Instrumental variable analyses of exposure – response relationships

Exposure-response, or pharmacokinetic-pharmacodynamic (PKPD), analyses provide quantitative information that can be used as a basis for decisions regarding dose, dosing schedule, and covariate-based or concentration-based individualization. It is typically applied without assessment of causality and investigation into how different origins of exposure variation (dose, covariates, random effects) can differentially relate to response. The risk of erroneous conclusions of standard PKPD analyses across a range of different causality scenarios will be presented in this talk. The randomized dose is a suitable variable for instrumental variable (IV) analysis, a method that can help identify causal effects. Two standard IV methods, adapted for exposure-response analysis, are predictor substitution and control function. An extended PKPD approach, the partitioned effect model, allows the estimation of separate exposure-response effects for IV, covariate and random effect relations. These methods can complement exposure-response analysis and provide an opportunity to better understand causality and improve the basis for dose selection. The partitioned effect model also allows an understanding of how different sources of exposure variability contribute to the exposure-response, aspects of relevance when designing studies and making decisions about dose individualization.