Living environments and mental symptoms and disorders

Abstract

Background

We live in different environments with distinct effects on mental health. However, the complex associations between living environments and mental symptoms and disorders remain elusive. This study aimed to comprehensively investigate how multiple environmental factors relate to mental symptoms and predict the onset of psychiatric disorders.

Methods

Using data from the UK Biobank, we systematically characterized ten living environmental exposures, encompassing both built and natural environment features. We comprehensively examined their associations with symptoms of anxiety, depression, and bipolar disorder, evaluating not only the overall effects but also stratified effects across sex and age subgroups to capture potential demographic-specific vulnerabilities. Furthermore, we investigated the predictive value of these baseline environmental exposures for the subsequent onset of eight major psychiatric disorders over time. To enhance precision, all analyses were complemented by subgroup-specific models, allowing us to identify both shared and distinct patterns of environmental influence across demographic strata.

Results

We identified 18 significant associations between environmental exposures and baseline mental symptoms, with area deprivation showing the most significant associations with depression and anxiety symptoms. Destination accessibility and slope emerged as novel protective factors, while area deprivation and air pollution accounted for more than 80% of the environmental contribution to model explanatory power. One sex-specific and three age-specific associations were observed, particularly for street network accessibility and area deprivation. Longitudinal analyses identified 25 significant associations between environmental predictors and incident mental disorders, involving eight environmental factors across eight disorders. Specially, destination accessibility was associated with a 56.6% reduced risk of schizophrenia. Two sex-specific and four age-specific predictions were also found.

Conclusions

This study uncovers previously unrecognized potential environmental factors associated with mental health outcomes and highlights distinct vulnerability patterns across demographic groups. By elucidating these nuanced associations, our findings offer a foundation for precision public health initiatives and underscore the potential of evidence-informed urban and environmental planning to mitigate mental health risks at the population level.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12888-026-08383-5.

Background

Mental symptoms and disorders are among the leading causes of disability worldwide and affect individuals across the lifespan. Nearly half the people are expected to develop one or more mental disorders by the age of 75 years [1], highlighting a profound burden on individuals, healthcare systems, and society at large. While genetic predisposition contributes to risk, increasing evidence underscores the pivotal role of modifiable environmental exposures, including both personal behaviors and broader living conditions, in shaping mental health outcomes [2]. Of particular interest are multidimensional living environments (e.g., urbanicity, air pollution, and greenspace) have been associated with mental symptoms [3–5]. Identifying and specifying the precise relationships between living environments and mental health outcomes are essential, as living environments affect all people and are modifiable, and thereby mental health can be improved by minimizing risk and enhancing protective exposures.

Living environments refer to all kinds of external physical, social, and economic surroundings that individuals are exposed to in their living areas, consisting of diverse but correlated exposures, such as area deprivation, air pollution, noise pollution, traffic intensity, greenspace, blue space, as well as destination and street network accessibility. Mental symptoms and disorders also include subdomains and subclasses. For example, mental disorders include anxiety, depression, bipolar disorder, addiction, schizophrenia, dementia, and others. In terms of the relationships between living environments and mental symptoms and disorders, previous studies have reported important associations [6–10], such as the associations of air pollution, residential greenness, and temperature with mental symptoms. However, several limitations in these previous association studies between living environments and mental health outcomes should be addressed. First, it remains unclear about the effects of several potentially important living environments (such as destination and street network accessibility) on mental health outcomes, which may discover new potential intervention-relevant factors. Second, few studies have combined the cross-sectional association analysis and the longitudinal prediction analysis to identify living environmental exposures with reliable effects on mental health outcomes, which may indicate reliable targets for prevention. Third, few studies have investigated the relative contribution of living environmental factors on the same mental health outcome, which are essential for designing precise and cost-effective interventions. Fourth, few studies considered the sex- and age-related variations (shared and distinct associations) in the association between living environments and mental health outcomes, which are crucial for planning precise interventions. Addressing these limitations is essential to advance our understanding of the associations between living environment and mental health, informing precise and cost-effective interventions for mental health promotion.

To address these critical gaps, we analyzed data from up to 430,499 participants in the UK Biobank [11], investigating the overall, age-specific, and sex-specific effects of ten living environmental exposures on three mental symptoms and eight mental disorders. We considered nine categories of living environmental exposures and used the confirmative factor analysis (CFA) to integrate the categories with multiple exposures, generating ten environmental assessments (five single and five composite exposures). Based on the baseline data, we investigated the overall effects of these environments on anxiety, depression, and bipolar symptoms in the total sample, and then investigated the sex and age shared and distinct associations. Based on the follow-up data of eight mental disorders, we employed propensity score matching (PSM) to create a sample size- and covariate-matched control group for each mental disorder group, and then used Cox proportional hazard regression to investigate the performance of each baseline exposure in predicting the subsequent onset of each mental disorder in the sample. We further performed the age- and sex-stratified analyses to explore the prediction performance in age and sex subgroups. By integrating multidimensional environmental profiling with symptom and disorder-level outcomes, we expect to improve the understanding of the associations between living environments and mental health, and to highlight the potential that modifying everyday environments may contribute to improved mental health outcomes, particularly in middle-aged and elderly people. The study workflow is illustrated in Fig. 1.

Fig. 1.

