Association between residential greenness and risks of diabetic complications among individuals with type 2 diabetes and the roles of healthy lifestyle behaviors

Abstract

Prospective evidence on the associations between residential greenness and risks of major diabetic complications remains limited. We explored the associations of residential greenness with incident diabetic macrovascular and microvascular complications among individuals with type 2 diabetes (T2D), and further assessed the roles of lifestyle factors in these associations. Our study included 13,848 patients with T2D. The Normalised Difference Vegetation Index (NDVI) within a 300-m buffer was used as the indicator of residential greenness. Eight lifestyle factors were included to construct a weighted healthy lifestyle score. Cox proportional hazard models were performed to assess the links between residential greenness and diabetic complication risks. We found that each interquartile range increase of NDVI within 300-m buffer was significantly associated with decreased diabetic complication risks, with hazard ratios (95% confidence intervals) of 0.921 (0.881, 0.964) for composite macrovascular complications, 0.886 (0.805, 0.975) for stroke, 0.901 (0.825, 0.985) for heart failure, 0.914 (0.866, 0.965) for coronary artery disease, 0.909 (0.865, 0.956) for composite microvascular complications, 0.842 (0.757, 0.937) for diabetic neuropathy, and 0.899 (0.841, 0.961) for diabetic nephropathy. Air pollutants and physical activity partly mediated these associations. Lifestyle factors modified some residential greenness-complications associations, and no significant associations between residential greenness and diabetic complications were observed among participants with poor lifestyles. Higher residential greenness levels were linked with lower risks of diabetic complications, and lifestyle factors might modify these associations.

Graphical abstract

Highlights

• •Residential greenness could be linked to reduced risks of diabetic complications.
• •Air pollutants and physical activity could mediate these associations.
• •Residential greenness showed no effect on diabetic complications in those with poor lifestyles.

1. Introduction

Diabetes is a serious chronic non-communicable disease; in 2021, there were over 500 million people living with diabetes worldwide. Type 2 diabetes (T2D) made up over 95.0% of all diabetes cases [1]. Complications of diabetes, including macrovascular and microvascular complications, are the major causes of disability and death in patients with T2D. Macrovascular complications include coronary artery disease (CAD), heart failure (HF), stroke, and peripheral artery disease (PAD) [2]. Studies have shown that people with diabetes have more than twice the risk of these complications compared to people without diabetes [3]. Diabetic microvascular complications, including diabetic retinopathy, diabetic nephropathy, and diabetic neuropathy, are also highly prevalent in individuals with T2D [4]. These microvascular complications can lead to serious consequences such as blindness, amputation, and kidney failure, reducing the quality of patients’ lives and imposing significant societal burdens [4]. Therefore, identifying factors associated with diabetic complications and developing targeted protective measures are important to reduce the incidence of macrovascular and microvascular complications in individuals with T2D.

Residential greenness plays a vital role in health promotion [5,6] and many studies have linked residential greenness to decreased risk of diabetes [[7], [8], [9]]. Existing research has also identified the multiple beneficial pathways of residential greenness, including reducing oxidative stress and systemic inflammation, alleviating psychological stress, mitigating the urban heat island effect and exposure to air pollution, optimizing blood glucose control, and promoting physical activity [[10], [11], [12], [13], [14]], which might potentially account for the links between residential greenness and incident diabetic complications. For example, current research has linked higher residential greenness exposure to lower blood glucose levels [11,15]—a key biomarker strongly associated with the incidence of diabetic complications [16]. Additionally, emerging evidence suggests that residential greenness might mitigate psychological stress [10,17], another critical factor in the development and progression of diabetic complications [18]. Recently, a Spanish cohort study found that increased residential greenness exposure was associated with decreased myocardial infarction (MI) risk among people with diabetes [19]. However, this finding was not replicated in an Australian cross-sectional study of 4166 T2D patients, which showed no significant associations between residential greenness and incident cardiovascular disease (CVD) or MI [20]. Regarding microvascular complications, a Chinese cross-sectional study demonstrated an inverse linear relationship between residential greenness and diabetic retinopathy prevalence [21]. However, current studies have several limitations, including the use of cross-sectional designs and limited samples in some studies, inconsistent findings across different populations, and a lack of research on other important diabetic complications (such as nephropathy and neuropathy). Therefore, prospective cohort studies with large samples are important for systematically investigating the relationships of residential greenness with the incidence of various complications in individuals with diabetes.

A growing number of studies have identified significant protective effects of healthy lifestyle factors on incident diabetic complications [22,23]. Notably, these factors may also interact with residential greenness to influence health outcomes. Regarding physical activity, individuals who engage in regular exercise may benefit more from residential greenness because they may spend more time outdoors and have greater exposure to it, whereas sedentary individuals might not experience the same protective effects [24]. Conversely, smoking, poor dietary habits, and poor sleep quality may exacerbate oxidative stress and systemic inflammation, and elevate blood glucose levels, potentially offsetting the anti-inflammatory and metabolic benefits associated with residential greenness [[25], [26], [27]]. For example, a cohort study in China identified that the association of residential greenness with reduced CVD mortality was more evident among those maintaining healthier lifestyles, including non-smokers and non-drinkers [28]. Similarly, a study of 403,748 Americans found significant residential greenness-mortality associations only among physically active individuals [29]. However, the roles of healthy lifestyles in the associations of residential greenness exposure with incident diabetic complications are unclear.

