Einsatz neuronaler Netze in der Notaufnahme: Eine Illustration anhand der Prognose der stationären Aufnahme

Background: A portion of hospital admissions derive from unscheduled inpatients admitted through the emergency department (ED). Preparation for admission is routinely based on the decision by the physician in charge. The resulting time for the receiving ward to prepare for the admission is sometimes not sufficient and delays may occur. Aim: The goal of this work is to demonstrate the benefits of using artificial neural networks (ANN) by illustrating its application within the context of predicting the probability that a patient in the ED will be admitted to the hospital. The question arises whether routinely collected patient data that are available in most EDs can already contribute to reducing the delay in the admission process. Materials and methods: On the basis of limited and routinely collected data from a hospital information system, an ANN has been developed for an ED to predict whether admission to an inpatient ward is necessary. The ANN is implemented using the open source software R. Results: Using routinely collected data, the ANN has an accuracy of 76.64%. The sensitivity, i.e., the share of correctly predicted admissions is 66.93%, which is lower than the specificity of 82.13% (share of correctly predicted discharges from ED). Discussion: The results show that an ANN can make a valuable contribution to improve process management regarding admissions from the ED even if only routinely collected data are used. It is expected that additional variables, such as a patient’s age, will increase the accuracy of the prediction. © 2022, The Author(s).