gms | German Medical Science

When data are sparse, extrapolation is a promising approach to utilizing related external information in an analysis [1]. On the technical side, an obvious way of considering external evidence is via the use of informative prior distributions. Care must however be taken to avoid overconfidence in results derived from "naive" pooling, and the possibility of a prior-data conflict should be anticipated.In the context of meta-analysis, one is quite commonly faced with a small number of studies, while potentially relevant and useful additional information may also be available. We describe a simple extrapolation strategy based on heavy-tailed mixture priors [2] for effect estimation in a meta-analysis. The model setup is easily interpretable and leads to robust inference. We illustrate the method using examples of extrapolation from adults to children, and utilizing the "bayesmeta" R package [3].

Die Autoren geben an, dass kein Interessenkonflikt besteht.

Die Autoren geben an, dass kein Ethikvotum erforderlich ist.