An attenuation in the log transformed EPCB OR also was observed after adjustment for standard factors: The unadjusted OR was 1.65 (95% CI: 1.38, 1.99) per 1 SD increase in log PCBs, whereas the adjusted OR for the PCB diabetes association was 1.23 (95% CI: 0.88, 1.72).
This was apparent in both the quintile and the log transformed EPCB models (p < 0.05 for sex x PCB interaction terms).
Total wet weight levels of the 35 PCB congeners in nanograms per gram (parts per billion) were summed (EPCBs).
Unconditional logistic regression modeling compared participants with and without diabetes by quintiles of the EPCBs. Participants were excluded from multivariable analyses if they had prediabetes (n = 171) because many of their disease and demographic characteristics were intermediate to those observed in individuals with diabetes and those with normoglycemia.
In a separate linear regression model, age, total lipids, total years living in Anniston, and nonwhite race showed significant positive associations with EPCBs, whereas we observed a significant inverse association for education.