In the present letter, a closed-form framework is derived for the system-level modelling, analysis and optimization of the energy efficiency (EE) in LoRa networks. The proposed framework is derived by exploiting stochastic geometry tools and associates the Long Range (LoRa) EE with the density of end-devices (EDs) and the ED transmit power. The analysis reveals the trends of the EE curve with respect to each of the two parameters while revealing the existence of an optimal transmit power that maximizes the EE. Thus, the proposed framework arises as a valuable tool, of significant practical value, for the optimization of the receiver/transmitter design, and of the network deployment in LoRa networks. The robustness of the framework is verified even at the asymptotic cases of fully-loaded or sparsely-loaded LoRa networks.