Capítulo 1: Relevancia Económica y social de la Contratación Pública
II. Relevancia económica y social de la contratación pública
The unique features of the analysis of LDL and HDL cholesterol in relation to exposure to PM10, NO2 and SO2 are described here. The principle difference in relation to the methods is that LDL and HDL, are not particularly labile characteristics and are considered to be determined by influences over time culminating in their status as a risk factor for future CVD events and are tied to other risk factors characterized as metabolic syndrome. We therefore consider cumulative exposure over time as the relevant time frame to evaluate the effects of PM exposures on the lipid profile. The two principle determinants of the lipid profile are LDL and HDL, where high levels of LDL and low levels of HDL confer greater risk.
Outcome Variables:
LDL –Cholesterol calculated from measured total cholesterol and fasting triglycerides using the equation developed by Friedewald, Levy and Fredrickson in mg/dl
Total Cholesterol - high density cholesterol - triglyceride/5. in mg/dl.158
Total Cholesterol (Hitachi 704Analyzer/Boehringer-Mannheim Diagnostics) in mg/dl Triglycerides. (Available only on 4yrs of age and older). (Hitachi 704
Analyzer/Boehringer-Mannheim Diagnostics) in mg/dl
Exposure variables:
I create a partitioned pollution exposure parameter for each of the prior 1 year average pollutant measurements. As an example, the county and within county parameter for PM (measured by PM10, NO2, and SO2) was calculated as follows: PM j = jth county mean of prior 1-year PM- grand mean of prior 1-year PM
PM ij =ith individual in county j prior 1-year PM – jth county mean prior 1-year PM
The resulting county level parameter represents the deviation of the county mean of the prior 1 year exposure to PM10, from the mean prior 1 year exposure across all counties. The resulting within county level parameters represents the individual’s deviation from the county mean of the prior 1-year PM10 measurements.
1. Statistical Analysis
Mixed models as described above were fit with LDL and HDL as the dependent outcomes. Each model was fit alternately with each partitioned pollution parameter in the random intercept model. The results of these models constitute the principle statistical results with the resulting effect estimates for county and individual level being the basis of inference regarding the effects of PM.
2. Sensitivity Analysis
Due to the assumption of the mixed models that differences in the county effects are not due to differences in the distribution within the counties of characteristics causally associated with the outcome, I alternately fit models that included alternate specifications of the most deterministic characteristics of the outcome based on substantive knowledge of the lipid outcomes. For cholesterol, the base model included a linear and quadratic term for both age and BMI as well as an interaction term for sex with the linear and quadratic form of age. In the other models I employed alternate specifications related to age, sex and BMI that included:1) a model without the interaction between sex and the age variables;2) a model with an interaction between sex and the BMI variables (instead of age); 3) A piecewise linear parameterization of age with the cutpoint corresponding to 60, reflecting the
overrepresentation of people of this age group in the NHANES data; 4) The piecewise linear parameterization of age with an interaction between sex and the piecewise age parameters; 5) The most naïve specification of age and BMI, a single continuous variable for each, without any interactions.
It is also assumed that no important variables are left out of the model that would make the errors at the individual level and the county level not exchangeable. I therefore ran
models that adjusted for additional possible confounders including: 1) a variable (1=yes, 0=no) for whether the participant had lived at the same address for more than a year 2) The sum total of the number of times exercised in the last month; 3) education; 4) Household size 5 or more (index) vs. 4 or less (referent); 5) Use of wood stove in the past 12 months; 6) Use of fireplace in the past 12 months; 7) Use of gas stove in the past 12 months; 8) Number of times eating seafood as an indicator of omega-3 fatty acid intake; 9) Coffee drinking that is associated with cholesterol levels; 10) Use of hypertension drugs.
To facilitate comparison, the resulting parameter estimates for county and within county level pollutant estimates derived from each of the alternate model specifications were plotted with their 95% confidence interval.
3. Effect measure modification
In studies of acute effects of air pollution, people with diabetes and older people have been found to be at elevated risk for adverse effects of particle matter air pollution. I therefore evaluated if there was evidence that the joint effect from each air pollutant and alternately diabetes and age, was different from their independent effects. To do this, I included interaction terms in each pollutant-lipid model. As a means to evaluate if there was statistical evidence that diabetes, or age did in fact modify the risk from air pollution, I used a cutoff of p less than 0.20, understanding that these tests are underpowered. The interaction terms included the within-county air pollutant parameter.