Study cohort
The Korean Urban Rural Elderly (KURE) study aimed to identify ways to prevent and effectively treat important chronic diseases among the older individuals in the Republic of Korea [14]. Through local advertisements, we recruited participants aged 65 years and older from northwest Seoul (urban areas: Eunpyung-gu, Mapo-gu, and Seodaemun-gu) and Incheon (rural area: Ganghwa). At baseline, the participants completed questionnaires (demographics, history of disease, and lifestyle behaviors), underwent anthropometric measurements (height, weight, and blood pressure), and blood tests. The follow-up survey was conducted between 2016 and 2019 at a 4-year interval from the baseline between 2012 and 2015. Exclusion criteria were (1) a self-reported history of diabetes mellitus, fasting blood glucose ≥ 126 mg/dL, or HbA1c ≥ 6.5% (48 mmol/mol) at baseline, (2) failure to undergo a follow-up survey, (3) unidentifiable residential address, and (4) missing values for glycemic indicators or covariates. This cohort study was approved by the Institutional Review Board of Yonsei University Health System, Severance Hospital (IRB No. 4-2012-0172/4-2022-0435) and adhered to the principles of the Declaration of Helsinki. We obtained informed consent for participation from all patients.
Air pollutants
The annual average concentrations of PM10, PM2.5, and NO2 were estimated at the residential addresses of participants using a validated exposure prediction model applied in previous cohort studies [15,16,17]. This nationwide prediction model was built in a universal kriging framework based on air quality regulatory monitoring data along with geographic predictors and spatial correlation. Geographic predictors were estimated by partial least squares from 320 geographic variables, including transportation, demographics, land cover, transportation facilities, emissions, greenness, and elevation. Model performance (cross-validation R2) for PM10, PM2.5, and NO2 in 2016 was 0.50, 0.37, and 0.81, respectively. We estimated air pollution concentrations for 1 year before the baseline and follow-up survey years (e.g., 2011 air pollution data used for the survey year of 2012). We also estimated concentrations for 5 years prior to the baseline survey years (e.g., 2011 − 2015 air pollution data used for the survey year of 2016). Because national air quality monitoring for PM2.5 began in 2015, PM2.5 concentrations in each survey year between 2012 and 2016 were replaced with 1-year concentrations in 2015.
Glycemic indicators
Glycemic indicators included fasting blood glucose, HbA1c, insulin, and homeostatic model assessment for insulin resistance (HOMA-IR) at baseline and follow-up. HOMA-IR was calculated using the equation: \(\:\left[Insulin\:\right(\mu\:U/mL)\:\times\:\:Fasting\:glucose\:(mg/dL\left)\right]/405\) [18].
Physical activity
The level of physical activity was calculated based on the metabolic equivalent of task (METs) and categorized as inactive, minimally active, and health-enhancing physical activity (HEPA) [19]. The HEPA group consisted of participants who engaged in vigorous activity of over 1,500 METs-min/week for at least 3 days or involved in any combination of walking, moderate-intensity, and vigorous activity for at least 3,000 METs-min/week for at least 7 days. The minimally active group included individuals engaging in at least 20 min of vigorous activity per day for 3 or more days, or at least 30 min of moderate-intensity activity or walking per day for 5 or more days, or involved in any combination of walking, moderate-intensity, and vigorous activity for at least 600 METs-min/week for 5 or more days. The remaining participants were considered inactive group.
Changes in physical activity levels were determined using baseline and follow-up data and were categorized in terms of (1) levels of physical activity maintenance and (2) the change in METs. The levels of physical activity maintenance consisted of the inactivity and moderate-to-vigorous groups. Due to the small number of individuals in the maintained HEPA group, we combined the maintained minimally active group with the maintained HEPA group into the moderate-to-vigorous group. The change in METs was categorized as decreased (METs difference < 0) and increased METs (METs difference > 0); here, we excluded individuals whose METs did not change (METs difference = 0).
