B0722
Title: Communicating likelihood profiles of single parameters of interest in complex models in federated research networks
Authors: Martijn Schuemie - Janssen Research and Development (United States) [presenting]
Abstract: Effects of medical intervention are increasingly studied in distributed research settings, using multiple clinical data sources such as electronic health records and administrative claims. Sharing individual patient data is seldom allowed, and instead, only summary statistics can be used to combine evidence across the network. Although the models in these studies can be complex, for example, Cox models conditioned on propensity score strata, or multi-variable Poisson models conditioned on individual patients, there is typically only a single parameter of interest, representing the effect of the exposure on the outcome. We present likelihood profiles as a generic approach to communicating information on this parameter between sites that is both privacy-preserving and efficient, requiring only a single round of communication. A likelihood profile is a simple representation of the likelihood of values of the parameter over a wide range, for example by sampling the likelihood on a grid and using linear interpolation. We demonstrate how this approach can be used in both fixed and (Bayesian) random-effects models, using both real and simulated data. Results show performance comparable to pooling data.