Therefore, based on a large-scale cohort, we explored the relationships between exposure to residential greenness and the risks of multiple diabetic complications in participants with T2D. In addition, we examined the interaction effects of residential greenness with lifestyle factors on incident diabetic complications.

2. Materials and methods

2.1. Study design

Participants in this research were drawn from the UK Biobank, a nationwide prospective cohort study previously detailed elsewhere [30]. Briefly, over 500,000 individuals aged 37 to 73 were enrolled between 2006 and 2010 and have been followed up for a long duration. Data collection included questionnaires, physical examinations, biological samples, and standard medical records. The UK Biobank protocol was approved by the North West Multicenter Research Ethical Committee (16/NW/0274), with informed consent obtained from all participants. The current analysis was conducted under application number 69741.

According to the algorithms provided by the UK Biobank study [31], we identified 24,064 individuals with T2D at baseline. Then, those without complete information on residential greenness (N = 172) or lifestyle behaviors (N = 6112), as well as those with histories of diabetic macrovascular complications (N = 3288) and microvascular complications (N = 644) were excluded. Finally, a total of 13,848 participants with T2D were included in this study, and Fig. S1 shows the flow chart of this study.

2.2. Residential greenness exposure

This study used Normalised Difference Vegetation Index (NDVI), derived from MODIS satellite data, to quantify the levels of residential greenness [32]. In this study, data were sourced from the 250-m resolution, 16-d composite remote sensing product (https://modis.gsfc.nasa.gov/data/). In satellite remote sensing images, NDVI is calculated as: NDVI = (NIR − Red)/(NIR + Red), where Red is the reflected value of the red light band and Near Infrared (NIR) represents the reflected value of the near-infrared band.

NDVI values range from −1 to +1, with higher values denoting denser vegetation and negative values typically representing water. In this study, NDVI values were limited to greater than 0. According to previous research [32], we calculated NDVI levels during the summer to characterize residential greenness in the study population. Based on the location coordinates of the participants’ home addresses provided by the UK Biobank, we matched each participant with their annual average residential greenness exposure levels from their enrollment year to 2020. In the main analysis, we examined NDVI within a 300-m buffer (NDVI_300m) at baseline (i.e., the average value of the enrollment year) as the level of residential greenness, since 300 m (a 5-min walk) is the threshold beyond which residential greenness accessibility begins to decline rapidly [33]. Additionally, we also calculated the time-weighted average NDVI_300m based on the duration of residence at each address during the follow-up period. The follow-up period spanned from enrollment until the occurrence of diabetic complications, death, or the end of the follow-up (whichever came first), and we used the time-weighted average NDVI_300m as the indicator of residential greenness level in the sensitivity analysis. We also calculated NDVI values within 500-m and 1000-m buffer zones at baseline and employed them as residential greenness indicators in sensitivity analyses.

2.3. Healthy lifestyle score

Based on previous research [22,34], eight modifiable behavioral factors were identified to create a healthy lifestyle score. These factors included smoking, physical activity, alcohol intake, waist circumference (WC), diet, sleep duration, sedentary behavior, and social connections. WC was measured using the Wessex non-elastic spring tape measure, while other lifestyle data were collected through self-report at baseline.

In this study, the healthy lifestyle factors included: (1) no current smoking, including never-smokers (those who had never smoked) and former smokers (those who had quit smoking); (2) participating in regular physical activity, such as a weekly total of 150 min of moderate or 75 min of vigorous exercise; (3) moderate alcohol consumption (up to 1 drink per day for female and 2 for male); (4) low WC (<80 cm for female and <94 cm for male); (5) maintaining a healthy diet (adequate intake from at least half of the ten recommended food groups); (6) sufficient sleep (7−8 h per day); (7) limited sedentary time (<4 h per day); (8) maintaining active social connections (not being isolated). Text S1 and Table S1 describe the assessment of these lifestyles in detail. In this study, each healthy lifestyle factor was assigned a score of 1 and an unhealthy lifestyle factor was scored as 0. The weighted healthy lifestyle scores [35] were subsequently calculated, as described in Text S2.

2.4. Definition of outcomes

Participants were followed up for incident diabetic complications, death, or the end date of this analysis (December 31, 2020). Through linkages to hospital admission records and death registers, incident diabetic macrovascular and microvascular complications were identified according to the International Classification of Diseases Tenth Revision. Details about the codes for these outcomes are described in Table S2.

2.5. Covariates and mediators

According to the directed acyclic graph (Fig. S2), the covariates include age (years), sex (male, female), ethnicity (White Europeans, non-White Europeans), residential area (urban, rural), educational levels (college or university degree, other), employment status (employed, other), family income (≥£31,000, <£31,000), high cholesterol (yes, no), hypertension (yes, no), family history of diabetes (yes, no), time spent outdoors in summer (hours/day), diabetes duration (years), diabetes medication usage (yes, no), and glycosylated hemoglobin A1c (HbA_1c) (mmol/mol).

