An Assessment of Bayesian Model-Averaged Logistic Regression for Intensive-Care Prognosis

Abstract

SUMMARYLogistic regression is the standard method for developing prognostic models for intensive care, but this approach does not take into account the uncertainty in the model selected and the uncertainty in its regression coefficients. This weakness can be addressed by adopting a Bayesian model-averaged approach to logistic regression; however, with respect to the dataset used for our study, we found maximum likelihood to be as effective as the more elaborate Bayesian approach, and an implementation of model averaging did not improve performance. Nevertheless, the Bayesian approach has the theoretical advantage that it can exploit prior knowledge about regression coefficient and model probabilities.