Rune Sejer Jakobsen
Department of Health Science and Technology, Aalborg University, Aalborg, Denmark
Ole Hejlesen
Department of Health Science and Technology, Aalborg University, Aalborg, Denmark
Mads Nibe Stausholm
Department of Health Science and Technology, Aalborg University, Aalborg, Denmark
Simon Lebech Cichosz
Department of Health Science and Technology, Aalborg University, Aalborg, Denmark
Download article
Published in: Proceedings from The 16th Scandinavian Conference on Health Informatics 2018, Aalborg, Denmark August 28–29, 2018
Linköping Electronic Conference Proceedings 151:12, p. 70-74
Published: 2018-08-24
ISBN: 978-91-7685-213-2
ISSN: 1650-3686 (print), 1650-3740 (online)
Evidence is increasing about an unsatisfying performance from the existing non-disease-specific scoring systems in the intensive care unit (ICU). Evidence is furthermore increasing about differences in the mortality rate between diabetics and non-diabetics dependent on the level of blood glucose (BG), but few scoring systems include these variables in the assessment of the patients. 142,404 ICU admissions were included from the eICU database in the development of an unsupervised trained Bayesian Network (BN). The BN suggested that abnormalities in the level of BG should be associated with differences in the mortality rate between diabetics and non-diabetics. The BN showed promising predictive ability with an AUC on 0.86 for predicting death (sensitivity: 75.06, specificity: 78.40 %). 48.43 % of the length of stays (LOS) were correctly predicted. The results were slightly below the results from the APACHE IV scoring system but showed great ability of risk stratification. The BN showed a potential for predicting the patient outcome and might enable an improved method for risk stratifying the patients admitted to the ICU.
Intensive Care Unit, APACHE IV, Mortality, Diabetes Mellitus, Blood Glucose.