Baseline nitrogen dioxide (NO₂) and fine particulate matter (PM_2.5) concentrations were obtained from the Department for Environment, Food and Rural Affairs (https://uk-air.defra.gov.uk), and detailed information is described in Text S3.

Table S3 lists the percentages of missing values of the covariates in this study. For variables with missing data rates of 5% or higher, missing values were assigned to a separate category. Otherwise, for missing data, mode values were used for categorical variables and mean values were used for continuous variables in this study.

2.6. Statistical analyses

Categorical data were summarized as frequencies and percentages, while continuous measures were reported as medians with interquartile ranges for skewed distributions or means ± standard deviations (SDs) for normally distributed variables. Group comparisons were performed using the Chi-square test for categorical measures, and using either the Kruskal–Wallis test or one-way ANOVA for continuous variables, as appropriate based on distributional assumptions.

To identify the links between residential greenness and the risks of diabetic complications, Cox proportional hazards models were fitted. The hazard ratios (HRs) and 95% confidence intervals (CIs) for each increase in interquartile range (IQR) for residential greenness were documented. Three models were constructed. Model 1 served as the unadjusted baseline model. Model 2 incorporated adjustments for age, sex, ethnicity, educational level, employment status, residential area, family income, family history of diabetes, time spent outdoors in summer, hypertension, and high cholesterol. Model 3 was adjusted for age, sex, ethnicity, educational level, employment status, residential area, family income, family history of diabetes, time spent outdoors in summer, hypertension, high cholesterol, diabetes duration, usage of diabetes medication, and HbA_1c level. The proportional hazards assumption was assessed using Schoenfeld residuals, and no violations were found. Restricted cubic spline (RCS) models with three knots (at the 10th, 50th, and 90th percentiles) were implemented to investigate the exposure-response associations of NDVI_300m with the risks of diabetic complications. We further investigated whether the associations between residential greenness and incident diabetic complications were mediated by air pollutants (PM_2.5 and NO₂) and physical activity (regular/not regular). Detailed information on mediation analysis is presented in Text S4. In addition, stratified analyses were conducted among participants of different sexes (male, female), age groups (<65 years old, ≥65 years old), residential areas (urban, rural), ethnic groups (White Europeans, non-White Europeans), glycemic control status (HbA_1c < 55 mmol/mol for optimal control vs. ≥55 mmol/mol), hypertension (yes, no), and high cholesterol (yes, no).

We then considered exposure levels above the mean value of NDVI_300m as high residential greenness exposure levels and assessed the HRs and corresponding 95% CIs of high residential greenness exposure, which were then applied to calculate the population attributable fraction (PAF) and corresponding 95% CIs (%) of diabetic complications potentially preventable under high residential greenness exposure. Considering the potential negative associations between residential greenness and diabetic complications, we further calculated the prevented fraction (PF) and corresponding 95% CIs (%) based on the calculated PAF and corresponding 95% CIs [36], which represents the percentage of complications that could be avoided if everyone were exposed to high levels of residential greenness: $PF=1-1/\left(1-PAF\right)$.

We further examined the connections between the weighted healthy lifestyle scores and incident diabetic complications. We evaluated potential effect modification by modeling the multiplicative interaction between residential greenness and the healthy lifestyle score. We also examined the interactions between each lifestyle factor and residential greenness in relation to incident diabetic complications.

Several sensitivity analyses were performed. 1) We excluded participants with the onset of complications within two years of follow-up. 2) We excluded participants who had lived at their current address for no more than 4 years. 3) We excluded participants with missing covariates. 4) We conducted analyses after performing multiple imputation on the missing covariates. 5) The competing risk of death was considered. 6) We further adjusted for healthy lifestyle scores based on Model 3. 7) We adjusted for specific types of diabetes medicine, including usage of insulin, metformin, and other diabetes medication, rather than the usage of any diabetes medication in Model 3. 8) We further adjusted for usage of anti-hypertensive medication and lipid-lowering medication based on Model 3. 9) We employed NDVI within 500-m buffer as the residential greenness indicator. 10) We employed NDVI within a 1000-m buffer as the residential greenness indicator. 11) We employed time-weighted NDVI_300m during the follow-up period as the indicator of residential greenness exposure. 12) Inverse probability weighting was employed to adjust for selection bias resulting from disparities in baseline characteristics between the final analytical sample and the initial study population. 13) Marginal structural Cox models were performed to identify the potential causal links between residential greenness and incident diabetic complications, and Text S5 describes this method in detail. 14) E-values of the associations of residential greenness exposure with incident diabetic complications were calculated to assess the effects of unmeasured or unadjusted confounders.

We conducted all statistical analyses with R software (version 4.2.2), adopting a two-sided P < 0.05 as the threshold for statistical significance.

3. Results

Table 1 summarizes the baseline characteristics of the study cohort, categorized according to NDVI_300m quartiles. The mean (SD) age of the 13,848 participants with T2D was 59.4 (7.1) years, and 8492 (61.3%) were males. Participants with higher NDVI_300m were older and more likely to be male, White Europeans, and highly educated. Moreover, they were more inclined to have higher incomes and spend more time outside in the summer. As listed in Table S4, participants included in this study had higher educational levels and income and experienced shorter diabetes duration compared with the broader cohort. Table S5 shows the distribution of NDVI within different buffers, and the mean value of NDVI_300m was 0.56.

