A0503
Title: Bayesian federated inference for estimating statistical models based on non-shared multicenter data sets
Authors: Marianne Jonker - Radboud university medical center (Netherlands) [presenting]
Hassan Pazira - Radboud University Medical Center (Netherlands)
Emanuele Massa - Radboud University (Netherlands)
Ton Coolen - Radboud University (Netherlands)
Abstract: Identifying predictive factors via multivariable statistical analysis is often impossible for rare diseases because the available data sets are too small. Combining data from different medical centers into a single (larger) database would alleviate this problem but it is, in practice, challenging due to regulatory and logistic problems. A Bayesian federated inference (BFI) framework is proposed. It aims to construct from local inferences in separate data centers what would have been inferred had the data sets been merged. It can cope with small data sets by inferring locally not only the optimal parameter values but also additional features of the posterior parameter distribution. The BFI methodology has been developed for generalized linear models and survival models, for homogeneous and heterogeneous populations, and for association and prediction models. The performance of the proposed methodology is quantified using simulated and real-life data.
