteoarthritis, rheumatoid arthritis, and cancer. Our analysis was based on the number of NCDs reported per participant and did not include information on which specific conditions each participant had.

Covariates

Covariates considered as potential confounders included age, sex, educational level, and net household income. Age was recorded at the time of the interview and analysed as both a continuous and categorical variable. For categorical analysis, age was divided into two groups based on the median of the distribution: 60–71 and 72–100 years. This stratification balanced participant distribution and provided insights into age-related health differences. Sex was classified as male or female, assuming a cisgender population. Educational attainment was categorized into three levels: low (incomplete or elementary education), middle (lower vocational to intermediate education), and high (secondary to university education). Household net income was categorized into three brackets: less than €1135, €1135–1816, and more than €1816. These cut-offs were based on the distribution of income in the study sample to reflect meaningful socioeconomic differences, while maintaining reasonably balanced group sizes. Reported income was based on respondents’ selection from 12 pre-defined income categories, which were converted to their class medians. For participants living with a partner, household income was adjusted using a 0.7 standardization factor to enable comparability with single-person households, consistent with established LASA methodology [21]. The selected brackets align approximately with the Dutch net modal income in 2007 [22].

Statistical analysis

Data analysis was conducted using R version 4.3.2. Most variables had <6% missing data, except income (12.9%), food environment score at 500 m (16.3%), and liveability score (22.8%). Together, these missing data contributed to 35% of cases being incomplete. To address this, multiple imputation by chained equations was employed. A total of 35 imputed datasets were generated and pooled for analysis, following the guideline that the number of imputations should be at least equal to the percentage of incomplete cases [23]. Baseline characteristics were summarized with mean (standard deviation) or median (interquartile range), and bivariate analyses used Pearson’s chi-squared, Kruskal–Wallis tests, and Wilcoxon rank sum tests.

Multicollinearity was managed by retaining variables with less missingness from highly correlated pairs (r > 0.80) and excluding those with a variance inflation factor (VIF) > 10. Linearity was verified with scatterplots, and no influential points (Cook’s distance > 0.5) were found. Errors were independent, and model fit was confirmed using the likelihood ratio chi-square and McFadden’s pseudo-R² tests against null models across all analyses.

Multivariable logistic regression was used to assess associations between environmental exposures at a 500 m buffer and having either a single NCD versus no NCD and having multi-morbidity versus no NCD, with individuals having no NCDs as the reference group. This approach allows the analysis to focus on a reference population presumed to represent ‘healthy’ individuals with no NCDs. Model A assessed the impact of individual exposures, while Model B combined 31 exposures into one comprehensive model. The models were adjusted for key covariates: age, sex, educational level, and household income. Results were presented as odds ratios (ORs) with 95% confidence intervals (CIs).

As secondary analyses, we ran Model B stratified by sex and by age group to gain insights into differences between subgroups. Sensitivity analyses included the investigation of a 1000 m buffer, comparing those with no NCDs to those with any NCD (having one or more NCDs), comparing those with one NCD to those with multi-morbidity, and using non-imputed datasets. Bonferroni correction was applied to adjust for multiple comparisons.

Results

Descriptive statistics

Among the 1,578 respondents included in the analytical sample, 55% were female, and the average age was 73.2 (SD 8.6) years (Table 2). Educational levels were not evenly distributed: 25% had low, 36% middle, and 38% high education. Approximately 40% of respondents were in the higher income category, earning more than €1816 per month. Overall, 77% reported at least one NCD (36% reported a single NCD, while 41% reported multi-morbidity). Statistical tests showed statistically significant associations between NCD status and sociodemographic variables, including sex, age, education, and income (Supplementary Table S2). Respondents with multi-morbidity were more likely to be female and had the highest average age (75.5 years, SD 8.4). Further, statistical tests indicated significant differences in NCD prevalence, education, and income by sex (Supplementary Table S3). Males were more prevalent in the no NCD group. Compared to males, females were on average older (mean age 73.9, SD 8.8), had a higher prevalence of multi-morbidity, and reported a lower household net income (Supplementary Table S3). Notable sex differences were also observed in educational attainment: 52% of males had high education compared to 27% of females, while 32% of females had low education versus 17% of men (Supplementary Table S3). Descriptive statistics of environmental exposures are detailed in Supplementary Table S1.

Table 2.

Characteristics of study participants

Characteristic	N	N = 1578 (100%)a
Sex (female)	1578	875 (55%)
Age	1578	73.2 (8.6)
Educational attainment	1578
Low education		401 (25%)
Middle education		570 (36%)
High education		607 (38%)
Monthly household net income	1375
Low <1135 €		327 (24%)
Middle 1135 €–1816 €		510 (37%)
High >1816 €		538 (39%)
Noncommunicable disease prevalence	1574
No NCD		355 (23%)
One NCD		572 (36%)
Two or more NCDs (multi-morbidity)		647 (41%)

^aMean (standard deviation); n (%).

