Residential greenness associate with psychiatric problem and physical–psychological multimorbidity among 6900 Chinese middle-aged and older adults

Abstract

Psychological problem and physical–psychological multimorbidity is rising among middle-aged and older adults in China. Whether residential greenness, air pollution and systemic inflammation jointly influence psychiatric outcomes remains unclear. Using a retrospective cross-sectional design based on 2015 China Health and Retirement Longitudinal Study (CHARLS) data, we investigated the link between the Normalized Difference Vegetation Index (NDVI) and psychiatric problems/comorbidities, and evaluated the roles of ambient pollutants (PM₂.₅, PM₁₀) and the inflammatory biomarker C-reactive protein (CRP). The final sample included 6900 adults aged ≥ 45 years. Key findings revealed that a 0.1-unit increase in NDVI was associated with an 18% lower risk of psychiatric problems (OR = 0.82, 95% CI 0.68–0.97, P = 0.024) and a 19% lower risk of physical–psychological multimorbidity (OR = 0.81, 95% CI 0.67–0.99, P = 0.035). Conversely, each unit increase in PM₂.₅ and PM₁₀ was linked to a 1% higher risk of psychiatric problems (PM₂.₅: OR = 1.01, 95% CI: 1.00–1.02, P = 0.033; PM₁₀: OR = 1.01, 95% CI 1.00–1.02, P = 0.042) and multimorbidity (PM₂.₅: OR = 1.01, 95% CI 1.00–1.03, P = 0.048; PM₁₀: OR = 1.01, 95% CI 1.00–1.02, P = 0.052). Inflammatory marker analysis showed that a 10-unit increase in CRP was associated with a 20% higher risk of psychiatric problems (OR = 1.20, 95% CI 1.04–1.39, P = 0.013) and a 22% higher risk of comorbidities (OR = 1.22, 95% CI 1.06–1.42, P = 0.007). Pathway analysis confirmed no significant evidence of mediation effects of PM₂.₅, PM₁₀, or CRP on the association between NDVI and physical–psychological multimorbidity. Cross-sectional analyses indicate an inverse association between residential greenness and psychiatric morbidity, yet longitudinal or interventional evidence is required to establish the clinical significance of this relationship and to determine whether air pollution or inflammation represents a causal mediator before findings can guide policy or practice.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-24005-8.

Introduction

According to the World Health Organization (WHO), the prevalence of psychiatric problems among older adults has been rising steadily over recent years. This situation underscores the urgency and significance of addressing this growing public health concern¹. Further analysis from the Global Burden of Disease Study 2021 projects the prevalence of major depressive disorders (MDD) among older adults through 2050, extending previous findings. The study indicates that the burden of MDD is increasing and is expected to keep rising, with significant growth in lower—income countries². In current medical practice, the primary diagnostic and treatment approaches for psychiatric problems are pharmacological therapy and psychotherapy. However, these interventions are limited by issues like low efficacy and medication dependency. Physical-psychological multimorbidity is used broadly to denote the co-occurrence of two or more chronic disorders in the same individual, provided that at least one is a mental health condition(co-occurrence of physical and psychological conditions)³. Therefore, it is crucial to explore positive associations for psychological health of middle-aged and older adults⁴.

Recent evidence converges on the neuropsychiatric toxicity of air pollution: prenatal PM₂.₅ exposure elevates offspring risks of autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD); chronic ozone inhalation activates the hypothalamic–pituitary–adrenal axis and sympathetic nervous system, provoking neuro-inflammation and oxidative stress that impair cognition; and a nationwide cohort linked long-term solid-fuel use to accelerated biological ageing and higher incidence of neuropsychiatric disease through systemic inflammation, oxidative damage and blood–brain-barrier disruption^5–8.

Existing studies also have identified residential greenness and air pollution as key determinants of psychological health. A significant positive association has been observed between environmental greenness and psychiatric problems, whereas air pollution has adverse effects on psychological outcomes, especially in research from low- and middle-income countries⁹.

Specifically, green environments foster social engagement and psychiatric support, whereas air pollution is associated with various psychiatric health issues. Thus, improving residential greenness may have a positive impact on the psychiatric health of middle-aged and older adults, which forms the basis of this study¹⁰. Similarly, individuals with better mental health and higher socioeconomic position (SEP) may be more likely to self-select into aesthetically advantageous neighbourhoods; this proposition is supported by multiple strands of evidence. First, empirical studies have consistently linked residence in high-SEP communities to more favourable mental health outcomes. Higher SEP is typically accompanied by qualitatively superior neighbourhood environments—characterised by greater green-space coverage and better communal amenities—that independently contribute to enhanced psychological well-being¹¹.

Satellite-derived NDVI, the canonical indicator of residential greenness, has been consistently linked to lower cardiovascular risk across diverse populations: a 0.1-unit rise in annual mean NDVI corresponded to a 9.3% reduction in overall CVD mortality (HR = 0.907, 95% CI 0.859–0.957) in the general population, while among diabetic patients each 0.01-unit NDVI increment was associated with a 6% decrease in myocardial-infarction risk (HR = 0.94, 95% CI 0.89–0.99), collectively underscoring the positive association of greater green exposure^12–14.