Table 1.

Descriptive characteristics of study participants based on quartiles of Normalised Difference Vegetation Index within a 300-m buffer.

Baseline characteristics	Total participants	Quartile 1	Quartile 2	Quartile 3	Quartile 4	P
Number	13,848	3464	3461	3462	3461	-
Age, years	59.4 ± 7.1	58.5 ± 7.5	59.3 ± 7.2	59.5 ± 7.0	60.2 ± 6.7	<0.001
Sex, male	8492 (61.3)	2041 (58.9)	2117 (61.2)	2149 (62.1)	2185 (63.1)	0.003
Ethnicity, White Europeans	12,320 (89.0)	2733 (78.9)	3119 (90.1)	3168 (91.5)	3300 (95.3)	<0.001
Education, college or university degree	3858 (27.9)	955 (27.6)	880 (25.4)	974 (28.1)	1049 (30.3)	<0.001
Employment status, employed	6156 (44.5)	1592 (46.0)	1464 (42.3)	1566 (45.2)	1534 (44.3)	0.014
Residential area, urban	12,228 (88.3)	3341 (96.4)	3202 (92.5)	3115 (90.0)	2570 (74.3)	<0.001
Family income, ≥£31,000	4807 (34.7)	1028 (29.7)	1122 (32.4)	1284 (37.1)	1373 (39.7)	<0.001
Family history of diabetes	4291 (31.0)	1107 (32.0)	1051 (30.4)	1103 (31.9)	1030 (29.8)	0.121
Hypertension	8180 (59.1)	2043 (59.0)	2106 (60.8)	2045 (59.1)	1986 (57.4)	0.035
High cholesterol	940 (6.8)	247 (7.1)	234 (6.8)	231 (6.7)	228 (6.6)	0.816
Time spent outdoors in summer, hours/day	4.0 [2.0, 5.0]	3.0 [2.0, 5.0]	4.0 [2.0, 5.0]	4.0 [2.0, 5.0]	4.00 [2.0, 5.0]	<0.001
Diabetes duration, years	5.0 [2.0, 9.0]	5.0 [2.0, 9.0]	5.0 [2.0, 9.0]	5.0 [2.0, 9.0]	5.00 [2.0, 9.0]	0.456
Diabetes medication usage	9334 (67.4)	2403 (69.4)	2355 (68.0)	2326 (67.2)	2250 (65.0)	0.001
HbA1c, mmol/mol	50.1 [43.1, 56.8]	50.5 [43.2, 57.3]	50.0 [43.2, 56.2]	49.8 [42.9, 56.9]	50.1 [42.9, 56.7]	0.495

During a median follow-up of 11.51 years, 3132 participants developed macrovascular complications, including 678 cases of stroke, 814 cases of HF, 2188 cases of CAD, and 565 cases of PAD; 2530 participants developed microvascular complications, including 1133 cases of diabetic retinopathy, 525 cases of diabetic neuropathy, and 1420 cases of diabetic nephropathy. We observed significant associations of residential greenness with diabetic macrovascular complications (Table 2). After fully adjusting for covariates, for each IQR increase in NDVI_300m, the HR (95% CI) was 0.921 (0.881, 0.964) for composite macrovascular complications, 0.886 (0.805, 0.975) for stroke, 0.901 (0.825, 0.985) for HF, and 0.914 (0.866, 0.965) for CAD. Similar associations were also observed when we fitted models with categorical NDVI_300m exposure, with P_trend < 0.001. However, we found no association between NDVI_300m and PAD risk. We further calculated the PFs for macrovascular complications associated with high residential greenness exposure (Table 2). Among all participants with T2D, an estimated 6.39% of new cases of composite macrovascular complications, 8.84% of HF, and 7.05% of CAD might be avoided by high residential greenness exposure. Figs. 1 and S3 demonstrate the linear relationships of residential greenness exposure with the risks of composite macrovascular complications, stroke, HF, and CAD (all P_{non-linearity} > 0.05).

Table 2.

Associations of residential greenness with the risk of macrovascular and microvascular complications among type 2 diabetes.

