The prediction, prevention, and management of infectious diseases in the United States is either geographically homogeneous or is coordinated through ad-hoc administrative regions, ignoring the intense spatio-temporal heterogeneity displayed by most outbreaks. Using influenza as a case study, we characterize a regionalization of the United States. Based on influenza time series constructed from fine-scale insurance claims data from 2002-2009, we apply a complex network approach to characterize regions of the U.S. which experience comparable influenza dynamics. Our results identify three to five epidemiologically distinct regions for each flu season, with all locations within each region experiencing synchronous epidemics, and with an average of a two week delay in peak timing between regions. We find that there is significant heterogeneity across seasons in the identity of the regions and the relative timing across regions, making predictability from one season to the next challenging. Within a given season, however, our approach shows the potential to inform on the shaping of regions over time, to improve resources mobilization and targeted communication. Our epidemiologically-driven regionalization approach could allow for disease monitoring and control based on epidemiological risk rather than geopolitical boundaries, and provides a tractable public health approach to account for vast heterogeneity that exists in respiratory disease dynamics.