The study workflow. (a) Study variables. We include ten living environmental factors, three mental symptoms, and eight mental disorders in this study. (b) Data and analyses. All data are obtained from the UK Biobank. We utilize 75 items of living environmental exposures to construct ten living environmental factors through a ten-fold CFA. The three mental symptoms are evaluated by questionnaires, while the eight mental disorders are diagnosed based on ICD-10. Based on the baseline data, we investigate the associations between environmental factors and mental symptoms, and the relative contribution of each environmental exposure to each symptom. Based on the follow-up data, we use propensity score matching to create the matched case and control groups and employ Cox proportional hazards regression to assess the performance of each baseline exposure in predicting the subsequent onset of each disorder. For both analyses, we also perform age- and sex-stratified analyses and investigate the shared and distinct associations between subgroups. Abbreviations: CFA, confirmatory factor analysis; HR, hazard ratio; ICD, International Classification of Diseases

Methods

Participants

The UK Biobank is a population-based cohort that recruited 502,415 participants from 22 centers in the United Kingdom [11, 12]. This study was approved by the National Health Service (NHS) Research Ethics Service (21/NW/0157), and written informed consent was obtained from all participants. We accessed to the data under application number 75,556. We excluded 468 participants without any living environmental exposure data, 46 participants who withdrew from the UK Biobank before the beginning of this study, 71,402 participants with missing ethnic background or covariate data including age, sex, site, and genetic principal component, as well as those who were not white British. The remaining 430,499 participants were aged 38–73 years at baseline and included 197,496 males and 233,003 females. The numbers of participants in statistical analyses are specified in Supplementary Tables.

Living environmental assessments

The UK Biobank provided nine categories (75 items) of living environmental exposures assessed at the end of baseline data acquisition (before 2011) [13]. They included the index of multiple deprivation (IMD; 1 item), terrain (1 item), traffic (7 items), greenspace and coastal proximity (5 items), normalized difference vegetation index (NDVI; 1 item), street network accessibility (18 items), destination accessibility (33 items), air pollution (6 items), and residential noise pollution (3 items) (Supplementary Table 1).

Area deprivation

IMD (https://www.gov.uk/government/collections/english-indices-of-deprivation) was used to measure area deprivation. IMD was assessed in England, Scotland, and Wales, respectively. The IMD score was calculated from seven sub-scores (crime, employment, education, health, housing, income, and living environment) in England; seven (access to services, crime, education, employment, health, housing, and income) in Scotland; and eight (access to services, community safety, education, employment, housing, health, income, and physical environment) in Wales.

Terrain and traffic

The terrain slope was defined as the mean slope value within a 500 m circular buffer centered at each participant’s residence. Traffic was evaluated by seven items, including traffic intensity on the nearest road, inverse distance to the nearest road, traffic intensity on the nearest major road, inverse distance to the nearest major road, the total traffic load on major roads, close to the major road, and the sum of road length of major roads within 100 m.

Air and noise pollution

Residential air pollution was evaluated in 2010 by six items, including nitrogen dioxide (NO₂), nitrogen oxide (NO), particulate matter (PM₁₀), PM_2.5, PM_2.5 absorbance; and PM_2.5−10. Residential noise pollution was evaluated by the average sound levels of the daytime, evening, and night-time.

NDVI and greenspace and coastal proximity

NDVI was calculated from the reflectance measures in color infrared satellite data. The mean NDVI within 500 m around the residence was used to assess the greenness of the living environment for each participant. The greenspace and coastal proximity were evaluated by five items with 300 m home location buffers, including the percentages of greenspace, domestic garden, water, natural environment, and the Euclidean distance to the coast.

Destination and street network accessibility

The destination accessibility was assessed by eight subclasses (33 items), including the distance to community, education, factory, services, waste and energy, transport, food, and emergency. This domain was constructed from distance-based measures, with higher values indicating greater distances to destinations. Street network accessibility was evaluated by five subclasses (18 items), including the centrality, link characteristics, simple radial measures, network detour, and network shape.

Mental symptom assessments

Mental symptoms were assessed by three measures, including depression, anxiety, and bipolar symptoms (Supplementary Table 2).

Depression and anxiety symptoms

Consistent with a previous study [14], depression and anxiety symptoms were assessed at baseline interview using the Patient Health Questionnaire (PHQ)-4 [15]. Depression was evaluated with two questions: “Over the past two weeks, how often have you felt down, depressed, or hopeless?” (Data-Field 2050) and “Over the past two weeks, how often have you had little interest or pleasure in doing things?” (Data-Field 2060). Anxiety was assessed with other two questions: “Over the past two weeks, how often have you felt tense, fidgety, or restless?” (Data-Field 2070) and “Over the past two weeks, how often have you felt tired or had little energy?” (Data-Field 2080). Each question was assessed using a four-point Likert scale, ranging from 1 (not at all) to 4 (nearly every day). Both depression and anxiety scores ranged from 2 to 8, with higher scores indicating more severe symptoms.

Bipolar symptom

According to a previous study [16], the bipolar symptom (resource: 158772) was assessed by the following five aspects: ever manic or hyper for two days (Data-Field 4642), ever highly irritable or argumentative for two days (Data-Field 4653), length of the longest manic or irritable episode (Data-Field 5663), severity of the manic or irritable episode (Data-Field 5674), and manic or hyper symptoms (Data-Field 6156). Based on these assessments, the bipolar score of each participant was calculated using five criteria (yes = 1 and no = 0): ever manic or hyper for ≥ 2 days; ever irritable or argumentative for ≥ 2 days; manic episodes at least three of “more talkative”, “more active”, “needed less sleep”, and “more creative/more ideas”; longest manic episode ≥ 1-week; and episode needed treatment or caused problems at work. Specially, if a participant had ever been manic or hyper < 2 days or irritable or argumentative < 2 days, the participant was scored as zero. Therefore, the bipolar scores ranged from 0 to 5, with higher scores indicating more severe symptoms.

