DescriptionAnalysis of biomarkers to detect levels of chemical exposure in humans is an important risk evaluation tool. For example, urine biomarkers such as 1-aminopyrene can be used to assess exposure levels to diesel exhaust (DE), an important public health concern. Toxic chemicals contained in DE and DE particles have demonstrated genotoxic and carcinogenic properties in experimental animals. A recent experiment evaluating the urine concentration of DE biomarkers was the impetus for this dissertation. One goal of the experiment, and the focus of this dissertation, was to characterize the excretion time course of the biomarker 1-aminopyrene. The times of maximum concentration in plasma or maximum excretion in urine have been typically summarized using non-parametric or asymptotic techniques based on individual subject-level values; however, there is limited information addressing confidence interval generation when sparse subject-level samples requiring population-modeling approaches are present. Therefore, there was a need to generate and evaluate an appropriate confidence interval approach when sparse sampling is present.
Pharmacokinetic (PK) modeling was used to fit a standard one-compartment urine excretion model to the data for estimation of the time of maximum excretion. Several variations of the PK model were explored and a model based on cumulative excretion rates was selected. Several statistical techniques for modeling PK data and calculating confidence intervals for the time of maximum excretion were compared including confidence intervals based on the first and second order delta methods, derived for this dissertation.
A comparison of confidence interval methods showed that when using: (1) within-subject Tmax values, coverages obtained using the non-parametric method were highest and often provide coverages close to the nominal 95% level; and (2) population-average Tmax values, confidence intervals generated using the first-order delta method provided the highest coverages, at approximately 93% when numerical approximation estimation methods were used. Subject response profiles for the 1-aminopyrene biomarker data were varied and led to a hypothesis that a mixture of more than one distribution of profiles may be present. Future exploration with data collected for more than 24-hours would be needed to further explore this hypothesis fully.