However, direct evidence linking NDVI to physical–psychological multimorbidity is scarce. Although existing studies have confirmed the effect of natural environments against single psychiatric problems, they have three main limitations: ① lack of mechanistic exploration of physical–psychological multimorbidity; ② insufficient evidence from Chinese aging cohorts; ③ failure to control for confounding factors in the urbanization process. Grounded in the socio-ecological model and exposome theory, we propose a theoretical framework in which the macro-level built environment (higher NDVI) leads to a reduced incidence of physical-psychological multimorbidity.Thus, it is urgent to validate the specific impact of NDVI on physical–psychological multimorbidity in Chinese older adults.

This study, using the nationally representative CHARLS 2015 sample, employed a retrospective cross-sectional design and achieved three objectives for the first time: ① accurately calculating individual NDVI exposure with 250 m resolution satellite data; ② analyzing psychiatric problems and profiles of physical-psychological multimorbidity; ③ applying multivariable adjusted models to control for confounders like air pollution and socioeconomic factors. The findings will offer evidence-based support for “green prescriptions” as interventions for physical-psychological multimorbidity and facilitate connections between urban planning and public health policies.

Methods

Study population​​

This study utilised the nationally representative cross-sectional sample from the 2015 wave of the China Health and Retirement Longitudinal Study (CHARLS). The source population comprised community-dwelling residents aged ≥ 45 years in 150 counties across 28 provinces of China. Participants were eligible if they: (1) completed a physical examination (height, weight, blood pressure); (2) had complete fasting venous blood biochemistry, including fasting plasma glucose, total cholesterol, triglycerides, high-density lipoprotein cholesterol, and low-density lipoprotein cholesterol; and (3) provided a valid residential address enabling geolocation within a 250 m buffer for NDVI and PM₂.₅ exposure linkage. Exclusion criteria were: (1) unsuccessful air-pollution exposure assessment (missing address or satellite data); (2) non-response to the question “Has a doctor ever told you that you had emotional, nervous, or psychiatric problems?”; or (3) self-reported physician-diagnosed schizophrenia or bipolar disorder (to exclude severe psychotic disorders). After these procedures, 6,900 participants remained for analysis (Fig. 1)¹⁵.

Fig. 1.

Flowchart of participant selection in the current study.

Variables

Psychiatric outcomes definition

This study defines psychiatric outcomes as psychiatric problems and physical-psychological multimorbidity.Psychiatric problems were ascertained by self-report of a physician diagnosis obtained through the question “Has a doctor ever told you that you had emotional, nervous, or psychiatric problems?”.

Physical-psychological multimorbidity was defined as coexisting ≥ 2 chronic diseases from 14 physician-diagnosed categories self-reported by participants: hypertension, diabetes mellitus, cancer, lung diseases, heart diseases, stroke, mental problems, arthritis, dyslipidemia, liver diseases, kidney diseases, gastrointestinal diseases, asthma, and memory-related diseases, based on self-reported diagnostic information from questionnaires. Physical-psychological multimorbidity was defined as the simultaneous presence of ≥ 2 chronic conditions, at least one of which was a physician-diagnosed psychiatric problems (emotional, nervous, or psychiatric problems).Outcomes were evaluated using standardized questionnaires from CHARLS, which included incident cases identified after excluding certain baseline data¹⁶.

NDVI and PM₁, PM₂.₅, PM₁₀, NO₂ exposure

Residential greenness was quantified as the annual mean NDVI (MODIS MOD13Q1) within a 250 m radius of each participant’s address, covering the year preceding visual impairment, loss to follow-up, or study end¹⁷.

Exposure to PM₁, PM₂.₅, PM₁₀, and NO₂ was estimated using a validated spatiotemporal model combining multi-source data. Ground-level monitoring data (2015) from 77 CAWNET (PM₁) and 1,497 CNEMC stations (PM₂.₅/PM₁₀/NO₂) were supplemented with satellite-derived daily AOD (MODIS, 1 km) and tropospheric NO₂ (OMI Level 3). These were temporally aligned with NDVI (2015 annual averages) to match CHARLS outcomes.A random forest model (R ranger package) integrated predictors like satellite data, meteorological variables (temperature/humidity), land use, road/population density to produce 1 km × 1 km concentration grids. Model performance was strong (PM₂.₅: R² = 0.92, RMSE = 8.2 μg/m³; tenfold cross-validation). Quality control included:Excluding hourly station data with > 10% daily missingness or extreme values (± 3 SDs from annual means).Ground-truth calibration via linear mixed-effects models (random station intercepts).Bootstrap resampling to assess exposure misclassification uncertainty.Methodological details follow Chen et al^18–20.

Covariate

CHARLS utilized a multistage stratified sampling method that was proportional to size, and data were gathered through standardized questionnaires, biological sample analyses, and satellite remote sensing techniques²¹.