Quality control and factor analysis of living environmental exposures

The UK Biobank database included 75 items (nine categories; Supplementary Table 1) of living environmental exposures. For each item, we estimated its median and median absolute deviation (MAD) in all participants with this item assessment, and then we removed values that exceeded six MAD (both directions) than the median. We then excluded six items with extremely skewed distribution using the “nearZeroVar” function from caret R package [17]. When a category remained less than three items after quality control, we directly used these items. For each category with three or more items, we conducted a ten-fold cross-validation confirmatory factor analysis (CFA) to generate a composite index for the category. In each round, the CFA model was trained using 90% of participants, and the trained model was then used to calculate the latent variable (composite index) of the category for the remaining 10% of participants. The evaluation criteria for the CFA model included the comparative fit index (CFI) > 0.95, Tucker-Lewis index (TLI) > 0.95, standard root mean square residual (SRMR) < 0.08, and root mean square error of approximation (RMSEA) < 0.06 [18]. We initially trained the CFA model using all qualified items in the category. If the resulting CFA model could not meet the model fit criteria, we iteratively removed the item with the smallest factor loading and repeated modelling until the CFA model satisfied the criteria. The factor loadings of the optimized CFA model were used to calculate the factor in the test dataset.

Overall associations between living environments and mental symptoms

After normalizing mental symptom data using the RankGauss method, the multivariate linear regression was performed to investigate the overall associations between living environments and mental symptoms in all participants with both data. As environmental and mental health data were acquired at baseline, we investigated these associations, while controlling for genetically determined sex, top 40 genetic principal components, and age, assessment sites, and state of mental disorders at baseline. The covariates for mental disorders were the same eight mental disorders as the Cox regression analyses (Supplementary Table 3). According to the state of diagnosis for each mental disease at the end of baseline data acquisition (December 31, 2010), we encoded the diagnosed patients as 1 and other participants as 0. We corrected for multiple comparisons across ten environmental exposures and three mental health phenotypes. Given that analyses were conducted in the total sample as well as in four stratified subgroups (sex- and age-stratified analyses), the Bonferroni-corrected significance threshold was set at P < 0.05/10/3/5 = 3.33 × 10^− 4.

The relative importance of environmental exposures

To evaluate the relative importance of environmental exposures in the associations with each mental symptom, we utilized the R package relaimpo [19], which decomposes the total R² value of a regression model into contributions from individual predictors. This approach allows us to quantify the relative importance of each environmental factor in explaining the variance of mental symptoms. Specifically, we applied the “lmg” method from the package, which calculates the average contribution of each predictor across all possible orderings, providing a robust measure of variable importance. In our model, we included only the environmental factors that showed significant associations with each mental symptom, without adjusting for additional covariates to specifically assess the total R² contribution of the relevant environmental exposures. We then compared the importance scores to identify the most influential environmental factors for this mental symptom.

Sex-stratified analyses for associations with mental symptoms

We performed sex-stratified analyses to investigate the associations (P < 3.33 × 10^− 4, Bonferroni corrected) between living environments and mental symptoms in the male and female subgroups, respectively. We adjusted for all covariates in overall association analyses except for sex.

Age-stratified analyses for associations with mental symptoms

We divided the participants into the middle-aged (< 60 years) and elderly ( ≧ 60 years) subgroups, and performed age-stratified analyses to investigate the associations (P < 3.33 × 10^− 4, Bonferroni corrected) between living environments and mental symptoms in the two age subgroups, respectively. The same covariates as the overall association analyses were adjusted in these subgroup analyses.

Shared and distinct associations between subgroups

To assess the heterogeneity of associations across age and sex subgroups, we conducted Cochran’s Q-test (CQ-test) to compare the effect sizes between subgroups. A shared association was defined as the association that was significant in both subgroups (P < 3.33 × 10^− 4, Bonferroni corrected; consistent direction of effects) and had a CQ-test P > 0.05. A distinct association was defined as the association that was significant in at least one subgroup (P < 3.33 × 10^− 4, Bonferroni corrected) and had a CQ-test P_c < 0.05 (Bonferroni corrected for the unique significant associations in sex-stratified analyses and the unique significant associations in age-stratified analyses).

Overall Cox proportional hazard regression analyses

The Cox proportional hazard regression model was used to investigate whether baseline living environmental exposures can predict the subsequent onsets of mental disorders categorized according to the 10th version of the International Classification of Diseases (ICD-10). We only included mental disorders that: (a) had a high prevalence in middle-aged and elderly people comparable to UK Biobank participants; (b) were potentially associated with the living environmental exposures; and (c) had enough samples for the analyses (> 400 effective cases). We finally included eight mental disorders, including anxiety, depression, schizophrenia, bipolar affective disorder, dementia, delirium, mental and behavioral disorders (MBDs) due to use of tobacco, and MBDs due to use of alcohol (Supplementary Table 3).