	Continuous, per IQR increase		Q1, NDVI < 0.50	Q2, 0.50 ≤ NDVI < 0.56	Q3, 0.56 ≤ NDVI < 0.63	Q4, NDVI ≥ 0.63	Ptrend	PF (95% CI), %
	HR (95% CI)	P	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)
Composite macrovascular complications (cases, 3132)
Model 1	0.927 (0.889, 0.968)	<0.001	Ref.	0.992 (0.899, 1.095)	0.894 (0.809, 0.987)	0.850 (0.770, 0.939)	<0.001
Model 2	0.920 (0.880, 0.963)	<0.001	Ref.	0.971 (0.879, 1.073)	0.899 (0.813, 0.995)	0.845 (0.761, 0.937)	<0.001
Model 3	0.921 (0.881, 0.964)	<0.001	Ref.	0.977 (0.885, 1.080)	0.894 (0.808, 0.989)	0.853 (0.769, 0.941)	0.001	6.39 (3.07, 9.50)
Stroke (cases, 678)
Model 1	0.887 (0.811, 0.970)	0.009	Ref.	0.988 (0.800, 1.220)	0.870 (0.703, 1.076)	0.803 (0.648, 0.995)	0.022
Model 2	0.880 (0.799, 0.968)	0.009	Ref.	0.980 (0.792, 1.213)	0.877 (0.707, 1.088)	0.800 (0.640, 1.000)	0.029
Model 3	0.886 (0.805, 0.975)	0.013	Ref.	0.984 (0.795, 1.217)	0.871 (0.702, 1.082)	0.810 (0.647, 1.013)	0.035	5.06 (−2.71, 12.65)
HF (cases, 814)
Model 1	0.907 (0.835, 0.986)	0.021	Ref.	0.965 (0.797, 1.167)	0.813 (0.670, 0.987)	0.762 (0.627, 0.926)	0.002
Model 2	0.896 (0.820, 0.980)	0.016	Ref.	0.930 (0.768, 1.127)	0.804 (0.660, 0.978)	0.749 (0.612, 0.917)	0.002
Model 3	0.901 (0.825, 0.985)	0.022	Ref.	0.941 (0.776, 1.141)	0.800 (0.658, 0.974)	0.762 (0.622, 0.934)	0.003	8.84 (2.40, 14.41)
CAD (cases, 2188)
Model 1	0.925 (0.880, 0.973)	0.003	Ref.	0.983 (0.874, 1.107)	0.877 (0.779, 0.989)	0.852 (0.759, 0.960)	0.002
Model 2	0.912 (0.864, 0.963)	0.001	Ref.	0.974 (0.864, 1.097)	0.889 (0.788, 1.003)	0.838 (0.740, 0.948)	0.002
Model 3	0.914 (0.866, 0.965)	0.001	Ref.	0.975 (0.865, 1.099)	0.886 (0.785, 1.000)	0.842 (0.743, 0.953)	0.002	7.05 (3.07, 10.68)
PAD (cases, 565)
Model 1	0.934 (0.846, 1.032)	0.179	Ref.	0.991 (0.784, 1.252)	0.997 (0.792, 1.254)	0.830 (0.654, 1.053)	0.148
Model 2	0.902 (0.811, 1.004)	0.059	Ref.	0.905 (0.716, 1.146)	0.952 (0.755, 1.200)	0.779 (0.610, 0.997)	0.082
Model 3	0.902 (0.811, 1.003)	0.057	Ref.	0.910 (0.719, 1.151)	0.932 (0.739, 1.176)	0.782 (0.611, 1.000)	0.075	3.58 (−5.16, 10.99)

Model 1 was a crude model; Model 2 was adjusted for age, sex, ethnicity, educational level, employment status, residential area, family income, family history of diabetes, time spent outdoors in summer, and prevalence of hypertension and high cholesterol; Model 3 was adjusted for age, sex, ethnicity, educational level, employment status, residential area, family income, family history of diabetes, time spent outdoors in summer, prevalence of hypertension and high cholesterol, diabetes duration, usage of diabetes medication, and HbA_1c level. HF, heart failure; CAD, coronary artery disease; PAD, peripheral artery disease; IQR, interquartile range; HR, hazard ratio; CI, confidence interval; PF, prevented fraction.

Fig. 1.

Exposure-response relationships between residential greenness and the risks of diabetic macrovascular (A) and microvascular (B) complications in individuals with type 2 diabetes. Model as defined in Model 3 of Table 2.

Residential greenness was also significantly associated with incident microvascular complications (Table 3). For each IQR increase in NDVI_300m, we observed statistically significant associations with decreased risks of composite microvascular complications (HR: 0.909; 95% CI: 0.865, 0.956), diabetic neuropathy (HR: 0.842; 95% CI: 0.757, 0.937), and diabetic nephropathy (HR: 0.899; 95% CI: 0.841, 0.961). Among all individuals with T2D, an estimated 7.46% of new cases of composite microvascular complications, 9.25% of diabetic neuropathy, and 9.04% of diabetic nephropathy could be avoided by high residential greenness exposure (Table 3). The RCS curves exhibit monotonic associations between residential greenness and incident composite microvascular complications, diabetic neuropathy, and diabetic nephropathy (all P_{non-linearity} > 0.05) (Figs. 1 and S4). We also found that air pollutants (PM_2.5 and NO₂) and physical activity played significant mediating roles in the connections between residential greenness and the risks of stroke, HF, CAD, composite macrovascular complications, diabetic neuropathy, diabetic nephropathy, and composite microvascular complications (Table S6).

Table 3.

Associations of residential greenness with the risk of macrovascular and microvascular complications among type 2 diabetes.

