Overadjustment bias and mediation analysis in PK/PD modeling

Description: Pharmacokinetic/pharmacodynamic (PK/PD) modelling benefits from the inclusion of covariates to enhance predictive capabilities and explain variability. However, the use of postbaseline or time-varying covariates introduces the risk of overadjustment bias, a term used in epidemiology for a situation where bias is increased when including a covariate in the analysis.

Although overadjustment bias is rarely discussed within the context of PK/PD modelling, it is by no means less important for this setting. In scenarios affected by overadjustment bias, the inclusion of covariates in a PK/PD model result in under-estimated treatment effects or a concentration-effect relationship that is too shallow. One of the situations where inclusion of the covariate would cause overadjustment bias, is when the candidate covariate lies on the causal pathway from the treatment to the outcome.

In this situation, one might be interested in the level in which this candidate covariate mediates the treatment effect on the outcome, which can be estimated using a mediation analysis from epidemiology. High levels of mediation—while not sufficient to demonstrate causality—may support the use of the intermediate as a surrogate endpoint or bridging biomarker. This has relevance in drug development, when clinical endpoints are rare or necessitate prolonged follow-up, but can also support biomarker development for personalised medicine.

This individual talk will explain the epidemiological concepts of overadjustment bias and mediation analysis to a pharmacometrics audience, supported by illustrative simulations. An important part of the talk will be translating these underdiscussed concepts into practical advice for pharmacometricians.