As all included environmental data were acquired at the end of the baseline (before 2011), we used the last date (December 31) of 2010 as a cut-off to select participants without mental disorders before environmental data acquisition. The participants were then followed up and their diagnoses were continuously updated (recorded across all hospital inpatient records in either the primary or secondary position, Data-Field 41270 and 41280). With the date of our approved access to the UK Biobank data as the endpoint (December 8, 2021), we estimated and obtained the number of effective cases for each mental disorder. In the Cox regression analyses, we defined healthy controls as those participants who have no death records or lost visits, and have never been diagnosed as any of the included mental disorders by December 8, 2021. Incident cases were defined as events occurring after the date of environmental exposure assessment (December 31, 2010). The follow-up period for each participant extended from December 31, 2010, until the date of the first diagnosis of the disorder.

Considering that the sample size of healthy controls was much larger than that of patients with each mental disorder, we utilized propensity score matching (PSM) to generate a control group for each patient group, and the healthy control group was matched to the patient group in sample size and covariates of age at baseline, genetic-determined sex, and top 40 genetic principal components. The R package MatchIt [20] was used to perform PSM (method = “nearest”, distance = “gam”, caliper = 0.1). Standardized mean differences (SMDs) in these covariates were used to assess the performance of PSM.

Using each living environmental exposure as a predictor, the R package survival [21] was used to calculate the hazard ratio (HR) and its 95% confidential intervals (CIs) of the exposure for each mental disorder. The Cox proportional hazards model with robust variance estimators can account for the potential within-cluster correlation and provide reliable standard errors, whereby statistical inference remains valid even when the matched pairs are not independent. The Schoenfeld residuals method [22] was used to assess the proportional hazards assumption and the Bonferroni method (P < 0.05/10/8/5 = 1.25 × 10^{− 4}) was used to adjust for multiple comparisons.

For each exposure-disorder pair, we also evenly divided the participants (patients and controls) into four quarters according to the exposure. In each Cox proportional hazard regression analysis, we only included two quarters of participants with the highest and lowest exposures, by which the environmental variable was transformed into a binary variable (the highest and lowest exposures). Then we conducted the Cox regression analyses and calculated HR and 95% CI for the exposure-disease pair. We used the Kaplan-Meier curves [23] to represent the clinical progression of mental disorders over time (days) and compared the difference in the clinical progression of each mental disease between quarters of individuals with the highest and lowest exposures. Furthermore, we conducted a series of sensitivity analyses to assess the robustness of the findings under alternative sample definitions and additional covariate adjustment strategies. Detailed descriptions are provided in the Supplementary Methods and Supplementary Table 4.

Sex-stratified Cox proportional hazard regression analyses

We performed sex-stratified Cox proportional hazard regression analyses to investigate whether baseline living environmental exposures can predict the subsequent onsets of mental disorders (P < 1.25 × 10^− 4, Bonferroni-corrected) in male and female subgroups, respectively. Before the analyses, for the male or female subgroup, we performed PSM to construct the 1:1 matched control group for each mental disorder with the same covariates as the overall analyses except for sex.

Age-stratified Cox proportional hazard regression analyses

We performed age-stratified Cox proportional hazard regression analyses to investigate whether baseline living environmental exposures can predict the subsequent onsets of mental disorders (P < 1.25 × 10^− 4, Bonferroni-corrected) in middle-aged and elderly subgroups, respectively. Before the analyses, for the middle-aged or elderly subgroup, we performed PSM to construct the 1:1 matched control group for each mental disorder with the same covariates as the overall analyses.

Shared and distinct predictions between subgroups

To evaluate whether the associations differed across age and sex subgroups in the Cox regression models, we used a CQ-test to compare the coefficients between subgroups. A shared prediction was defined as the prediction that was significant in both subgroups (P < 1.25 × 10^− 4, Bonferroni corrected; consistent direction of effects) and had a CQ-test P > 0.05, while a distinct prediction was defined as the prediction that was significant in at least one subgroup (P < 1.25 × 10^− 4, Bonferroni corrected) and had a CQ-test P_c < 0.05 (Bonferroni corrected for the unique significant predictions in sex-stratified analyses and the unique significant predictions in age-stratified analyses).

Results

Construction of living environmental factors

The UK Biobank database included 75 items (nine categories) of living environmental exposures (Supplementary Table 1). For each item, we removed the participants with extreme values from all participants with this item assessment, and then estimated the variance of the item in the remaining participants. After removing six items with near zero variance, three categories were represented by one item, one category by two items, and five categories by more than two items. For each category with more than two items, we conducted a ten-fold CFA to create a composite index to represent this category (Supplementary Table 5). We finally included ten environmental exposures involving nine categories and 46 items (23 were further excluded during CFA model optimization) (Fig. 2, Supplementary Fig. 1 and Supplementary Table 1). The ten exposures were area deprivation, destination accessibility, air pollution, noise pollution, traffic intensity on the nearest major road (traffic intensity), inverse distance to the nearest major road (traffic distance), slope, normalized difference vegetation index (NDVI), greenspace and coastal proximity (green & blue spaces), and street network accessibility. We then calculated Spearman correlations between these environmental exposures, and found weak to moderate correlations (absolute rho ≤ 0.69; Supplementary Fig. 2), indicating that they are distinct but correlated exposures.

Fig. 2.

