A Bayesian approach for detecting a disease that is not being modeled

Aronis, John M.; Ferraro, Jeffrey P.; Gesteland, Per H.; Tsui, Fuchiang; Ye, Ye; Wagner, Michael M.; Cooper, Gregory F.

The data consists of patient-care reports from emergency departments in the Intermountain Healthcare system in Salt Lake County, Utah from June 1, 2010 through May 31, 2015. These emergency departments capture about 55% of emergency department visits in Salt Lake County. Each report is processed with natural language processing software [16] to extract a set of 65 medical findings that clinicians determined are relevant to the diagnosis of influenza-like illnesses (listed in Table 1). The data were joined to a database of results of laboratory tests. There were a total of 944, 562 patient records. After a strict syntax check of records 3, 063 were removed due to inappropriate laboratory codes leaving 941, 499 reports. From these, we selected 32, 249 reports of patients who definitively tested positive for exactly one of influenza, respiratory syncytial virus (RSV), parainfluenza, or human metapneumovirus (hMPV), or were negative for all four tested diseases.