Personalized Medicine intends to provide to each patient and his/her disease the best fitting treatment. Biomarkers are used to inform the individual treatment choice. The analysis of a biomarker treatment interaction in a regression model is the formal process to derive a treatment decision rule. Alternatively, the potential outcome (PO) paradigm is a more natural approach to derive a multivariate distribution of counterfactual outcomes under several treatments. The talk generalizes the concept of a bivariate Gaussian PO-model using copula theory. It also discusses aspects of model diagnostics.  Finally the marginal approach using interaction terms in regression models is compared to the PO strategy. Implications for study designs will be dscussed.

Einladung von Prof. Werner Brannath