Construction of living environmental factors. We include nine categories (75 items) of living environmental exposures from up to 430,499 participants from the UK Biobank. After removing six items with near zero variance, a ten-fold CFA is performed for the five categories of environments containing three or more items. Finally, a total of ten living environmental factors are constructed involving 46 items, including three categories with one item (NDVI, slope, and area deprivation), one category with two items (traffic intensity and traffic distance), and five categories with a composite index created by ten-fold CFA to represent the category of exposures (air pollution, green & blue spaces, destination accessibility, street network accessibility, and noise pollution). Abbreviations: CFA, confirmatory factor analysis; Normalized Difference Vegetation Index

Overall associations between living environments and mental symptoms

We used the baseline depression, anxiety, and bipolar symptoms to assess mental health (Supplementary Table 2) and investigated the overall associations between ten living environments and three mental symptoms in the total sample. Given that the association analyses were conducted in the total sample and age- and sex-stratified subgroups, the Bonferroni-corrected significance threshold was set at P < 0.05/10/3/5 = 3.33 × 10^{− 4}. We identified 18 significant overall associations (Fig. 3a and Supplementary Table 6), involving eight environmental factors and all mental symptoms. Area deprivation, air pollution, traffic intensity, and traffic distance had adverse effects on mental symptoms, whereas NDVI, slope, destination accessibility, and green & blue spaces demonstrated beneficial effects on mental symptoms. Area deprivation showed the most significant associations with depression (β = 0.097, P < 1 × 10^{− 300}) and anxiety (β = 0.096, P < 1 × 10^{− 300}) symptoms. For each symptom, we further calculated the total R² of all related environmental factors that showed a significant correlation with the symptom. We then estimated the relative contribution of each related environmental factor to the symptom using the R package relaimpo [19]. The related environmental factors accounted for a total R² of 1.53% for anxiety, 1.57% for depression, and 0.78% for bipolar symptoms (Fig. 3b and Supplementary Table 7). Within the environmental contribution to model explanatory power, area deprivation accounted for the largest relative contribution, accounting for 85.6%, 86.2%, and 96.9% of the variances in anxiety, depression and bipolar symptoms, respectively. Air pollution is the second important contributor, accounting for 8.5%, 8.8%, and 2.1% of the variances in these symptoms.

Fig. 3.

Significant associations between living environments and mental symptoms. (a) Overall associations between living environments and mental symptoms (P < 3.33 × 10^− 4). The size of the outermost circle represents the absolute T value of each association. (b) The relative contributions of environmental factors to mental symptoms. (c) Significant shared (CQ-test: P > 0.05) and distinct (CQ-test: P < 1.25 × 10^− 3) associations between males and females. (d) Significant shared (CQ-test: P > 0.05) and distinct (CQ-test: P < 1.25 × 10^− 3) between elderly and middle-aged individuals. Abbreviations: NDVI, Normalized Difference Vegetation Index

Additionally, we assessed the associations between environmental exposures and three mental symptoms in non-White British participants, using the significant associations identified in White British participants as a reference and adjusting for the same covariates, to evaluate the generalizability of our findings across ethnic groups. Among the 18 significant associations, 11 remained significant in non-White British participants (P < 0.05/18 = 2.78 × 10^− 3). Consistent with the primary analysis, area deprivation exhibited the strongest associations, with significant effects on depression (β = 0.13, P = 1.17 × 10^− 145) and anxiety (β = 0.12, P = 8.48 × 10^− 125) symptoms. Importantly, the direction of all 18 significant associations was consistent between the two groups (Supplementary Table 8).

Sex-stratified associations with mental symptoms

We performed sex-stratified analyses to investigate the associations (P < 3.33 × 10^− 4, Bonferroni-corrected) between living environments and mental symptoms in male and female subgroups, respectively. We identified 17 significant associations in females and 20 significant associations in males (Supplementary Tables 9–10). For the 21 unique associations, we employed Cochran’s Q-test (CQ-test) to examine the differences in effect sizes between subgroups. A shared association was defined as the association that was significant in both subgroups (consistent direction of effects) and had CQ-test P > 0.05, and a distinct association was defined as the association that was significant in at least one subgroup and had CQ-test P < 0.05/(21 + 19) = 1.25 × 10^− 3 (Bonferroni corrected for 21 unique significant associations in sex-stratified analyses and 19 unique significant associations in age-stratified analyses). We found 13 sex-shared and one sex-distinct associations (Fig. 3c and Supplementary Table 11). The shared associations included the associations of area deprivation, traffic intensity, NDVI, and green & blue space with both anxiety and depression symptoms, associations of traffic distance, air pollution, and destination accessibility with anxiety symptom, as well as associations of area deprivation and air pollution with bipolar symptom. The sex-distinct association between street network accessibility and anxiety symptom exhibited a stronger effect (CQ-test: P = 4.62 × 10^− 4) in males (β = 0.012, P = 9.05 × 10^− 7) compared to females (β = 0.0004, P = 0.87).

Age-stratified associations with mental symptoms

We divided the participants into the middle-aged (< 60 years) and elderly ( ≧ 60 years) subgroups, and performed age-stratified analyses to investigate the associations (P < 3.33 × 10^− 4, Bonferroni-corrected) between living environments and mental symptoms in the two subgroups, respectively. We identified 17 associations in the middle-aged subgroup and 16 associations in the elderly subgroup (Supplementary Tables 12–13). The 19 unique associations included ten age-shared (CQ-test: P > 0.05) and three age-distinct (CQ-test: P < 1.25 × 10^− 3, Bonferroni-corrected) associations (Fig. 3d and Supplementary Table 14). The age-shared associations included those of NDVI, traffic distance, air pollution, and green & blue space with anxiety and depression symptoms, as well as the associations between destination accessibility and depression symptom, and between air pollution and bipolar symptom. As for age-distinct associations, area deprivation exhibited stronger associations with anxiety (middle-aged: β = 0.103, P < 1 × 10^− 300; elderly: β = 0.089, P = 3.78 × 10^− 281; CQ-test: P = 1.84 × 10^− 5), depression (middle-aged: β = 0.107, P < 1 × 10^− 300; elderly: β = 0.082, P = 3.31 × 10^− 240; CQ-test: P = 1.73 × 10^− 14), and bipolar (middle-aged: β = 0.075, P = 3.17 × 10^− 80; elderly: β = 0.048, P = 2.41 × 10^− 32; CQ-test: P = 2.50 × 10^− 6) symptoms in middle-aged individuals compared to elderly individuals.

