The evolution of the etiological agents of disease presents one of the greatest challenges for their control, and makes essential complementing standard epidemiological investigations with broader approaches that allow for evolutionary change. Given the stunning genetic diversity that is possible for many such agents, such as the influenza virus, it is impossible to represent all of the diversity manifest at the level of amino acid sequences. We show that drift-variant influenza strains naturally cluster into groups which are associated with functionally important epitopic regions. Dominant clusters typically replace each other every 2-5 years, and this feature is fundamental to the development of vaccination strategies. We furthermore show that stochastic fluctuations can greatly magnify small interference effects among strains, or even among subtypes, leading for example to competitive exclusion in situations where such effects would be unexpected based on the usual deterministic models. We suggest that this effect might be involved in the explanations of some persistent empirical anomalies.