	Continuous, per IQR increase		Q1, NDVI <0.50	Q2, 0.50 ≤ NDVI <0.56	Q3, 0.56 ≤ NDVI <0.63	Q4, NDVI ≥0.63	Ptrend	PF (95% CI), %
	HR (95% CI)	P	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)
Composite microvascular complications (cases, 2530)
Model 1	0.903 (0.862, 0.946)	<0.001	Ref.	0.961 (0.863, 1.071)	0.875 (0.784, 0.975)	0.774 (0.692, 0.865)	<0.001
Model 2	0.900 (0.857, 0.946)	<0.001	Ref.	0.953 (0.854, 1.063)	0.881 (0.789, 0.984)	0.775 (0.690, 0.871)	<0.001
Model 3	0.909 (0.865, 0.956)	<0.001	Ref.	0.973 (0.872, 1.085)	0.882 (0.789, 0.985)	0.792 (0.705, 0.891)	<0.001	7.46 (3.73, 10.78)
Diabetic retinopathy (cases, 1133)
Model 1	0.888 (0.828, 0.952)	0.001	Ref.	0.893 (0.757, 1.053)	0.958 (0.815, 1.120)	0.705 (0.595, 0.835)	0.001
Model 2	0.916 (0.850, 0.987)	0.021	Ref.	0.934 (0.791, 1.104)	1.012 (0.861, 1.189)	0.752 (0.630, 0.898)	0.015
Model 3	0.931 (0.864, 1.004)	0.064	Ref.	0.969 (0.820, 1.146)	1.040 (0.885, 1.225)	0.797 (0.660, 0.953)	0.044	2.71 (−3.53, 8.23)
Diabetic neuropathy (cases, 525)
Model 1	0.845 (0.764, 0.934)	0.001	Ref.	0.862 (0.680, 1.093)	0.855 (0.678, 1.079)	0.660 (0.516, 0.845)	0.002
Model 2	0.834 (0.749, 0.928)	0.001	Ref.	0.829 (0.653, 1.053)	0.851 (0.612, 1.076)	0.619 (0.502, 0.839)	0.002
Model 3	0.842 (0.757, 0.937)	0.002	Ref.	0.853 (0.672, 1.084)	0.851 (0.672, 1.078)	0.670 (0.518, 0.867)	0.004	9.25 (1.18, 15.99)
Diabetic nephropathy (cases, 1420)
Model 1	0.891 (0.837, 0.948)	<0.001	Ref.	0.983 (0.852, 1.135)	0.817 (0.705, 0.947)	0.748 (0.645, 0.867)	<0.001
Model 2	0.893 (0.836, 0.955)	0.001	Ref.	0.963 (0.833, 1.113)	0.832 (0.717, 0.966)	0.759 (0.651, 0.886)	<0.001
Model 3	0.899 (0.841, 0.961)	0.002	Ref.	0.979 (0.846, 1.131)	0.831 (0.716, 0.965)	0.776 (0.665, 0.906)	<0.001	9.04 (4.24, 13.34)

Models identical to those in Table 2.

Stratified analyses were performed (Tables S7 and S8), and we found that ethnicity might modify the associations of residential greenness with incident composite microvascular complications and diabetic nephropathy (P_interaction < 0.05). Moreover, the associations of residential greenness with incident diabetic complications were stronger among those with optimal glycemic control, although no significant interactions were identified. We also performed several sensitivity analyses (Tables S9–S22). The results of our study remained consistent when using the time-weighted average NDVI_300m during the follow-up period as the indicator of residential greenness (Table S19). Inverse probability-weighted adjusted models also produced estimates consistent with the primary results (Table S20). The results of marginal structural Cox models suggested potential causal links between residential greenness and the risks of diabetic complications (Table S21).

Tables S23 and S24 show significant associations between healthy lifestyle scores and multiple diabetic complication risks, except diabetic retinopathy. As shown in Table 4, healthy lifestyle scores could modify the associations of residential greenness with incident HF, composite microvascular complications, diabetic retinopathy, and diabetic nephropathy (all P_interaction < 0.05). Notably, the associations of residential greenness and all diabetic complications became statistically non-significant in the lowest healthy lifestyle score quartile. Stratified analyses by individual healthy lifestyle factors were also performed (Tables S25 and S26). Associations of residential greenness with incident composite microvascular complications and diabetic nephropathy were significantly stronger in participants with healthy sedentary behavior and sleep duration (all P_interaction < 0.05), while WC and sleep duration could modify residential greenness-HF association (all P_interaction < 0.05).

Table 4.

Hazard ratios and 95% confidence intervals for the associations between residential greenness and macrovascular and microvascular complications risk among type 2 diabetes patients, stratified by quartiles of healthy lifestyle score.