Covariates
Demographics, socioeconomic factors, history of disease, lifestyle behaviors, blood pressure, and lipid profiles were considered covariates. Age (years), systolic blood pressure (SBP), diastolic blood pressure (DBP), as well as triglyceride, high-density lipoprotein (HDL) cholesterol, and low-density lipoprotein (LDL) cholesterol levels were included as continuous variables. LDL cholesterol levels were calculated using the Friedewald equation [20]. Categorical variables included sex (male or female), household income (quartile), physical activity (inactive, minimally active, or HEPA), smoking status (never, former, or current smoker), and current alcohol consumption status (none, monthly or less, or at least once a week).
Statistical analysis
Linear mixed-effects models were used to examine the longitudinal associations of the previous 1-year exposure to air pollution with glycemic indicators, considering each participant as a random effect. All covariates except sex were considered as time-varying variables. Glycemic indicators, body mass index, and lipid profiles were natural log-transformed owing to a skewed distribution. Model 1 was adjusted for age and sex. Model 2 was adjusted for Model 1 variables plus household income, SBP, DBP, body mass index, triglycerides, HDL cholesterol, LDL cholesterol, level of physical activity, smoking status, and current alcohol consumption status. All air pollutant concentrations were standardized. The glycemic indicator change was expressed as a percentage change per 1-standard deviation (SD) increase in each air pollutant and its 95% confidence interval (CI). Percentage changes were calculated using the formula: \(\:\left({exp}^{\beta\:}-1\right)\times\:100\). To perform a complete-case analysis, we excluded outcome variables, exposure data, and covariates from the analysis if any were missing. Selection of study participants is shown in Supplementary Material 1.
Additionally, we examined the associations of air pollution concentrations for 5 years prior to the baseline survey with glycemic indicators after IPTW [21]. This approach enabled us to improve causal inference by minimizing the impact of differences in baseline characteristics between participants living in high- and low-pollution areas. For IPTW, we estimated propensity scores by constructing multivariable logistic regression models, including all covariates except for physical activity. The propensity score was defined as the probability of being assigned to the higher or lower 5-year exposure group. The higher exposure group was participants with the 66 percentile or higher concentrations of PM10 (≥ 50.1 µg/m3), PM2.5 (≥ 23.9 µg/m3), and NO2 (≥ 32.5 ppb). The lower exposure group was those with lower than 33 percentiles of PM10 (< 47.3 µg/m3), PM2.5 (< 22.7 µg/m3), and NO2 (< 26.9 ppb). The numbers of these subsets were 633, 632, and 632 in the higher exposure groups and 612, 611, and 612 in the lower exposure groups for PM10, PM2.5, and NO2, respectively. Linear mixed-effects models were constructed by considering each matched pair as a random effect and simultaneously accounting for repeated measures within individuals. Furthermore, we included physical activity level as a time-varying covariate in the model.
Using the above lower and higher exposure subsets, we estimated the impact of physical activity level changes on the association between air pollution and glycemic indicators. We repeated the above linear mixed-effects model analyses with IPTW after stratification by changes in physical activity levels (levels of physical activity maintenance and the change in METs), including Model 2 covariates except for physical activity. Significant between-group differences were tested using the formula proposed by Altman and Bland [22].
Given the observed positive associations of PM10 and NO2 with insulin and HOMA-IR in the increased METs group, we conducted post-hoc analysis to explore non-linear relationships between the levels of METs increase and glycemic indicators at follow-up and compared the patterns between the lower and higher exposure groups. Participants with increased METs (METs difference between baseline and follow-up > 0) were only included in this post-hoc analysis. The non-linear relationships were estimated using a generalized additive model (GAM), including METs as a spline independent variable (degrees of freedom = 3) and each glycemic indicator as the dependent variable. The GAM was adjusted for the corresponding glycemic indicator at baseline and all covariates (as parametric variables) in Model 2 except physical activity.
All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA). Statistical significance was set at two-sided p < 0.05.
Add Comment