Current large-scale coastal flood risk assessments are typically based on scenarios considering a range of spatially uniform return periods (RP). These assessments do not account for the spatial variability of real flood events, and only estimate average annual losses. In this study, we address these limitations by developing a novel event-based probabilistic framework to capture the spatial dependence structure of coastal floods, and use it to investigate the effects of spatial dependence on national flood risk estimates globally. We show that the widely-used RP-based approach gives lower damage estimates for relatively low return periods while higher damages are estimated for medium-to-large return periods. The intersection point where lower damage estimations turn into higher damage estimations varies across countries and is primarily dependent on local flood protection standards. We also provide the first global mapping of differences in risk indicators between these two approaches in terms of expected annual damages (EAD) and 1-in-200-year damages. We show that spatial dependence has minor effects on the EAD but the RP200 damage is estimated higher for 76% of global countries by the RP-based approach. Accounting for flood protection standards is found to increase these differences. Lastly, we demonstrate the added value of our approach by showing the flood damages of the simulation year with the highest combined annual damages at a subnational scale for each continent. Our framework provides more accurate large-scale coastal flood risk estimates, which can aid in better regional planning decisions, more precise insurance pricing, and improved emergency responses.