Data exploration in meta-analysis with smooth latent distributions

Abstract

Meta-analysis with discrete outcomes is interpreted as the estimation (in one or two dimensions) of a non-parametric smooth latent distribution of event probabilities (or rates). A simple but efficient EM algorithm is presented. A fine grid is used and fast smoothing is done by penalized least squares. Data exploration is the primary goal, but the estimated distribution can also be used to compute useful statistics of treatment effects. Copyright © 2007 John Wiley & Sons, Ltd.