Overall associations between living environments and mental disorders

We conducted the Cox proportional hazard regression analyses to investigate whether ten baseline environmental exposures can predict the subsequent onsets of eight mental disorders (Supplementary Table 3) based on the follow-up (from January 2011 to December 2021) data from the UK Biobank. These mental disorders included anxiety, depression, bipolar disorder, mental and behavioral disorders due to the use of alcohol (MBDs-alcohol), MBDs-tobacco, schizophrenia, dementia, and delirium. We defined healthy controls as the participants without death records or lost visits and having never been diagnosed as any of mental disorders by December 2021. For each mental disorder, the survival time of each patient was extended from December 31, 2010, until the date of the first diagnosis of the disorder. As the number of healthy controls was much larger than the number of patients with each mental disorder, we used PSM to generate a healthy control group for each patient group matched in sample size and 42 covariates (Supplementary Table 15 and Supplementary Figs. 3–7). In the Cox regression analyses, we adjusted for ten exposures and eight mental disorders across the overall analysis, and age-stratified and sex-stratified subgroup analyses, resulting in a Bonferroni-corrected significance threshold of P < 0.05/10/8/5 = 1.25 × 10^− 4.

We subdivided the participants (patients and controls) evenly into four subgroups (quarters) according to each environmental exposure, and then investigated disease risk differences between quarters of participants with the highest and lowest exposures. We conducted Cox hazard regression analyses and calculated hazard ratios (HRs) and 95% confidence intervals (CIs) for ten baseline exposures in predicting the subsequent onsets of eight mental disorders. We used the Kaplan-Meier curves [23] to represent the clinical progression of mental disorders over time (days) and compared the difference in clinical progression of each disorder between the quarters of individuals with the highest and lowest exposures. We found 25 significant associations between living environments and mental disorders (Fig. 4a and Supplementary Table 16). Area deprivation, air pollution, noise pollution, traffic distance, and traffic intensity were the risk factors, while slope, destination accessibility, and green & blue spaces were protective factors for mental disorders. We identified four exposures with above twofold risk for disorders. Compared to the quarter of individuals with the lowest exposure, the quarter with the highest exposure to area deprivation demonstrated an increased risk for schizophrenia (HR = 3.53, P = 4.18 × 10^{− 14}), MBDs-tobacco (HR = 2.81, P < 1.00 × 10^{− 300}) and bipolar disorder (HR = 2.01, P = 1.06 × 10^{− 11}) (Fig. 4b), and the quarter of individuals with highest exposure to air pollution also showed increased risk for schizophrenia (HR = 3.12, P = 2.73 × 10^{− 10}) (Supplementary Fig. 8). We also identified one prediction with below 50% risk for disorder. The quarter of individuals with the highest exposure to destination accessibility (HR = 0.43, P = 7.94 × 10^{− 8}) showed a reduced risk for schizophrenia than the quarter with the lowest exposure (Supplementary Fig. 8). The results of the sensitivity analyses are provided in the Supplementary Results and Supplementary Tables 17–24, further supporting the robustness and stability of the findings.

Fig. 4.

Baseline living environments predict subsequent onsets of mental disorders. (a) The significant predictions of baseline living environments for the subsequent onset of mental disorders. HR values are displayed, with different colors representing distinct disorders. (b) Kaplan-Meier curves illustrate the impact of area deprivation on the risk of bipolar disorder, schizophrenia and MBDs-tobacco. The curves represent the quarters of individuals with the lowest (orange) and highest (green) exposures, and shaded regions indicate 95% CIs. Abbreviations: CI, confidence interval; HR, hazard ratio; MBDs, mental and behavioral disorders

Sex-stratified associations with mental disorders

We performed sex-stratified Cox proportional hazard regression analyses to investigate whether ten baseline living environmental exposures can predict the subsequent onsets of eight mental disorders (P < 1.25 × 10^− 4, Bonferroni-corrected) in male and female subgroups, respectively. We identified 31 significant predictions in males and 15 in females (Supplementary Tables 25–26). For the 33 unique predictions from the two subgroups, the CQ-test was used to quantify the sex differences in effect sizes of these predictions. A shared prediction was defined as the prediction that was significant in both subgroups (consistent direction of effects) and had CQ-test P > 0.05, while a distinct prediction was defined as the prediction that was significant in at least one subgroup and had CQ-test P < 0.05/(33 + 34) = 7.46 × 10^− 4 (Bonferroni-corrected for 33 unique predictions in sex-stratified analyses and 34 unique predictions in age-stratified analyses). By definition, we identified nine sex-shared predictions and two sex-distinct predictions (Fig. 5a and Supplementary Table 27). The sex-shared predictions were the predictions of area deprivation to anxiety, dementia, depression, and delirium, the predictions of slope and traffic intensity to MBD-tobacco, and the predictions of slope to depression, destination accessibility to MBD-alcohol, and air pollution to delirium. The sex-distinct predictions were the predictions of destination accessibility and air pollution to depression. Destination accessibility showed stronger protective effect (males: HR = 0.754, P = 4.28 × 10^− 14; females: HR = 0.906, P = 3.31 × 10^− 4; CQ-test: P = 8.02 × 10^− 5) (Fig. 5b) and air pollution showed increased risk (males: HR = 1.414, P = 1.42 × 10^− 22; females: HR = 1.211, P = 3.58 × 10^− 13; CQ-test: P = 4.12 × 10^− 4) on depression in males compared to females.

