A study of the transferability of influenza case detection systems between two large healthcare systems

This study evaluated the accuracy and transferability of Bayesian case detection systems (BCD) that used clinical notes from emergency department (ED) encounters to detect influenza cases. We developed BCDs at the University of Pittsburgh Medical Center and Intermountain Healthcare in Utah using local data and evaluated their ability to discriminate (1) influenza from non-influenza, and (2) influenza from non-influenzal influenza-like illness (NI-ILI) in test cases from both locations. We also measured the same discriminations when automatic case detection was applied at different times relative to the date of ED registrations. This study demonstrated the high performance of automatic influenza detection in two large healthcare systems, and identified the factors that influence transferability, adding to the evidentiary support for the use of automatic case detection from clinical notes in national influenza surveillance.