Association of Long-term Exposure to Ambient Air Pollutants With Risk Factors for Cardiovascular Disease in China

Key Points Question Which cardiometabolic risk factors are associated with ambient air pollution in China? Findings In this population-based cross-sectional study including 15 477 Chinese adults, the strongest associations for exposure to air pollution were detected for individuals with hyperbetalipoproteinemia and the weakest associations for those with overweight or obesity. The metabolic risk factors may have exacerbated the associations of air pollutants with the prevalence of cardiovascular disease. Meaning These findings may help researchers, physicians, and policy makers evaluate the hazardous effects of air pollutants more completely and design targeted strategies for primary prevention of cardiovascular disease.

The operation of the monitoring stations strictly followed quality assurance/quality control (QA/QC) procedures set by the State Environmental Protection Administration of China (SEPAC,1992). The environmental monitoring centers in each of the three study cities conducted regular performance audits and precision checks on the airmonitoring equipment. Quarterly performance audits are conducted to assess data accuracy on PM10, SO2, NO2, and O3 monitoring systems.

1) The calculation method
The calculation method is performed according to Chinese National standards (GB8170-87). The unit of monitored pollutants is mg/m 3 accurate to the third decimal. The units can also be expressed as μg/m 3 , depending on the pollutant's concentration. For concentrations that were too low to be measured, half of the lowest checking limit of the equipment will be used as the measured value.
2) Outliers When the measured concentration is too low (e.g. background value), a negative value can be obtained because of the zero drift of the monitor. There is no physical meaning to this value. This negative value can be regarded as a value of "unable to measure." For the monitoring station with an automatic calibration system, if equipment zero drift/span drift exceeds the control range during the period of zero/span calibration, the data from the time it becomes out of control until the equipment is recovered should be regarded as invalid data. The data cannot be used statistically.
The data during the period of zero calibration/span calibration should be regarded as invalid data. It cannot be used statistically, but a flag should be made on these data and the records stored as evidence.
When values are missing because of a loss of power, any data received by the central control station during the period of the loss of power should be regarded as invalid data. The period of loss of power should be counted at the start of power outage until complete warm-up of equipment. The data cannot be used statistically.
Because pollutant concentrations change over time and change slowly, there should be no swift change in pollutant concentration in the results of normal monitoring. Either a swift change or no change indicates that there is an equipment problem. The problem should be identified, and the data between the start of problem to recovery should be regarded as outliers. These data cannot be used statistically.

3) Statistics of monitoring data
One time value The central control station uses an average of 15 minutes of pollutant concentrations measured at the branch station as a one-time value. The central control modifies this value and judge whether this value is an outlier using the report software.

eMethods 3. Detailed Information on the 2-Level Binary Logistic Regression Model
At the participant level, we predicted the logit of the prevalence of a cardiovascular disease (CVD) or its risk factors as a function of k covariates (X1 ….Xk) as follows: logit [Probability (Yij)] = αj + β1X1ij + ….+ βkX1ij + eij (1) The variable (Y) in equation 1 is prevalence of CVD or its cardiometabolic risk factor, the subscript j is for study districts (j=1,…, 11) or communities (j=1,…,33), the subscript i is for participants (i=1,..nj), αj are intercepts at the district or community level, β1 …βk are regression coefficients for covariates, and eij are the random errors, assumed to have means of zero and constant variance. The αj are random coefficients because they are assumed to vary across districts or communities.
At the district or community level, we regressed the district-specific or community-specific intercepts αj on the district-specific or community-specific pollutant level (Zj) to explain variations of αj, as follows: Equation 2 predicts CVD or its risk factor prevalence in a district or community by air pollutant concentration Zj. If γ1 is positive then districts or communities with higher pollutant levels have a higher prevalence of CVD or its risk factor (adjusting for covariates). Conversely, if γ1 is negative, then the prevalence is lower in districts or communities with a higher pollutant level (adjusting for covariates). The u-terms uj are random errors at the district or community level, assumed to be independent and have mean of zero and constant variance. These random errors characterize the variation between districts or communities and are assumed to be independent from eij at the participant level. Note that α, β1,…, βk, and γ1 are not assumed to vary across districts or communities. Therefore, they have no subscript j to indicate to which district or community they belong; they are referred to as fixed effects given that they apply to all districts or communities.
Substituting equation 2 into equation 1 yields a single regression equation: The terms in the first and second parentheses in equation 3 are often respectively called the fixed (ordeterministic) and random (or stochastic) parts of the model.

eFigure 1. Associations Between Air Pollution and Cardiometabolic Risk Factors Stratified by Sex
A, for particles with aerodynamic diameter ≤1.0 µm (PM1); B, for particles with aerodynamic diameter ≤2.5 µm (PM2.5); C, for particles with aerodynamic diameter ≤10 µm (PM10); D, for sulfur dioxide (SO2); E, for nitrogen dioxide (NO2); F, for ozone (O3). The effect estimates (odds ratios and 95% confidence intervals) were scaled to 10 μg/m 3 in air pollutants and were adjusted by age, smoking status, alcohol consumption, household income, controlled diet of low calories and low fat, sugar-sweetened soft drink intake, exercise, career, education, gross domestic product, greenness level, family history of cardiovascular disease, and residuals from regression model of highly correlated air pollutants. (*) indicates P<0.05 for the interaction.

eFigure 3. Associations Between Air Pollution and Cardiomebolic Risk Factors Stratified by Family History of Cardiovascular Disease (CVD)
A, for particles with aerodynamic diameter ≤1.0 µm (PM1); B, for particles with aerodynamic diameter ≤2.5 µm (PM2.5); C, for particles with aerodynamic diameter ≤10 µm (PM10); D, for sulfur dioxide (SO2); E, for nitrogen dioxide (NO2); F, for ozone (O3). The effect estimates (odds ratios and 95% confidence intervals) were scaled to 10 μg/m 3 in air pollutants and were adjusted by age, gender, smoking status, alcohol consumption, household income, controlled diet of low calories and low fat, sugarsweetened soft drink intake, exercise, career, education, gross domestic product, greenness level, and residuals from regression model of highly correlated air pollutants. (*) indicates P<0.05 for the interaction.