Background: Within-person variability in biomarkers results in random error that can attenuate estimates of association. Little information on such variability is available for a number of nutrition-related biomarkers.
Methods: Blood samples obtained 2 to 4 weeks apart were analyzed for tocopherols, carotenoids, ascorbate, lipids, cotinine, C-reactive protein, and oxidative stress. Subjects (n = 206 men and women, mean age 45.4 years) were either smokers or passively exposed to smoke. We calculated intraindividual and interindividual variability and the number of measurements required to reduce attenuation.
Results: For most biomarkers, 2 measurements would be required to limit the attenuation of correlation coefficients to no lower than 90% of the true correlation. If only one measurement were obtained, observed correlations would be approximately 80-88% of true correlations. For regression coefficients, 3 or 4 measures would be required. Exceptions were ascorbic acid and malondialdehyde, for which a single measure resulted in little attenuation.
Conclusions: For most serum markers, collection of 2 or more measurements per person is desirable to increase the ability to detect associations between biomarkers and health-related variables. If only one measure is possible, sample sizes should be planned to permit detection of associations that are likely to be observed, not the theoretical true associations. The results of this study, in which measurements were obtained 2 to 4 weeks apart, are relevant for epidemiologic research in which the exposure of interest is the subject's baseline or current status. It is likely that within-person variability would be greater over a period of months or years.