	Healthy lifestyle score quartile				Pinteraction
	Q1, score <1.23	Q2, 1.23 ≤ score <3.74	Q3, 3.74 ≤ score <5.86	Q4, score ≥5.86
Composite macrovascular complications	0.980 (0.905, 1.060)	0.982 (0.903, 1.068)	0.864 (0.783, 0.952)	0.861 (0.770, 0.963)	0.073
Stroke	0.936 (0.801, 1.093)	0.897 (0.737, 1.091)	0.902 (0.741, 1.098)	0.734 (0.565, 0.955)	0.432
HF	1.006 (0.869, 1.164)	0.877 (0.733, 1.050)	0.855 (0.708, 1.034)	0.751 (0.602, 0.938)	0.033
CAD	0.976 (0.888, 1.072)	0.957 (0.861, 1.063)	0.849 (0.749, 0.963)	0.859 (0.765, 0.964)	0.062
PAD	0.955 (0.806, 1.131)	0.911 (0.738, 1.125)	1.013 (0.805, 1.275)	0.646 (0.475, 0.879)	0.541
Composite microvascular complications	1.002 (0.917, 1.095)	0.988 (0.889, 1.098)	0.822 (0.750, 0.900)	0.829 (0.730, 0.940)	0.001
Diabetic retinopathy	1.020 (0.875, 1.188)	0.921 (0.806, 1.053)	0.930 (0.791, 1.094)	0.807 (0.688, 0.947)	0.024
Diabetic neuropathy	0.933 (0.778, 1.119)	0.803 (0.660, 0.977)	0.885 (0.695, 1.127)	0.750 (0.568, 0.991)	0.477
Diabetic nephropathy	0.996 (0.889, 1.117)	0.979 (0.861, 1.113)	0.763 (0.667, 0.872)	0.789 (0.661, 0.941)	0.006

Model as defined in Model 3 of Table 2.

As shown in Fig. 2, compared with participants with low residential greenness exposure and low healthy lifestyle scores, those with high residential greenness exposure and high healthy lifestyle scores had lower risks of composite macrovascular and microvascular complications, with HRs (95% CIs) of 0.532 (0.428, 0.661) and 0.518 (0.401, 0.668), respectively. Similar joint effects on the risks of stroke, HF, CAD, PAD, diabetic neuropathy, and diabetic nephropathy were also identified (Tables S27 and S28).

Fig. 2.

Joint associations of residential greenness and healthy lifestyle with the risks of composite macrovascular complications (A) and microvascular complications (B) among patients with type 2 diabetes. Model as defined in Model 3 of Table 2. ∗P < 0.05.

4. Discussion

This is the first study based on a large-scale cohort to identify the relationships between residential greenness exposure and the incidence of multiple diabetic complications in individuals with T2D. Our results showed that increased residential greenness levels were significantly associated with decreased risks of composite macrovascular complications, stroke, HF, CAD, composite microvascular complications, diabetic neuropathy, and diabetic nephropathy. PM_2.5, NO₂, and physical activity could mediate these associations. Healthy lifestyle scores modified the relationships between residential greenness and the risks of HF, composite microvascular complications, diabetic retinopathy, and diabetic nephropathy. Furthermore, residential greenness was not associated with incident diabetic complications among participants with poor lifestyle behaviors.

Currently, there is limited research focusing on the relationships between residential greenness and the risks of developing diabetic complications. A Chinese cross-sectional study used NDVI_300m as the residential greenness indicator and found that residential greenness was inversely associated with diabetic retinopathy prevalence [21]. In contrast, our large-scale cohort study did not find this association. The reasons for the inconsistency in results might be differences in study design, adjusted covariates, and ethnicity of the study population. Another study conducted in Spain found that for every 0.01 increase in NDVI_300m, the MI risk among participants with diabetes was reduced by 6% [19]. After adjusting for multiple covariates, our study also linked residential greenness with decreased risks of macrovascular complications, including stroke, CAD, and HF, demonstrating the multifaceted health benefits of exposure to the natural environment.

Research has shown that controlling blood glucose levels in patients with T2D significantly reduces the risks of multiple diabetic complications [37,38]. Higher NDVI has been linked with lower blood glucose levels and lower HbA_1c, which might be potential mechanisms underlying the connections between residential greenness and incident diabetic complications [11,39]. Residential greenness was also found to be negatively associated with perceived psychological stress [40], a key factor contributing to poor blood glucose control and the development of complications in patients with diabetes [18,41]. Moreover, significant associations between residential greenness and biomarkers of inflammation and oxidative stress were identified [42,43], which were both involved in the development of diabetic complications [4].

While stringent glycemic control remains the primary approach for preventing diabetic complications [37,38], our study identifies residential greenness as a potential complementary environmental factor. Importantly, we found that the effects of residential greenness were stronger among participants with optimal glycemic control at baseline, although the interactions were not statistically significant. This suggests that residential greenness functions as an independent and complementary factor, which remains beneficial even for patients who have already achieved optimal glycemic control. Therefore, enhancing residential greenness presents a potential strategy to mitigate diabetic complication risks associated with environmental factors, providing a complementary approach to conventional pharmacological and lifestyle management [44,45].

Our study also identified that high residential greenness exposure corresponded to PFs of 6.39% to 9.25% for diabetic complications, which might provide more information on public health significance. With an estimated 508 million T2D cases in 2021 and a projection to 1.27 billion by 2050 [1], and given the high prevalence of complications (e.g., approximately 27% for macrovascular and 50% for microvascular) [46], our findings indicate that residential greenness holds significant potential for reducing the burden of diabetic complications. This potential benefit is especially urgent given the backdrop of global urbanization, which is progressively diminishing human exposure to residential greenness [[47], [48], [49]]. Therefore, given the potential causal relationships (caution is warranted), our findings highlight the need for proactive measures to preserve and enhance residential greenness to counter this trend and mitigate the future impacts of preventable diabetic complications.

