Abstract

Background: It has been speculated on theoretical grounds that biomarkers are superior surrogates for chemical exposures to air samples in epidemiology studies.

Methods and Results: Biomarkers were classified according to their position in the exposure-disease continuum—that is, parent compound, reactive intermediate, stable metabolite, macromolecular adduct, or measure of cellular damage. Because airborne exposures and these different biomarkers are time series that vary within and between persons in a population, they are all prone to measurement error effects when used as surrogates for true chemical exposures. It was shown that the attenuation bias in the estimated slope characterising a log exposure-log disease relation should decrease as the within- to between-person variance ratio of a given set of air or biomarker measurements decreases. To gauge the magnitudes of these variance ratios, a database of 12 077 repeated observations was constructed from 127 datasets, including air and biological measurements from either occupational or environmental settings. The within- and between-person variance components (in log scale, after controlling for fixed effects of time) and the corresponding variance ratios for each set of air and biomarker measurements were estimated. It was shown that estimated variance ratios of biomarkers decreased in the order short term (residence time ⩽2 days) > intermediate term (2 days < residence time ⩽2 months) > long term biomarkers (residence time >2 months). Overall, biomarkers had smaller variance ratios than air measurements, particularly in environmental settings. This suggests that a typical biomarker would provide a less biasing surrogate for exposure than would a typical air measurement.

Conclusion: Epidemiologists are encouraged to consider the magnitudes of variance ratios, along with other factors related to practicality and cost, in choosing among candidate surrogate measures of exposure.