The covariates taken into account were diverse: meteorological factors such as temperature and humidity sourced from the China Meteorological Administration; sociodemographic variables including age, sex, urban or rural residence, education level, marital status, and household expenditure; behavioral factors like smoking, alcohol consumption, and physical activity measured in metabolic equivalents (METs); metabolic indices comprising waist circumference, blood pressure, and glucose levels; and medication history focusing on the use of antihypertensive and hypoglycemic drugs. Metabolic syndrome was defined according to modified Joint Interim Society (JIS) criteria, which required a waist circumference of at least 90 cm for men and 80 cm for women, along with the presence of two or more metabolic abnormalities. The use of solid fuels encompassed coal, crop residues, or wood, while clean energy sources included natural gas, biogas, and liquefied petroleum gas (LPG). Definitions for physical activity, cardiometabolic multimorbidity (CMM), and social isolation were consistent with existing literature^16,22–24.

Covariates were selected a priori from established theory, prior evidence, practical relevance, and significant yet non-collinear associations with psychiatric outcome.

Ethic statement

This study followed the rules in the Declaration of Helsinki. The Ethics Committee at Peking University approved it. The approval number is IRB00001052-11015. All people in CHARLS signed a document to agree. The data was made anonymous before sharing. Because we used data that was already de-identified, we did not ask for new consent. The Ethics Committee at 925 Hospital checked and approved this process²⁵.

Statistical methods

All statistical analyses were conducted using R version 4.3.1 and free statistics version 2.2. Baseline characteristics among study participants were compared using independent t-tests for normally distributed variables, Mann–Whitney U tests for variables exhibiting skewness, and χ² tests for categorical variables. The association between environmental factors and the Normalized Difference Vegetation Index (NDVI) was evaluated using Spearman correlation coefficients.

Statistical analysis employed multivariable logistic regression with three hierarchical models: unadjusted (Model 1), demographics-adjusted (Model 2: age, sex, education, marital status), and fully adjusted (Model 3: Model 2 plus lifestyle, socioeconomic, and clinical variables). Urban–rural gradients were captured by distance to city center and population density; household expenditure (log-transformed) served as an income proxy. Missing data were addressed by multiple imputation using chained equations (mice), producing five imputed datasets with 50 iterations each. The imputation model incorporated all analytic covariates”.

A significance level was set at two-tailed P < 0.05, with a P-value range of 0.05 ≤ P < 0.10 indicating marginal trends. Effect sizes were reported as odds ratios (ORs) with 95% confidence intervals (CIs) corresponding to unit increases in pollutant concentrations (PM₂.₅ and PM₁₀ at 1 μg/m³; CRP at 10 mg/L), and NDVI (0.1 units).We used the 10-question Center for Epidemiologic Studies Depression Scale (CES-D-10) in CHARLS to identify depression. A total score of 10 or more usually indicates a risk of depression. We then looked at the link between depressive symptoms and NDVI, and how the co-occurrence of physical and depression symptoms might change this relationship for sensitivity analysis²⁶.

Pathway analysis used NDVI as the independent variable and psychiatric problems/comorbidities as the dependent variables, incorporating three mediation pathways. We evaluated potential mediation by PM2.5, PM10, and CRP in the NDVI-psychiatric outcomes relationship using Preacher-Hayes bootstrapping (10,000 resamples) with full covariate adjustment. Paths included:NDVI → Mediators (pollutants/CRP);Mediators → Outcomes (psychiatric problems/multimorbidity);NDVI → Outcomes (direct effect), KHB decompositionaddressed scaling in logistic models, and bias-corrected 95% CIs determined significance²⁷. Given that cross-sectional data cannot verify the temporal sequence, the mediating effects reported in this paper are merely statistical associations, rather than evidence of a causal chain.

Results

Study population characteristics

The current investigation encompassed a total of 6900 participants, with 6751 individuals classified within the normal group and 149 within the psychiatric problem group, revealing notable heterogeneity (refer to Table 1). Regarding demographic characteristics, the psychiatric problem group exhibited a statistically significant greater mean age (61.9 ± 8.9 years compared to 59.7 ± 9.3 years, P = 0.005) as well as a higher male representation (66.4% vs. 49.8%, P < 0.001). While there were no significant variations in residential location, alongside an increased reliance on solid fuel sources (51.7% vs. 41.0%, P = 0.009). The prevalence of current smoking was markedly higher in the psychiatric problem group (62.4% vs. 45.1%, P < 0.001); however, no significant disparities were noted in terms of alcohol consumption frequency or educational attainment. Importantly, multimorbidity rates were considerably elevated in the psychiatric problem group (88.6% vs. 49.7%, P < 0.001), as were median levels of C-reactive protein (1.7 mg/L vs. 1.4 mg/L, P = 0.015), suggesting distinct differences in systemic inflammatory states.

Table 1.

Baseline characteristics of study participants by psychiatric problem status.

