Study population
All participants were from the Henan rural cohort study (Registration number: ChiCTR-OOC-15006699), established in five rural areas of Henan Province, China during the period from 2015 to 2017. Its main objective was to examine the burden of multiple chronic diseases in the population, explore potential risk factors, clarify trends in non-communicable diseases, and develop disease risk scoring models suitable for Chinese people. Detailed descriptions of the cohort design and study population were previously published elsewhere [23, 24].
A total number of 39,259 participants in the Henan Rural Cohort Study were recruited, and each participant signed an informed consent form. After excluding participants who did not accept blood glucose measurement and those who lacked other vital data, 39,019 participants were available for analyses actually.
Residential green and blue spaces assessments
Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) are employed to characterize the residential green space. These two vegetation indices are derived from satellite images, and obtained by the Moderate Resolution Imaging Spectroradiometer (MODIS) [25]. NDVI is a widely used index describing the relative overall vegetation density and quality, which is calculated based on land surface reflectance of near-infrared (NIR) and visible red (RED) wavelengths. EVI is similar to NDVI, but a blue light band (BLUE) is introduced to correct the atmospheric and soil background [26]. The values of both vegetation indices ranged from -1 to 1, with negative values representing water, zero value denoting bare soil, and higher, positive values indicating dense green vegetation. We calculated the average green exposure of each participant in the 500 m buffer zone around their residential address within three years before inclusion in the cohort.
We implemented the Soil and Water Assessment Tool (ArcSWAT-2012) in ArcGIS (10.2) to calculate the distance between the participant’s house and the nearest water body that represented the residential blue space. Using the remote sensing monitoring data provided by the Resource and Environmental Science Data Center of the Chinese Academy of Sciences (http://www.resdc.cn), we developed a land-use map of Henan Province (Figure 1). Then, we divided land use types into six types based on the Land Use and Cover Change (LUCC) classification system: forestland, grassland, cultivated land, water area, artificial surface, and unused land.
Outcome assessments
Following the Clinical Chemistry Blood Sample Collection and Processing guidelines [27], we collected blood samples from participants who had fasted for at least eight hours. The samples were immediately sent to the laboratory for serum and plasma separation. A Roche Cobas c501 (Switzerland) analyzer was utilized for testing of fresh serum specimens to determine the fasting blood glucose levels of the participants.
Following the recommendations of the International Classification of Diabetes Mellitus Diagnostics [28], we diagnosed T2DM in all subjects that met any of the following three criteria: (1) Defects in insulin secretion or action of the participants were not caused by type 1 diabetes mellitus and gestational diabetes mellitus; (2) The patient had been diagnosed with T2DM or had been taking antidiabetic drugs directed by doctors; (3) The fasting blood glucose level exceeded 7.0 mmol/L.
Covariates
Based on the evidence in the existing literature, we selected some variables as covariates [29]. Demographic covariates included age (years) and sex (male, female). Additionally, health status covariates were assessed, including body mass index (BMI) (kg/m2) and family history of diabetes mellitus (no, yes). The following socioeconomic covariates were examined: marital status (married/living together, divorced/widowed/separated/unmarried), education level (no school or primary school, middle school, junior college or higher), and monthly income level (low, medium, or high). We also tested the following health behavior covariates: smoking status (never, former, or current), drinking status (never, former, or current), high-fat diet (average consumptions of meat from livestock and poultry by each participant of more than 75 g per day) (no, yes), fruit and vegetable intake (average intake of fruit and vegetable by each participant of more than 500 g per day) (no, yes), and physical activity (low, medium, or high) [30].
Statistical analyses
After cleaning up the cohort's baseline data and environmental exposure data, we described age, sex, fasting blood glucose level, residential green space and other characteristics of all included participants. We explored the relationship between the residential environment and fasting blood glucose levels through scatter plots and performed a natural logarithmic conversion of the fasting glucose levels to improve the normality before the regression analysis. Mixed effect models were implemented to elucidate the effects of residential environment on T2DM and fasting blood glucose levels. Three types of models (Crude model, Model 1 and Model 2) were performed in our analyses. A minimal set of potential confounders was adjusted in model 1: age, sex, BMI, monthly income, and physical activity. Model 2 was further adjusted for education level, marital status, smoking, drinking, high-fat diet, vegetables and fruit intake, and family history of diabetes. Then, we analyzed the interaction effects through adding confounding factors as cross-product terms. The analysis results were presented as OR value and percentage change, respectively.
To examine the robustness of our results, we conducted a series of sensitivity analyses by removing subjects with a family history of diabetes, hypertension, or dyslipidemias. Different green exposure time values (e.g., 1-, 2-, and 4-year average values) were replaced to test the sensitivity of the results. All statistical analyses were completed using R 3.6.1 (https://rstudio.com).