Fig. 5.

Baseline living environments predict subsequent onsets of mental disorders in sex- and age-stratified analyses. (a) Significant shared (CQ-test: P > 0.05) and distinct (CQ-test: P < 7.46 × 10^− 4) associations between males and females. (b) Kaplan-Meier curves illustrate the impact of destination accessibility on depression risk in males and females. The curves represent the quarters of individuals with the lowest (orange) and highest (green) exposures, and shaded regions indicate 95% CIs. (c) Significant shared (CQ-test: P > 0.05) and distinct (CQ-test: P < 7.46 × 10^− 4) associations between elderly and middle-aged individuals. (d) Kaplan-Meier curves depict the risk prediction of area deprivation for depression in middle-aged and elderly individuals, with visualized quarters for the lowest (orange) and highest (green) exposures. Shaded regions denote 95% CIs. Abbreviations: CI, confidence interval; HR, hazard ratio

Age-stratified associations with mental disorders

We performed age-stratified Cox proportional hazard regression analyses to investigate whether ten baseline living environmental exposures can predict the subsequent onsets of eight mental disorders (P < 1.25 × 10^− 4, Bonferroni-corrected) in middle-aged and elderly subgroups, respectively. We identified 25 significant predictions in the elderly and 18 in middle-aged individuals, as well as seven age-shared (CQ-test: P > 0.05) and four age-distinct (CQ-test: P < 7.46 × 10^− 4, Bonferroni-corrected) predictions (Fig. 5c and Supplementary Tables 28–30). Age-shared predictions included the predictions of slope and air pollution to depression, area deprivation and air pollution to delirium, destination accessibility and air pollution to MBD-alcohol, and slope to MBD-tobacco. In the age-distinct predictions, area deprivation had higher risk for depression (middle-aged: HR = 1.937, P = 1.35 × 10^− 117; elderly: HR = 1.548, P = 6.88 × 10^− 48; CQ-test: P = 7.94 × 10^− 8) (Fig. 5d), anxiety (middle-aged: HR = 1.867, P = 4.46 × 10^− 85; elderly: HR = 1.495, P = 7.80 × 10^− 37; CQ-test: P = 9.32 × 10^− 7), and MBD-tobacco (middle-aged: HR = 3.063, P = 4.59 × 10^− 264; elderly: HR = 2.497, P = 1.19 × 10^− 158; CQ-test: P = 1.25 × 10^− 5) in middle-aged individuals than in elderly individuals. Additionally, NDVI only reduced the risk for MBD-tobacco (CQ-test: P = 2.92 × 10^− 4) in middle-aged individuals (HR = 0.857, P = 4.09 × 10^− 7) rather than in elderly individuals (HR = 1.013, P = 0.706).

Discussion

In up to 430,499 participants from the UK Biobank, we investigated the associations between ten living environments and three mental symptoms, identifying 18 significant associations. We also examined the predictions of the ten baseline living environmental exposures to the subsequent onset of eight mental disorders, uncovering 25 significant predictions. The sex- and age-stratified analyses further revealed the shared and distinct associations and predictions between subgroups. The discovered associations between living environments and mental symptoms and disorders could enhance mental health by informing more targeted and effective public health strategies.

In the cross-sectional analysis, we found 18 significant associations between living environments and mental symptoms. Although prior studies have reported associations between area deprivation, air pollution, traffic intensity and distance, NDVI, and green and blue space with mental symptoms [24–27], we also discovered several new associations. For example, higher destination accessibility was identified as an exposure for all three mental symptoms, indicating that the availability of essential services, social interactions, and recreational opportunities may be linked to mental health issues. We also identified slope as a protective factor for anxiety symptom, possibly due to the increased outdoor walking time in individuals living in hilly area [28]. These findings may inform novel interventions for improving mental health. Slope and destination accessibility may influence health through several interrelated pathways. Neighborhoods with greater walkability may encourage walking and other forms of active transport, thereby increasing daily physical activity levels and contributing to better physical and mental health outcomes. Previous studies have consistently shown that walkable environments are associated with higher levels of walking and physical activity as well as reduced risks of chronic diseases [29–31]. In addition, accessible and walkable neighborhoods may promote social interaction, neighborhood trust, and social cohesion, which are also important determinants of health and well-being. Moreover, the inclusion of multiple exposures in the same study offers a unique opportunity to explore the relative contribution of individual exposures to mental symptom. Among the significant exposures, we found that area deprivation accounted for 86.2%, 85.6%, and 96.9% of the variances in depression, anxiety, and bipolar symptoms. Although the association between area deprivation and mental health has been explained by the social drift/selection hypothesis (individuals with mental health problems are more likely to move to or reside in impoverished areas) and/or the unobserved familial factors [32–34], the social causality hypothesis (adverse social and economic factors may increase the risk of mental health problems) is also suggested to operate independently or concurrently with the social selection hypothesis [35–40]. If the causal influence of area deprivation on mental health is established, tackling social inequality would become a priority for improving mental health. While area deprivation and air pollution accounted for the largest relative contributions in our models, other urban environmental factors may still have important practical implications. In particular, factors such as green space, transportation, and destination accessibility represent modifiable features of the living environment that can be targeted through urban planning and public health interventions. Moreover, these environmental characteristics may act synergistically with socioeconomic conditions to influence mental health outcomes. In addition to sex-shared and age-shared associations, we also found sex-distinct and age-distinct associations, indicating more precise interventions. Street network accessibility is a crucial aspect of urbanicity, which has been associated with negative mental health outcomes [41]. We extended the observation by providing sex-specific information (higher street network accessibility was a risk factor for anxiety symptom in males). We also found stronger associations between area deprivation and mental symptoms in middle-aged compared to elderly individuals, informing the vulnerable population for interventions.