Variables	Total (n = 6900)	Normal	Psych problem	P-value
Age (years)	59.7 ± 9.3	59.7 ± 9.3	61.9 ± 8.9	0.005
BMI	25.0 ± 33.0	25.0 ± 33.3	23.6 ± 4.1	0.609
Sex, n (%)				< 0.001
Female	3437 (49.8)	3387 (50.2)	50 (33.6)
Male	3463 (50.2)	3364 (49.8)	99 (66.4)
Residence, n (%)				0.527
Urban	4321 (62.6)	4224 (62.6)	97 (65.1)
Rural	2579 (37.4)	2527 (37.4)	52 (34.9)
Marital_status, n (%)				0.077
Married and living with a spouse	6526 (94.6)	6389 (94.6)	137 (91.9)
Married but living without a spouse	366 (5.3)	355 (5.3)	11 (7.4)
Single, divorced, and windowed	8 (0.1)	7 (0.1)	1 (0.7)
Education_Status, n (%)				0.14
Elementary school or below	4516 (65.4)	4410 (65.3)	106 (71.1)
Middle school or above	2384 (34.6)	2341 (34.7)	43 (28.9)
Smoking_Status, n (%)				< 0.001
Non-smoker	3759 (54.5)	3703 (54.9)	56 (37.6)
Smoker	3141 (45.5)	3048 (45.1)	93 (62.4)
Drinking_Status, n (%)				0.494
Non-drinker	4444 (64.4)	4354 (64.5)	90 (60.4)
Drink but less than once a month	574 ( 8.3)	562 (8.3)	12 (8.1)
Drink more than once a month	1882 (27.3)	1835 (27.2)	47 (31.5)
Regional_Category, n (%)				0.001
East	2416 (35.0)	2383 (35.3)	33 (22.1)
Midland	2453 (35.6)	2382 (35.3)	71 (47.7)
West	2031 (29.4)	1986 (29.4)	45 (30.2)
Cooking_Fuel_Use, n (%)				0.009
Clean fuel	4056 (58.8)	3984 (59)	72 (48.3)
Solid fuel	2844 (41.2)	2767 (41)	77 (51.7)
Waist circumference, cm	85.8 ± 12.8	85.8 ± 12.8	85.4 ± 14.2	0.721
Anti hypertensive medicine use, n (%)				0.117
No	5194 (75.3)	5090 (75.4)	104 (69.8)
Yes	1706 (24.7)	1661 (24.6)	45 (30.2)
Anti Diabetes medicine use, n (%)				0.188
No	6436 (93.3)	6301 (93.3)	135 (90.6)
Yes	464 ( 6.7)	450 (6.7)	14 (9.4)
MetS, n (%)				0.081
No	5488 (79.5)	5361 (79.4)	127 (85.2)
Yes	1412 (20.5)	1390 (20.6)	22 (14.8)
NVDI	0.29 ± 0.1	0.29 ± 0.1	0.27 ± 0.09	0.051
NO2 (μg/m3)	29.7 ± 9.5	29.7 ± 9.5	30.3 ± 9.4	0.472
O3 (μg/m3)	83.1 ± 7.6	83.1 ± 7.6	82.4 ± 7.3	0.265
PM1 (μg/m3)	31.3 ± 10.3	31.2 ± 10.3	32.1 ± 10.6	0.297
PM10 (μg/m3)	95.1 ± 33.5	95.1 ± 33.5	99.2 ± 33.0	0.132
PM2.5 (μg/m3)	56.0 ± 19.0	56.0 ± 19.0	58.5 ± 18.9	0.114
SO2 (μg/m3)	31.1 ± 14.6	31.1 ± 14.6	31.5 ± 14.4	0.749
Comorbidity, n (%)				< 0.001
No	3412 (49.4)	3395 (50.3)	17 (11.4)
Yes	3488 (50.6)	3356 (49.7)	132 (88.6)
C reactive protein (mg/l)	1.4 (0.8, 2.6)	1.4 (0.7, 2.6)	1.7 (1.0, 3.2)	0.015
Physical activity	15,078.0 (8932.5, 20,160.0)	15,078.0 (8932.5, 20,178.0)	15,252.5 (8355.0, 18,038.2)	0.476
Annual household expenditure	8851.0 (4930.0, 15,556.0)	8864.0 (4948.8, 15,548.0)	7730.0 (3720.0, 15,567.0)	0.048

Although PM2.5 and PM10 concentrations demonstrated a rising trend within the psychiatric problem group, this was not statistically significant (P > 0.05). Furthermore, no significant differences in physical activity levels or annual family expenditures were identified between the groups (P > 0.05), although the psychiatric problem group reported a lower median annual expenditure (7730 yuan vs. 8864 yuan, P = 0.048).

Associations of ambient air pollutants with NDVI and psychiatric outcomes

Correlation analyses revealed inverse associations between NDVI and ambient air pollutants: NO₂ (r = – 0.29), O₃ (r = – 0.37), PM₁ (r = – 0.32), PM₁₀ (r = – 0.33) and PM₂.₅ (r = – 0.36) (all p < 0.001). SO₂ displayed weaker negative correlations (r = – 0.14 to – 0.09; Figure S1).

Hierarchical logistic regression (Table 2 and Table S4) revealed that only PM₂.₅ exhibited a consistent exposure–response relationship with psychiatric outcomes. In the fully-adjusted model (Model 3), each 1 μg/m³ rise in PM₂.₅ increased the odds of physical-psychological multimorbidity (OR = 1.01; 95% CI 1.00–1.03; p = 0.048) and isolated psych problems (OR = 1.01; 95% CI 1.00–1.02; p = 0.033). PM₁₀ showed a similar but weaker association for psych problems (OR = 1.01; 95% CI 1.00–1.02; p = 0.042). Estimates were stable across strata, although confidence intervals widened with adjustment. NO₂, O₃, PM₁ and SO₂ yielded null results (all 95% CIs crossed 1.00; p > 0.05). Fine particulate matter thus appears to be an independent negative factor for physical-psychological multimorbidity.