Association with noncommunicable diseases

The final associations between having a single NCD or multi-morbidity versus no NCD (Model B) are visualized as forest plots (Figs 1 and 2). Given the discovery-based nature of this study, our primary aim was to identify potential associations that merit further investigation, with statistically significant findings offering stronger support for such links.

Figure 1.

Forest plot of the multivariable logistic regression of all environmental exposures and no NCD versus single NCD.

Figure 2.

Forest plot of the multivariable logistic regression of environmental exposures and no NCD versus multi-morbidity.

Built environment variables

A higher density of green space at a 500 m buffer was associated with a lower odds of having a single NCD (OR: 0.52, 95% CI: 0.33–0.83), and remained significant after Bonferroni’s correction for multiple testing. There was a similar direction but non-significant trend for multi-morbidity (OR: 0.81, 95% CI: 0.53–1.24) (Supplementary Table S4). This trend remained in the no NCD versus any NCD analysis (Supplementary Table S6). This was consistent for females and stronger for those aged 72–100 (Supplementary Tables S8 and S9). However, this was not observed at the 1000 m buffer or when comparing a single NCD to multi-morbidity (Supplementary Tables S5 and S7). Other built environment exposures did not show statistically significant associations with single NCD or multi-morbidity.

Physico-chemical environment variables

No statistically significant associations were found for physico-chemical environment exposures and the odds of single NCDs or multi-morbidity. Although estimates for these variables were directionally consistent across different model comparisons (Supplementary Tables S4, S6, and S7), all CIs included the null, and thus no meaningful associations can be reported based on these results.

Sociodemographic environment variables

Among the sociodemographic environment variables examined, only households with a higher number of cars demonstrated a statistically significant association, being linked to lower odds of having a single NCD (OR: 0.14, 95% CI: 0.04–0.55) (Supplementary Table S4). Stratified analysis suggested this association was primarily observed among males (Supplementary Table S8). Although similar directions were seen in other comparisons and buffer sizes, these were not statistically significant (Supplementary Tables S5–S7). Variables such as motorcycle ownership and house values appeared to show ORs of 1.00 with tight CIs (e.g. OR: 1.00, 95% CI: 1.00–1.00); however, further inspection revealed that this was due to rounding. These variables were retained for completeness but were not further interpreted.

Findings from the non-imputed dataset sensitivity analysis remained consistent with the primary analysis findings (Supplementary Table S10).

Discussion

We used a discovery-based approach to examine associations between various exposome factors and NCD prevalence among older adults in the Netherlands, stratified by sex and age. Our findings highlight the role of greenery density and neighbourhood-level car ownership in shaping NCD outcomes. Furthermore, several demographic variables were relevant in describing our population and those reporting higher multi-morbidity.

Educational attainment was unevenly distributed, with only one-quarter of respondents classified as having low education and nearly 40% classified as having high education. Men were more likely to have a higher education, and women were more likely to have a lower education. Women were also older, had lower household incomes, and reported more multi-morbidity. These descriptive findings may reflect broader structural inequalities during this time.

Results suggest that higher greenery density is associated with lower odds of having a single NCD, with a similar trend observed for multi-morbidity, among both men and women. Previously, greenery has been shown to reduce outdoor temperatures while also minimizing exposures to urban hazards like some air and noise pollutants, thereby lowering the risk of NCDs [7, 24–27]. Further, having access to green spaces has previously been found to promote physical activity and foster social interaction that supports physical and mental well-being [27, 28]. This effect was particularly strong for the older adult group (aged 72–100) and was more pronounced within a 500 m buffer compared to 1000 m, suggesting that ‘older’ older adults benefit more from nearby greenery potentially due to limited mobility compared to their ‘younger’ older adult counterparts [6, 29]. The less pronounced association with multi-morbidity could suggest that greenery primarily aids in preventing initial NCDs.

Our findings show an association between households with high car ownership and lower odds of NCDs. This association may reflect additional dimensions of socioeconomic advantage not fully accounted for, serving as a proxy for higher SES. In the Netherlands, the costs of owning and maintaining a vehicle are among the highest in Europe [30]. This association likely reflects the advantages of living in more affluent neighbourhoods, where access to health-promoting features, such as proximity to essential services and green spaces, is more common, while exposures to environmental toxins are higher [27, 31]. SES exerts a pervasive influence on health, often creating systematic barriers that individuals with lower SES cannot overcome, resulting in disproportionate exposure to adverse living conditions. The use of this indicator allowed us to capture the influence of socioeconomic factors, beyond what traditional SES measures might reveal. These findings underscore the need for equitable public health policies that ensure access to healthy environments is not limited to those with higher SES. This is particularly critical for older adults, many of whom depend on fixed incomes and have limited capacity to improve their SES later in life.

