Guidelines for assessing Pharmacokinetic Bias in Epidemiology

Melvin E. Andersen

Biomarkers of exposure can be measured at lower and lower levels due to advances in analytical chemistry. Using these highly sensitive methods, some epidemiology studies report associations between plasma and urinary biomarkers and health outcomes at biomarker levels much below those associated with effects in animal studies. In some cases, these associations may arise from pharmacokinetic (PK) bias, i.e., a situation where a confounding factor or the health outcome itself alters pharmacokinetic processes affecting biomarker levels.  Quantitative assessment of PK bias combines PK or physiologically based pharmacokinetic (PBPK) modeling and statistical methods describing outcomes across large numbers of individuals in simulated populations.  We recently published a manuscript in Environmental Research involving a collaborative team of toxicologists and epidemiologists that provided an overview of PK bias examined in studies with both persistent chemicals, such as perfluoroalkyl substances (PFAS), and short-lived compounds, such as phthalates.  We proposed a set of steps to consider in deciding whether PK bias is likely to play a role in low exposure associations between chemicals and health effects. These steps include guidance for developing quantitative PK modeling for simulated populations to assess the possibility of PK bias more quantitatively.  

Table 1. Steps in assessing the likelihood of pharmacokinetic bias  

Considerations in assessing the possibility of PK Bias to explain observed associations Steps to identify and quantify PK Bias 
1. Are the associations of health outcomes and biomarker levels qualitatively consistent with observations from toxicology studies? Ask if the human outcomes are similar to those seen or expected based on knowledge of modes of action (MOAs) noted in animals. 
2. Is the level of observed biomarker similar to internal exposures/doses causing responses in toxicology studies? Calculate a measure of dose-related coherence and assess the departure from a coherent interspecies relationship, i.e.,1, 2, 3, or more orders of magnitude. 
3. Were associations noted for other compounds with similar PK characteristics? Examine whether there were other compounds evaluated where similar results would be expected if there was a PK bias in epidemiologic studies. 
4. In considering possible PK bias, what are the key determinants of the chemical’s PK behaviors and life stage differences showing the association? Develop a hypothesis for how key PK determinants would be affected by confounders of the health outcome and influence the observed association. 
5. Would the change in biomarker levels expected from alterations in PK behavior account for the magnitude of association seen in the study?  Implement the hypothesis quantitatively and use PK modeling approaches to simulate a population to compare with epidemiologic data.  

As defined in point 3 (Table 1), an interesting aspect of these assessments is the possibility of extending from association of an outcome seen with a compound with certain PK characteristics to other compounds with similar PK properties.  The PK bias analyses we described used a variety of PK models and emphasized the PK bias analyses are best performed by teams of epidemiologists, toxicologists, and quantitative modelers working to assess coherence between biomarker levels in both toxicity studies and human populations and then implementing quantitative analyses of causes of bias. Studies of the health effects of low levels of exposure with many compounds will be improved by developing some confidence that PK bias did not play significant roles in the observed associations.  

Our team is skilled in PK and PBPK modeling and provides advice on value PK bias studies in particular cases. Email me @ mandersen@scitovation.com for more information.