Late Breaking Abstract - Dynamic Early Warning Score versus National Early Warning Score-2 for predicting death or intensive care unit admission in respiratory patients

Background: The National Early Warning Score-2 (NEWS2) is used to detect patient deterioration in UK hospitals but relies on a snapshot assessment. We developed a Dynamic Early Warning Score (DEWS) to take into account changes in clinical observations over time, and assessed its predictive accuracy in respiratory patients. Methods: Clinical observations data were extracted for 31590 respiratory in-patient episodes (53.5% female, mean age 69.9) at a large teaching hospital (Nottingham University Hospitals NHS Trust) over a six-year period. The dataset comprised 1037349 date and time-stamped observations sets. Data from 2015-2019 and from 2020 were used for model development and validation respectively. The primary outcome was death or intensive care unit (ICU) admission within 24 hours of an observation set. This was positive in 2.3% of the observations sets. Using similar methodology to Zhu et al (Resuscitation. 2020; 157: 176-84) we incorporated the rolling average, standard deviation and trend of each clinical observation into a logistic regression model. Results: The area under the receiver operating curve (AUC) for predicting the primary outcome was 0.904 for DEWS versus 0.862 for NEWS2 in the training dataset, and 0.902 for DEWS versus 0.854 for NEWS2 in the validation dataset. A NEWS2 score of ≥5 had a sensitivity of 0.765 and specificity of 0.806. The DEWS cut-point with equivalent sensitivity (0.765) had a superior specificity of 0.883. Conclusions: DEWS has superior performance to NEWS2 in a respiratory population with respect to predicting death or ICU admission within 24 hours.