The simultaneous clustering of observations and features ofdatasets (known as co-clustering) has recently emerged as a central topic inmachine learning applications. However, most models focus on continuousdata in stationary scenarios, where cluster assignments do not evolve overtime. We propose in this paper the dynamic latent block model (dLBM),which extends the classical binary latent block model, making amenable suchanalysis to dynamic cases where data are counts. Our approach operates ontemporal count matrices allowing to detect abrupt changes in the way existingclusters interact with each other. The time breaks detection is performedthrough clustering of time instants, that allows for better model parsimony.The time dependent counting data are modeled via non-homogeneous Poissonprocesses (HHPPs), conditionally to the latent variables. In order to handlethe model inference, we rely on a SEM-Gibbs algorithm and the ICL criterionis used for model selection. Numerical experiments on simulated data highlightthe main features of the proposed approach and show the interest of dLBMwith respect to related works. An application to adverse drug reactionin pharmacovigilance is also proposed, where dLBM was able to recognizeclusters in a meaningful way that identified safety events that were consistentwith retrospective knowledge. Hence, our aim is to propose this dynamicco-clustering method as a tool for automatic safety signal detection, to supportmedical authorities.