Table 2.

Association between air pollutants and Physical-psychological multimorbidity in different models.

Variable	Model 1		Model 2		Model 3
	OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value
NO2	1.01 (0.99 ~ 1.02)	0.541	1.01 (0.99 ~ 1.02)	0.541	1.01 (0.99 ~ 1.03)	0.33
O3	0.98 (0.96 ~ 1.01)	0.138	0.98 (0.96 ~ 1.01)	0.153	0.98 (0.96 ~ 1.01)	0.191
PM1	1.01 (0.99 ~ 1.02)	0.351	1.01 (0.99 ~ 1.03)	0.348	1.02 (0.99 ~ 1.04)	0.162
PM10	1 (1 ~ 1.01)	0.243	1 (1 ~ 1.01)	0.233	1.01 (1 ~ 1.02)	0.052
PM2.5	1.01 (1 ~ 1.02)	0.16	1.01 (1 ~ 1.02)	0.16	1.01 (1 ~ 1.03)	0.048
SO2	1 (0.99 ~ 1.01)	0.85	1 (0.99 ~ 1.01)	0.826	1 (0.99 ~ 1.02)	0.811

Associations between NDVI and psychiatric outcomes

Multivariable logistic regression analyses examined associations between neighborhood greenness (NDVI) and physical-psychological multimorbidity (Table 3) and psychiatric problems (Table S1), adjusting for progressively comprehensive covariate sets. In the fully adjusted model (Model 3), a significant inverse association emerged between continuous NDVI (per 0.1 unit increase) and both outcomes. Physical-psychological multimorbidity risk decreased by 19% (OR = 0.81, 95% CI 0.67–0.99, p = 0.035) and psychiatric problem risk decreased by 18% (OR = 0.82, 95% CI 0.68–0.97, p = 0.024). Analysis using categorical NDVI (High vs. Low) showed a non-significant trend towards reduced risk for multimorbidity (Model 3 OR = 0.73, 95% CI 0.50–1.05, p = 0.093) but a significant positive association for psychiatric problems (Model 3 OR = 0.68, 95% CI 0.48–0.97, p = 0.033). Quartile analysis revealed a notable pattern: compared to the lowest quartile (G1), the highest NDVI quartile (G4) was significantly associated with a 45% lower risk of physical-psychological multimorbidity (Model 3 OR = 0.55, 95% CI 0.31–0.98, p = 0.042), while the association with psychiatric problems approached significance (Model 3 OR = 0.61, 95% CI 0.35–1.04, p = 0.068). The intermediate quartiles (G2, G3) showed no significant associations for either outcome. Tests for linear trend across quartiles were not statistically significant in any model for either outcome (e.g., Physical-psychological multimorbidity Model 3 p-trend = 0.070).

Table 3.

Multivariable regression analysis of NDVI association with physical-psychological multimorbidity.

Variable	Model 1		Model 2		Model 3
	OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value
NVDI (Per0.1 unit)	0.87 (0.73 ~ 1.04)	0.115	0.87 (0.73 ~ 1.04)	0.126	0.81 (0.67 ~ 0.99)	0.035
NVDI_GROUP
Low	1(Ref)		1(Ref)		1(Ref)
High	0.82 (0.58 ~ 1.17)	0.271	0.82 (0.58 ~ 1.16)	0.261	0.73 (0.5 ~ 1.05)	0.093
NIDV(Quartile)
G1	1(Ref)		1(Ref)		1(Ref)
G2	1.13 (0.7 ~ 1.81)	0.627	1.16 (0.72 ~ 1.87)	0.552	1.14 (0.7 ~ 1.84)	0.606
G3	1.33 (0.83 ~ 2.12)	0.234	1.37 (0.86 ~ 2.2)	0.185	1.25 (0.75 ~ 2.08)	0.402
G4	0.58 (0.32 ~ 1.02)	0.061	0.58 (0.33 ~ 1.03)	0.064	0.55 (0.31 ~ 0.98)	0.042
P for trend	0.9 (0.77 ~ 1.05)	0.165	0.9 (0.77 ~ 1.05)	0.177	0.86 (0.73 ~ 1.01)	0.07

Model 1 Crude.

Mode 2 Age, Sex, Marital status, Education Status, Smoking Status, Drinking Status, Antihypertensive medicine, Anti Diabetes medicine.

Model 3 Age, Sex, Marital status, Education Status, Smoking Status, Drinking Status, Antihypertensive medicine, Anti Diabetes medicine, dry, temperature, precipitation, CRP, METS.

