“A New Approach to the Generalizability of Randomized Trials”

Causal effects often differ between populations. When causal knowledge is obtained in one setting (for example in the study population of a large randomized controlled trial), extrapolation will usually be necessary in order to justify application of that knowledge in a different setting (for example in a clinically relevant target population). Several different frameworks have been proposed in order to formalize the extrapolation problem. These frameworks have different implications for how researchers should reason about what differences between populations must be accounted for. We propose a new type of causal model in order to formalize the problem, and show how this approach differs from other frameworks including the graphical models proposed by Bareinboim and Pearl. In contrast to these graphical models, our framework can be used to reproduce certain recommendations in the Cochrane Handbook.

Lecturer: Anders Huitfeldt

Location: Zweigbibliothek Campus Charité Mitte – Medizinische Bibliothek der Charité- Just above the CCM medical library. Address: Philippstr. 11, 10115 Berlin.