We also found that the relationships between residential greenness and diabetic complications were partly mediated by air pollution. This finding aligns with previous research that has documented the effects of residential greenness on other adverse health outcomes, indicating that residential greenness may reduce air pollutant concentrations through the filtering effect of vegetation [50]. This holds particular importance given that air pollution continues to be a leading factor in the global disease burden [[51], [52], [53]]. Existing evidence shows that the mechanisms underlying the relationships between air pollution and incident macrovascular and microvascular complications include oxidative stress, endothelial dysfunction, systemic inflammation, and endocrine disruption [54,55], which partially explain the association between residential greenness and diabetic complications. Physical activity could also mediate the residential greenness-diabetic complications associations. Residential greenness provides residents with more opportunities to engage in physical activity, which is important for managing blood glucose levels and preventing diabetic macrovascular and microvascular complications [22,56].

Currently, more and more studies are focusing on the impacts of modifiable lifestyle factors on health. Our study included eight lifestyle behaviors and constructed a weighted healthy lifestyle score. We found significant associations between higher healthy lifestyle scores and lower incidence of diabetic complications, and further identified that both individual healthy lifestyle factors and the healthy lifestyle scores could modify the associations between residential greenness and some diabetic complication risks. Healthy lifestyles may work synergistically with residential greenness by exerting anti-inflammatory, oxidative-stress-reducing, metabolic-improving, and vascular-protective effects, thereby collectively reducing the risks of diabetic complications [[57], [58], [59]]. In addition, among participants with the lowest healthy lifestyle scores, the protective effects of residential greenness on the risks of developing all diabetic complications were very limited. This emphasizes the importance of combining residential greenness level improvements with lifestyle interventions to reduce the risks of diabetic complications.

Our findings highlight residential greenness as a modifiable environmental factor for reducing diabetic complication risks. Urban policies should prioritize residential greenness expansion in high-risk areas, while healthcare providers could incorporate environmental assessments into patient care, advising increased residential greenness exposure alongside lifestyle modifications. The synergistic benefits of residential greenness and healthy behaviors further support integrated public health strategies targeting both environmental and individual-level risk factors. These measures may reduce complication risks and healthcare burdens, offering scalable solutions for diabetes management.

Our study had the major advantage of drawing on a large-scale cohort to determine, for the first time, the relationships between residential greenness and incident diabetic complications. This study also determined the effect modification of healthy lifestyle factors on the relationships between residential greenness levels and diabetic complication risks. In addition, we conducted several sensitivity analyses to assess the robustness of our primary findings. Nevertheless, several limitations also existed in this study. First, although residential greenness exposures were assessed using multi-scale NDVI (300-m, 500-m, and 1000-m buffer zones), this approach does not capture the quality, type, or actual accessibility of residential greenness, nor does it account for individual mobility patterns. This likely resulted in non-differential misclassification, which would tend to underestimate the true effects. Future research should incorporate multidimensional indicators to comprehensively evaluate the health effects of residential greenness. Second, collecting lifestyle information based on questionnaires inevitably leads to reporting bias. Besides, as lifestyle information was collected only at baseline, we could not account for possible changes over time. Future studies with repeated, objective measures of behavior are needed to accurately quantify how lifestyle modifies the residential greenness-diabetic complications relationship. Third, although multiple covariates were adjusted in our study, the influence of unmeasured or unadjusted confounders on the results cannot be completely ruled out. For example, detailed information on diabetes treatment (e.g., medication adherence, duration, and dosage) was not available, which may have led to residual confounding. Nevertheless, the high E-values in Table S22 indicated the robustness of our results against residual confounders. Lastly, since our study population was predominantly White Europeans, the findings may not be directly applicable to other populations. Future studies in diverse ethnic and cultural settings are essential to validate the universality of these associations.

5. Conclusions

Overall, this study identified significant relationships between residential greenness exposure and decreased risks of diabetic complications in individuals with T2D, and found that healthy lifestyles might modify some of these associations. Among individuals with unhealthy lifestyles, residential greenness was not associated with incident diabetic complications. Our results highlight that simultaneously promoting the accessibility of residential greenness and encouraging healthy lifestyles may help to alleviate the burden of diabetic complications and improve the quality of life for individuals with T2D. Future research could incorporate the measurements of biological mediators such as inflammation and oxidative stress to elucidate the underlying mechanisms linking residential greenness to diabetic complications, thereby strengthening the plausibility of these associations.

CRediT authorship contribution statement

Ning Chen: Writing – original draft, Formal analysis, Data curation, Conceptualization. Feipeng Cui: Formal analysis, Data curation. Yudiyang Ma: Formal analysis, Data curation. Jianing Wang: Formal analysis, Data curation. Linxi Tang: Formal analysis, Data curation. Lei Zheng: Formal analysis, Data curation. Meiqi Xing: Formal analysis, Data curation. Xinru Zhao: Formal analysis, Data curation. Yaohua Tian: Writing – review & editing, Supervision, Conceptualization.

Conflict of competing interest

The authors declare they have nothing to disclose.

Appendix A. Supplementary data

The following is/are the supplementary data to this article:

Data availability

The data for this article can be accessed through the UK Biobank (www.ukbiobank.ac.uk) upon request.