Breakpoint analysis revealed significant non-linear relationships between environmental exposure and psychiatric outcomes. For physical-psychological multimorbidity (Table 4), a threshold effect occurred at exposure level 0.344 (95% CI 0.342–0.346). Below this breakpoint, each unit exposure increase showed non-significant positive association with multimorbidity (OR = 1.25, 95% CI 0.83–1.88, p = 0.286, per 0,1 unit). Above the threshold, exposure demonstrated a non-significant positive trend (OR = 0.255, 95% CI 0.044–1.49, p = 0.129). The piecewise model significantly improved fit over linear regression (Likelihood Ratio test: p = 0.046; Non-linearity Test 1: p = 0.041).

Table 4.

Breakpoint analysis results for Physical-psychological multimorbidity.

Item	Breakpoint. OR (95%CI)	P value
E_BK1	0.344 (0.342 ,0.346)	NA
slope1(per 0.1 unit)	1.25 (0.83 ~ 1.881)	0.286
slope2(per 0.1 unit)	0.255 (0.044 ~ 1.488)	0.129
Likelihood Ratio test	–	0.046
Non-linear Test*1	–	0.041
Non-linear Test*2	–	0.061

For psychiatric problems (Table S2), a breakpoint occurred at lower exposure levels (0.328, 95% CI 0.325–0.330. Below-threshold associations were non-significant (OR = 1.16, 95% CI 0.77–1.74, p = 0.490, per 0,1 unit)), while above-threshold exposure showed marginally non-significant protection (OR = 0.319, 95% CI: 0.091–1.123, p = 0.075). The non-linear model provided significantly better fit than linear specification (Likelihood Ratio test: p = 0.015; Non-linearity Test 1: p = 0.034). Both outcomes demonstrated statistically significant threshold effects despite non-significant segment slopes. These results further reinforced the notion of nonlinear changes at breakpoints, suggesting that the mechanisms through which variables affect psychiatric problems may differ before and after the breakpoint, which aligns with the findings from curve fitting analyses illustrated in Fig. 2 and Fig. 2S.

Fig. 2.

Curve Analysis of NVDI and Psychological Problems.

Subgroup analysis of NDVI and psychiatric outcomes

Subgroup analyses (Fig. S3 and Fig. 3) evaluated how selected factors relate to psychiatric problems and physical-psychological multimorbidity.Smoking status and cooking-fuel type emerged as significant effect modifiers, underscoring their potential influence on mental health. In contrast, stratification by sex, age group, or residence revealed no significant differences in the associations, indicating that larger or more refined studies are warranted to clarify these null findings.

Fig. 3.

Subgroup Analysis of NVDI and Physical-psychological multimorbidity.

Associations of inflammatory markers and air pollutants with psychiatric outcomes

Multivariable logistic regression showed a linear, positive association between serum CRP and psychiatric outcomes (per 10-unit increase: multimorbidity OR = 1.22, 95% CI 1.06–1.42, p = 0.007; psychiatric problems OR = 1.20, 95% CI 1.04–1.39, p = 0.013) (Table 5 and Table S3). In this quartile-based analysis, no statistically significant linear dose–response trend was observed between C-reactive protein levels and psychiatric outcomes.

Table 5.

Multivariable regression analysis of C reactive protein association with physical-psychological multimorbidity.

Variable	Model1		Model 2		Model 3
	OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value
C Reactive Protein(per 10 units)	1.28 (1.12 ~ 1.47)	< 0.001	1.23 (1.06 ~ 1.42)	0.006	1.22 (1.06 ~ 1.42)	0.007
C Reactive Protein(Quartile)
G1	1(Ref)		1(Ref)		1(Ref)
G2	1.26 (0.74 ~ 2.13)	0.401	1.18 (0.69 ~ 2.01)	0.543	1.22 (0.71 ~ 2.07)	0.473
G3	1.03 (0.6 ~ 1.77)	0.913	0.95 (0.55 ~ 1.64)	0.858	0.99 (0.57 ~ 1.71)	0.964
G4	1.79 (1.1 ~ 2.89)	0.018	1.53 (0.94 ~ 2.51)	0.088	1.46 (0.86 ~ 2.46)	0.157
P for trend	1.19 (1.02 ~ 1.39)	0.029	1.13 (0.97 ~ 1.32)	0.127	1.1 (0.93 ~ 1.31)	0.246

Pathway analysis of NDVI and psychiatric outcomes via PM₁₀, PM₂.₅, and CRP

The Pathway analysis analysis investigated the connection between NDVI and physical-psychological multimorbidity, considering PM10, PM2.5, and CRP as potential mediators. The findings revealed that none of these mediators exhibited significant mediation effects of NDVI on physical-psychological multimorbidity (Table 6). In the present pathway analysis, the bootstrap 95% confidence intervals for both the indirect and total effects spanned zero, with all p-values > 0.05. Consequently, we found no empirical support for PM10, PM2.5, or CRP serving as single, intermediate mediators of the relationship between residential greenness and the development of physical-psychological multimorbidity.

Table 6.

Pathway Analysis Results for the Relationship between Physical-psychological multimorbidity , PM10,PM₂.₅, CRP and NVDI.

