Abstract
The most effective target for automated influenza surveillance systems is not known. In this work, we simulated a prospective surveillance system operating on authentic historical series of daily casecounts. We determined how long the system would take to detect an injected outbreak of synthetic cases. For influenza epidemics where >/= 5% of cases develop pneumonia, we found shorter outbreak detection delays when surveillance targeted only patients with pneumonia rather than all patients with acute respiratory infections.