All Times EDT
Keywords: marginal, conditional, estimands
ICH E9 R1 specified the importance of precisely quantifying the treatment effect for clinical trials – to inform patient choices and facilitate clinician’s evidence-based decision making. In the era of precision medicine, many targeted therapies are expected to derive differential treatment benefit based on patients’ biological signature. The PD-1/PD-L1 class in oncology is a prime example. The marginal and conditional estimands are both relevant as they describe distinct estimands – “the average treatment effect for a targeted population” and “the expected treatment effect for patients with a fixed set of baseline characteristics”. FDA’s draft guidance on covariate adjustment encourages judiciously use baseline covariates to enhance efficiency. Many traditional population-level summaries, despite undisputable usefulness in practice, have some undesirable properties, for example, non-collapsibility, not subgroup mixable. Issues, such as delayed treatment effect, further complicate the choice of effective estimands to enable effective decision making in drug development. In this talk, we would like to investigate methods which allow marginal effect estimation starting from covariate adjusted models and to compare benefit and risks of using multiple efficacy measures – the ultimate choice should still be driven by the clinical objective, and properties of the molecule.