To validate and expand the cross-sectional association analyses between living environments and mental symptoms, we further investigated whether baseline living environmental exposures could predict the subsequent onset of mental disorders. The potential confounding effects were eliminated by creating a control group with matched sample size and covariates. We identified 25 significant predictions, including that area deprivation predicted the subsequent onset of mental disorders. As longitudinal analysis may reduce the likelihood of inverse causality, our findings may provide evidence for the social causality hypothesis. In addition to confirming several previous findings, such as the links between area deprivation and air pollution with mental disorders [42–44], we also provided several new findings. For instance, we found that traffic intensity and distance could predict the risk for MBDs rather than other mental disorders. Among the three protective factors (slope, green and blue space, and destination accessibility), the strongest effect (HR = 0.434) was observed for destination accessibility concerning schizophrenia. These new findings may inform the disorder-specific interventions. In sex-stratified analyses, we found that higher destination accessibility showed a stronger protective effect and higher air pollution had a stronger adverse effect on depression in males compared to females. In age-stratified analyses, however, we identified that area deprivation had a higher risk for anxiety, depression, and MBD-tobacco, while NDVI exhibited a lower risk for MBD-tobacco in middle-aged compared to elderly individuals. These findings may provide key information for population-specific interventions for mental health improvement.

Several limitations should be acknowledged when interpreting the findings. First, while we included a broad range of living environmental factors, such as socioeconomic conditions, built environments, natural surroundings, and pollution, these cannot cover all living environmental factors that may be linked to mental health. Second, we focused primarily on the linear relationships between living environments and mental health outcomes. This strategy offers an interpretable framework for assessing the associations; however, it cannot capture the complex non-linear associations between environments and mental health. Third, the environmental exposures were only assessed at baseline, which cannot be employed to identify the temporal changes and sensitive age windows of the environmental effects on mental health. Fourth, the PHQ-4 is a brief screening instrument rather than a fully validated dimensional symptom scale, and its use as a continuous measure may limit variance and attenuate associations. Future studies employing more comprehensive symptom assessments are warranted to confirm our findings. Similarly, the bipolar symptom measure used is not a standard validated scale. Although it captures self-reported bipolar-like symptoms and demonstrates acceptable construct validity in our sample, these findings should be interpreted with caution, and future research using validated dimensional measures is recommended. Fifth, while we performed a sensitivity analysis using UK Biobank Home Location History data to account for residential moves, some exposure misclassification may still be present, potentially affecting the estimates. Sixth, despite using propensity score matching and sensitivity analyses, residual confounding due to unmeasured health-related factors cannot be entirely excluded. Therefore, while our results appear robust, caution is warranted when interpreting causal relationships. Seventh, our analyses were restricted to White British participants to minimize potential confounding from population stratification, which could influence both environmental exposures and mental health outcomes. Although similar associations were observed in non-White British participants, the findings may not be fully generalizable to other ethnic groups. Finally, given the relatively small schizophrenia subsample, these findings should be interpreted with caution and considered exploratory.

Conclusion

Using ten variables to reflect distinct aspects of living environments, we identified 18 significant associations between living environmental exposures and three mental symptoms, and 25 significant predictions for eight mental disorders. We also found the shared and distinct associations between sex and age subgroups. These findings provide evidence for developing tailored interventions to optimize mental health and prevent mental disorders across diverse populations.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Chunshui Yu and Yingying Xie designed the study. Yingying Xie and Jiawei Liu analyzed the data. Yingying Xie and Chunshui Yu wrote the manuscript. Chunshui Yu, Wen Qin, Feng Liu, Meng Liang and Jiayuan Xu supervised this work. All authors critically reviewed the manuscript.

Funding

We acknowledge funding from the National Natural Science Foundation of China (82430063, 82030053, 81425013, 82371924, 82502309); Tianjin Key Medical Discipline Construction Project (TJYXZDXK-3–008 C); National Key Project of “Inter-governmental International Scientific and Technological Innovation Cooperation” (2023YFE0199700); the Tianjin Applied Basic Research Diversified Investment Foundation (21JCYBJC01360); Science & Technology Development Fund of Tianjin Education Commission for Higher Education (2019KJ195); the Tianjin Medical University “Clinical Talent Training 123 Climbing Plan”; and the Tianjin Young Talents in Science & Technology.

Data availability

The data of this study can be requested from the UK Biobank.

Declarations

Ethics approval and consent to participate

This study used data from the UK Biobank, which has approval from the National Health Service (NHS) Research Ethics Service (Ref: 21/NW/0157). All participants provided written informed consent. This research was conducted under UK Biobank project 75556 and adhered to the principles of the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.