Urban flood evacuation simulations commonly rely on two-dimensional (2D) representations of urban environments, even though real cities are inherently three-dimensional (3D). Because flood inundation is closely shaped by the 3D structure of urban space, this mismatch may bias estimates of evacuation feasibility and timing while limiting the evaluation of disaster response strategies. This paper therefore proposes a conceptual framework for 3D urban evacuation simulation under flooding by coupling an urban digital twin (UDT)-based 3D navigation network with an agent-based model (ABM). The navigation network represents multi-level navigable spaces and their connectivity as a 3D graph with mode-dependent constraints for walking, driving, and public transport. The ABM incorporates variables associated with the vertical dimension of evacuation processes. An illustrative example is provided to demonstrate the framework. The main contributions are: 1) a conceptual framework integrating UDTs and ABM for 3D urban evacuation simulation; 2) a hierarchical definition and specification of navigable spaces and their connectivity for constructing a 3D graph-based navigation network; and 3) the identification of variables for representing 3D evacuation processes. The framework helps enable more effective disaster prevention and response strategies.