Model	Effect	Standard Error (S.E.)	95% confidence interval	p-value	z-value	Bootstrap 95% CI
NDVI → PM10 → Physical-psychological multimorbidity
Indirect effect (ab)	− 0.0016	0.0014	[− 0.0046, 0.0010]	0.2562	− 1.1354	[− 0.0046, 0.0010]
Direct effect (c')	− 0.0046	0.0042	[− 0.0146, 0.0014]	0.2754	− 1.0908	[− 0.0146, 0.0014]
Total effect (c)	− 0.0061	0.0041	[− 0.0157, − 0.0005]	0.1375	− 1.4854	[− 0.0157, − 0.0005]
NDVI → PM2.5 → Physical-psychological multimorbidity
Indirect effect (ab)	− 0.0016	0.0015	[− 0.0046, 0.0012]	0.2764	− 1.0884	[− 0.0046, 0.0012]
Direct effect (c')	− 0.0045	0.0042	[− 0.0146, 0.0014]	0.2848	− 1.0697	[− 0.0146, 0.0014]
Total effect (c)	− 0.0061	0.0041	[− 0.0158, − 0.0005]	0.1398	− 1.4766	[− 0.0158, − 0.0005]
NDVI → CRP → Physical-psychological multimorbidity
Indirect effect (ab)	0.0001	0.0001	[0.0000, 0.0002]	0.3039	1.0281	[− 0.0000, 0.0002]
Direct effect (c')	− 0.0061	0.004	[− 0.0159, − 0.0006]	0.1296	− 1.5158	[− 0.0159, − 0.0006]
Total effect (c)	− 0.006	0.004	[− 0.0158, − 0.0006]	0.132	− 1.5064	[− 0.0158, − 0.0006]

Sensitivity analysis

Sensitivity analyses comparing the same NDVI measures across two endpoints showed that a 0.1-unit increase in NDVI was often linked to a slightly higher risk of depression symptoms (OR = 1.06, P < 0.05). However, there was no clear link between NDVI and physical–depression multimorbidity (OR about 1.00, P > 0.9). A clear trend across NDVI quartiles was only seen for depression symptoms and became weaker after adjusting for weather, inflammation, and metabolic factors (Table S5, Table S6).

Discussion

This study used data from the 2015 CHARLS. It had a cross-sectional design. It included 6,900 older adults aged 45 and above. The goal was to analyze the relationship between NDVI and mental health issues or other health conditions. The research combined satellite data with biomarker information. This allowed for a detailed look at environmental effects on mental health. The main findings revealed that for every 0.1-unit increase in NDVI, there was a significant 19% decrease in the risk of physical–psychological multimorbidity (OR = 0.81, 95% CI 0.67–0.99) and an 18% decrease in the risk of psychiatric problems (OR = 0.82, 95% CI 0.68–0.97).After multivariable adjustment, NDVI remained protective, these findings suggest that greater vegetation coverage has positive associations on psychiatric health. Because exposure and outcome were measured concurrently, reverse causality cannot be ruled out. All observed associations are therefore exploratory and require longitudinal confirmation. Our data reveal only a weak, non-significant trend toward a protective effect of green space on multimorbidity, without detectable mediation by air pollution or systemic inflammation. Sensitivity analyses revealed that NDVI exhibited a weak yet significant positive association with “pure” depressive symptoms (OR ≈ 1.06 per 0.1-unit increase) but was essentially unrelated to physical–depression multimorbidity (OR ≈ 1.00). These divergent patterns indicate that the greenness effect is contingent upon outcome definition, yielding null findings for comorbid depression and an overall weak, inconsistent relationship. Thus, the magnitude, direction, and statistical significance of the NDVI–depression link are not robust across varying diagnostic criteria.

The study also revealed that for every 0.1-unit increase in NDVI, there was a significant reduction in the risks of psychiatric issues and physical-psychological multimorbidity. This finding aligns with the research conducted by Esmat et al., which emphasizes the positive association of environmental psychiatric factors on mental health²⁸. A study found no consistent link between residential greenness and adolescent mental health trajectories, but exploratory analysis showed higher childhood residential greenness correlated with reduced externalizing symptoms over time in males (not females), noting the complex relationship needs further research²⁹.

First, residential greenness offers residents enhanced social opportunities and psychiatric support. Public green spaces, including greenbelts and parks, function not only as areas for physical exercise but also as vital platforms for social interaction. When residents participate in activities within these green spaces, they foster connections with others, leading to increased social support, which positively influences their psychiatric health^30,31. Second, residential greenness can enhance psychiatric health by mitigating noise and air pollution. Research indicates that regions with increased greenery often experience reduced noise levels and improved air quality. These environmental factors are linked to decreased psychiatric distress and more favorable psychiatric health outcomes³². Air pollution has a detrimental effect on psychiatric health, extending beyond its well-known association with respiratory diseases to include a range of psychiatric issues such as depression and anxiety. Research indicates that exposure to polluted air can negatively influence mental well-being by elevating oxidative stress and triggering inflammatory responses in the body. These biological mechanisms may contribute to the development or exacerbation of mental health problems, highlighting the importance of addressing air quality not only for physical health but also for psychiatric well-being^33,34. Low-grade inflammation is a key mediator linking the natural environment and social factors to diabetes pathogenesis through aerobiological and epigenetic pathways^35,36. Systematic reviews indicate that exposure to both outdoor (urban parks) and indoor (office plants) green spaces significantly improves mental and physical health—reducing anxiety, depression, and blood pressure—while simultaneously enhancing social cohesion and environmental quality³⁷. Consequently, integrating green spaces into public health policy and urban planning is strongly recommended.

