Abstract
To overcome the low predictive value of SIRS or MEWS, several studies were conducted to predict sepsis. However, the label of sepsis was based on SIRS or ICD code; the time point of the label is obscure whether or not represent sepsis.This study aims to evaluate the performance of a continuous detecting bacteremia based on blood culture by deep learning model for in-hospital patients.Our dataset retrospectively includes 36,023 patients, who have undergone general surgeries from January 2008 to January 2018 in Asan Medical Center in Korea, which is a tertiary hospital. The performance of our model for detecting bacteremia achieves AUROC 0.97 with an average precision-recall 0.17. For predicting bacteremia within the previous 24-hour period, our model achieved AUROC 0.91 with an average PR 0.13. The important variables for detecting bacteremia, according to the occlusion analysis, were vital signs, eGFR, and WBC counts. When the model was incorporating time-series patterns of vital signs and labs, the performance was higher than using only the cross-sectional data. This study demonstrates that the use of a deep-learning model can assist the clinicians to evaluate patients for infection and suggest when to investigate blood cultures and prescribe antibiotics. In the future, with the use of our model in the EMR system, the clinical practice can enhance with a real-time monitoring of patients.
Footnotes
Cite this article as: European Respiratory Journal 2020; 56: Suppl. 64, 3103.
This abstract was presented at the 2020 ERS International Congress, in session “Respiratory viruses in the "pre COVID-19" era”.
This is an ERS International Congress abstract. No full-text version is available. Further material to accompany this abstract may be available at www.ers-education.org (ERS member access only).
- Copyright ©the authors 2020