13.02.2013 Views

Air Quality Criteria for Lead Volume II of II - (NEPIS)(EPA) - US ...

Air Quality Criteria for Lead Volume II of II - (NEPIS)(EPA) - US ...

Air Quality Criteria for Lead Volume II of II - (NEPIS)(EPA) - US ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

AX6-152<br />

Table AX6-5.1 (cont’d). Effects <strong>of</strong> <strong>Lead</strong> on Blood Pressure and Hypertension<br />

Reference, Study<br />

Location, and<br />

Period Study Description Pb Measurement Findings, Interpretation<br />

Europe (cont’d)<br />

Staessen et al.<br />

(1996a) (cont’d)<br />

Multiple regression models were used to test the<br />

association between natural log trans<strong>for</strong>med<br />

blood Pb (mean <strong>of</strong> baseline and follow up Pb)<br />

and blood pressure (systolic and diastolic),<br />

stratified by sex, then further stratified by use <strong>of</strong><br />

antihypertensive medications in men and<br />

menopausal status in women. Age and agesquared<br />

(calculated in quintiles) were <strong>for</strong>ced into<br />

the models, then remaining covariates were<br />

stepwise added to the model. Though not<br />

explicitly stated, natural log blood Pb (mean <strong>of</strong><br />

baseline and follow up) was likely <strong>for</strong>ced in last.<br />

Other candidate covariates were BMI,<br />

hemoglobin or hematocrit, serum gammaglutamyltransferase<br />

activity (an index <strong>of</strong> alcohol<br />

use) and serum calcium, 24 h urinary sodium and<br />

potassium excretion, energy expenditure,<br />

exposure to heavy metals (at the workplace),<br />

social class, smoking and drinking habits,<br />

menstrual status in women, and use <strong>of</strong><br />

antihypertensive medications, oral contraceptives,<br />

and hormone replacement therapy. In ambulatory<br />

blood pressure models, differences between<br />

baseline and follow up blood pressure models<br />

were constructed in the same way. For the<br />

difference models “concurrent variations in blood<br />

Pb concentrations” were used, presumably<br />

difference in baseline and blood Pb.<br />

The study does not use the full power <strong>of</strong> repeated measurements in the<br />

analyses. For problems encountered when collapsing repeated<br />

measurements to difference measures, see Møller (1992) above. Stepwise<br />

regressions are prone to capitalizing on chance results due to multiple<br />

testing <strong>of</strong> the same data and almost always produce a different mix <strong>of</strong><br />

covariates when they are stratified. Thus, it was puzzling to find that where<br />

in<strong>for</strong>mation on the effects <strong>of</strong> stepwise covariate addition to models was<br />

available in this article, that the same covariates were listed <strong>for</strong> both models<br />

based on the stratification variable. There is excessive reliance on<br />

fractionation <strong>of</strong> the data set due to multiple stratification, sometimes<br />

reducing the number <strong>of</strong> subjects in a model to as few as 171. Even the<br />

models using the most subjects had only 359 subjects. Low power to detect<br />

significant effects cautions against any interpretation <strong>of</strong> non-significant<br />

results. The time-integrated model used 10 baseline blood pressure<br />

measurements and 5 follow up blood pressure measurements, thus<br />

weighting the avg toward baseline blood pressure. The entry <strong>of</strong> the<br />

biochemical correlate <strong>of</strong> alcohol use in most <strong>of</strong> the models suggests that Pb<br />

effects and Pb-containing alcohol effects on blood pressure were confused,<br />

especially given the European setting and the time period during which the<br />

study was conducted. Control <strong>for</strong> use <strong>of</strong> hypertensive medication rarely<br />

entered models and partial control <strong>for</strong> this variable was achieved only by<br />

stratified analyses, further reducing power to detect significant effects in the<br />

remaining subgroup. No justification was given <strong>for</strong> stratified analyses.<br />

Incomplete in<strong>for</strong>mation in statistical methods and results complicates<br />

interpretation. It was uncertain if stepwise regression was used <strong>for</strong> logistic<br />

models. No comparisons were per<strong>for</strong>med to assess possible bias due to<br />

subject attrition over the course <strong>of</strong> the study.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!