The clinical implications and potential impacts of this study warrant in-depth discussion. By systematically evaluating the comprehensive effects of the NDVI and air pollution on psychiatric health among older adults, this study suggests the significant role that environmental psychiatric factors play in managing psychiatric health³⁸.

This finding not only offers a fresh perspective for psychiatric health interventions but also provides scientific evidence that can inform clinical practices and the development of public health policies. Unlike existing literature, this study uniquely integrates high-resolution satellite remote sensing data with biomarker information, allowing for a more thorough assessment of environmental psychiatric exposure. Additionally, building on prior European cohort studies that have examined similar pathways, the present study adds new cross-sectional evidence from a middle-aged and older Chinese population and quantifies the relative contribution of the hypothesised pathways. These insights pave the way for future research, especially in examining the mechanisms that underlie the effects of environmental psychiatric interventions on psychiatric well-being.Clinically, enhancing greening levels and air quality in residential environments may serve as an effective strategy to alleviate psychiatric problems among older adults³⁹.

Above the breakpoint, each 0.1-unit rise in NDVI was associated with a 75% lower risk of physical-psychological multimorbidity (OR = 0.255, 95% CI 0.04–1.49), although the wide CI reflects the smaller number of participants in the highest greenness stratum. Additionally, public health planning should integrate policies and measures aimed at reducing air pollution to effectively address the high prevalence of psychiatric issues in vulnerable groups⁴⁰. Future research should establish a prospective cohort with quarterly GPS-linked individual exposure monitoring and repeated mental-health assessments to confirm temporal ordering. A nested multi-omics sub-cohort will integrate DNA-methylation, neuro-imaging and gut-microbiome signatures to mechanistically dissect the biological pathways linking green-space exposure, systemic inflammation and brain health.

Limitations

Limitations come from a study based in a single area with mostly Han participants across 150 counties, which makes it hard to apply findings everywhere. The study’s cross-sectional design also means we can’t determine cause and effect and it might be affected by unmeasured factors like community safety, social bonds, genetics, and preferences for green spaces. The only psychiatric data was based on self-report, which probably underestimates how common mental health issues really are because of stigma and different diagnostic standards. For better accuracy, future research should use established tools like PHQ-9 or GAD-7. The satellite data on PM₂.₅ levels with annual updates may underestimate the indirect effect (about 26%), and the yearly NDVI data within a 250-m radius doesn’t fully capture individual exposure due to movement, indoor environments, commuting, and seasonal changes. Plus, without repeated biological markers, it’s hard to track long-term changes. Because of this, the NDVI threshold value found (0.344) and the proposed pathway—from green space to air pollution, to inflammation, and then to mental health—need to be tested in long-term, diverse groups with careful control for factors like noise, quality of green spaces, diet, and genetics.

Abbreviations

CHARLS

China health and retirement longitudinal study

NDVI

Normalized difference vegetation index

PM2.5

Particulate matter with a diameter of 2.5 µm or less

PM10

Particulate matter with a diameter of 10 µm or less

NO₂

Nitrogen dioxide O₃:ozone

SO₂

Sulfur dioxide

CRP

C-reactive protein

OR

Odds ratio

CI

Confidence interval

MDD

Major depressive disorder

CVD

Cardiovascular disease

METs

Metabolic equivalent of task

JIS

Joint Interim societies

LPG

Liquefied petroleum gas

STET

Spatiotemporal extreme random trees model

SD

Standard deviation

Author contributions

F.W. was responsible for designing the study and collecting the data. L.LY offered financial support and played a role in analyzing and interpreting the data. C.GB, Z.JL, and D.Y. were tasked with preparing the figures and tables. Y.G. helped with data collection. All authors were involved in writing and revising the manuscript.

Funding

Funds of 925hospital (2025-YNKT-04/05).

Data availability

The data used in this study are publicly available. They come from the charls database. You can access it at: https://charls.pku.edu.cn/.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The Medical Ethics Committee of the 925th Hospital approved this study (approval No. YNKT202504). As the data utilized were anonymized and accessible to the public via the CHARLS website, the 925th Hospital Review Board classified this study as “non—human subjects” research. Prior to the study, written informed consent was acquired from each participant. This research complied with the principles outlined in the Declaration of Helsinki and received approval from the Ethics Committee of Peking University (IRB00001052-11015). All participants of the China Health and Retirement Longitudinal Study (CHARLS) gave their written informed consent, and the data were anonymized prior to public dissemination. This secondary analysis used fully de-identified, publicly available CHARLS data; individual identifiers were never accessed. Procedures conformed to the CHARLS Data Use Agreement: no individual-level disclosure and files stored on encrypted servers.

Consent for publication

Each author has reviewed and endorsed the manuscript’s final version and agrees to its dissemination. The corresponding author attests that all authors are aware of the submission and have authorized its publication.