Other environmental variables included in our analyses did not show meaningful associations with NCD outcomes. These non-significant findings are presented in the forest plots and supplementary tables for transparency and completeness. Although not discussed in depth, they may offer useful directions for future research and should be interpreted cautiously, particularly given the discovery nature of this study.

While our study focused exclusively on older adults, our findings contribute to a broader understanding of how the urban environment, particularly green space density, can support or hinder healthy ageing. Wang et al. state that contact with green spaces in childhood has several benefits and having a higher access to green spaces during childhood and adulthood may contribute to more successful ageing, in line with the benefits recognized by exposure to greenery [27]. Wang et al further advocate for the development of ‘longevity-ready’ cities, which aim not only to accommodate older adults but to proactively shape environments from childhood onward to support long-term health and well-being. Moreover, there are evolving risks posed by climate change, such as increased heatwaves and pollution, making it clear that the urban physico-chemical environment is not static [27].

This analysis uses the exposome concept to explore various environmental exposures comprehensively within a unified model. However, the inclusion of numerous exposures challenged the statistical power. A larger sample size would likely have helped to address this issue. Various exposures, such as food environment and walkability, were excluded due to high multicollinearity, which limits the exploration of all planned factors. Despite this, the approach allowed for the simultaneous interpretation of multiple exposures, important for real-world interventions like urban planning. The data, being from 2008 to 2009, may not fully reflect current conditions, affecting the generalizability of the findings. Thus, these should be interpreted with caution regarding their applicability to today’s older adult population and their current experiences.

This analysis compared the prevalence of single NCD and multi-morbidity versus no NCDs in older adults. While self-reported NCDs may have a response bias due to respondents interpreting the definition of NCD differently, a strict NCD definition likely mitigated this issue. Future research should use objective measures such as data obtained from health registries to validate self-reports. Although many results were statistically non-significant with small effect sizes, their potential impact may be significant given the broad influence of environmental factors on entire populations. Given this large reach, even minor effects can have considerable public health implications.

The methodological approach of this study is a key strength, as it comprehensively explores real-life exposures as suggested by the exposome concept. Unlike studies that examine individual factors in isolation, our approach captures a broad range of exposures. Further, our study extends the scope of research by analysing both multi-morbidity and any single NCD instead of focusing on a specific NCD.

Given the discovery nature of this study, further research in different settings and using longitudinal study designs is needed to better understand how environmental and sociodemographic factors influence NCD odds among older adults. Importantly, our findings highlight that even when accounting for a broad range of environmental exposures, higher neighbourhood greenery density and households with a higher number of cars emerged as significant protective factors, underscoring their potential relevance for healthy ageing. While other exposures did not show statistically significant associations, their inclusion contributes to a comprehensive exposome-informed approach. Furthermore, although our study focused on older adults, a longevity-ready perspective supports designing urban environments with intergenerational benefits, to reduce the burden of NCDs and promote health longevity throughout the life-course. This perspective calls for greater community engagement, including input from all age groups, in shaping urban futures. Our findings on older adults can thus be seen not only as a call to improve current conditions for our older community, but also as evidence to support early and sustained investment in healthier urban design from the outset of life.

Supplementary data

Supplementary data are available at EURPUB online.

Conflict of interest: None declared.

Funding

This project was partly funded by STAGE. STAGE has received funding from the European Union’s Horizon Europe Research and Innovation Programme under grant agreement n°101137146. UK participants in Horizon Europe Project STAGE are supported by UKRI grant n°10112787 (Beta Technology), n°10099041 (University of Bristol) and n°10109957 (Imperial College London). Geo-data were collected as part of the Geoscience and Health Cohort Consortium (GECCO), which was financially supported by the Netherlands Organization for Scientific Research (NWO), the Netherlands Organization for Health Research and Development (ZonMw), and Amsterdam UMC. More information on GECCO can be found on http://www.gecco.nl.

Data availability

Individual-level data from the LASA cohort study cannot be provided due to privacy regulations. Environmental-level data from GECCO can be provided upon reasonable request (www.gecco.nl).

Key points.

• Higher residential greenery density was associated with lower odds of NCDs in older adults, reinforcing the health benefits of green space in later life.

• Findings were more pronounced for exposures measured within smaller buffer zones, suggesting that immediate local environments may play a more critical role in shaping health outcomes in older populations.

• The association between environmental exposures and health outcomes differed by sex and age group, underscoring the need for tailored, life-course sensitive public health strategies.

• Neighbourhood-level household car ownership, used as a proxy for socioeconomic status, revealed significant health disparities, highlighting the need for equitable urban and public health planning that prioritizes older adults regardless of socioeconomic background.