Introduction Guidance from the National Institute of Health and Care Excellence in 20071 has led to the almost universal use of early warning scores (EWS) derived from vital signs observations in hospitals in the UK to highlight patients at risk of deterioration. Lack of high quality prospective studies limits our understanding of the impact of using such monitoring systems on outcomes and working patterns. No EWS has been validated in respiratory patients despite widespread use. Our aim was to examine the ability of both the locally used EWS and National Early Warning Score (NEWS) to predict patient deterioration and associated burden of escalations generated in a respiratory cohort.
Methods Vital signs observations and outcomes for all admissions under the respiratory department at a tertiary referral centre between April 2015 and March 2016 were analysed. Predicted and actual escalation patterns in relation to primary endpoint of mortality were examined comparing NEWS to local EWS. Patients documented as receiving end of life care were removed from analysis.
Results Over 12 months there were 165,184 observations sets during 5293 admissions, with a mean of 38 observations per admission (standard deviation 50). Occurrence of primary endpoint of in-hospital death was 6.74%. 13% of observations triggered clinical escalation to a registered nurse or beyond, with mean of 1075 per month. 112 (31%) patients who died did not trigger escalation on their final set of observations, 1 patient was escalated despite scoring below protocol threshold. Applying NEWS criteria retrospectively predicts 6 patients who died would not be escalated, while generating a mean of 12,409 escalations of vital signs observations per month to registered nurse or beyond, 1,621 in patients who went on to die in hospital.
Conclusion Our data suggests that neither scoring system provides effective monitoring in patients with respiratory disease, falling short on either sensitivity or specificity for predicting in-hospital death. As more data becomes available, modelling may allow more accurate prediction systems to be developed.