We present our use of analytical mean-field approaches in investigating how the interplay between graph topology and excitable dynamics produce spatio-temporal patterns. We first detail the derivation of mean-field equations for a few simple model situations, mainly 3-state discrete-time excitable dynamics with an absolute or a relative excita-tion threshold. Comparison with direct numerical simulation shows that their solution satisfactorily predicts the steady-state excitation density. In contrast, they often fail to capture more complex dynamical features, however we argue that the analysis of this failure is in itself insightful, by pinpointing the key role of mechanisms neglected in the mean-field approach. Moreover, we show how second-order mean-field approaches, in which a topological object (e.g. a cycle or a hub) is considered as embedded in a mean-field surrounding, allow us to go beyond the spatial homogenization currently associated with plain mean-field calculations. The confrontation between these refined analytical predictions and simulation quantitatively evidences the specific contribution of this topological object to the dynamics. Mathematics Subject Classification (2010). Primary 05C82; Secondary 92C42.