Predicting acute neurological diseases with Bayesian networks

Emergency physicians in small primary care hospitals seeing patients with acute neurological symptoms have difficulty differentiating ischemic from hemorrhagic strokes and from stroke mimics. Telestroke consults with experienced neurologists supplemented by computerized decision support may aid in this time critical situation. Here we present a Stroke Bayesian Network (SBN) based on a na¨ıve Bayesian classifier to predict the most likely stroke etiology-ischemia, hemorrhage or stroke mimic-in an emergency room (ER) setting. As a proof of concept, this probabilistic network was evaluated in a pilot study on a cohort of 44 acute neurological patients admitted to three primary care hospitals associated with the TASC telestroke network in Saxony-Anhalt. In this cohort, the SBN correctly classified 31 of 36 ischemic stroke patients, and all five stroke mimics, but failed to identify three hemorrhages. For the frequent and significant ischemic stroke type, 97% classification precision and 86% sensitivity were obtained. To properly evaluate the SBN performance, a randomized controlled clinical trial should be conducted on a cohort of patients admitted to the ER.