Emergency events characterized by high uncertainty and complexity bring tremendous pressure and challenges to our society. Emergency decision-making (EDM) is an effective way to mitigate the losses caused by emergency events. The generation of alternatives and the selection of the best emergency plan are crucial to the successful management of an emergency event. To improve the efficiency of EDM, this study proposes a novel two-stage EDM method. In the first stage, to fully represent emergency events and measure their similarities, we adopt heterogeneous multi-attribute information in the dynamic case-based reasoning (CBR) model to generate alternatives. After that we provide an adaptive reaching process model with a dynamic interactive strategy to obtain similar historical emergency events. In the second stage, we apply the prosper theory and the linguistic term information to select the optimal emergency plan. To test the robustness of the two stage EDM methodology, we verify the feasibility and effectiveness of the method with a case study about evaluating emergency plans for public health emergencies. Lastly, we conclude this work with more sensitivity analysis and some discussions about future research.