Computer-aided national early warning score to predict the risk of sepsis following emergency medical admission to hospital: A model development and external validation study

Abstract

BACKGROUND: In hospitals in England, of age and older discharged over 0.777, M2 0.791). For NEWS of 5 or patients’ vital signs are monitored and 24 months from 2 acute care hospital higher, sensitivity increased (YH: 47.24% summarized into the National Early centres (York Hospital [YH] for model v. 50.56% v. 52.69%; NH: 37.91% v. Warning Score (NEWS); this score is development and a combined data set 43.35% v. 48.07%), the positive likeli-more accurate than the Quick Sepsis-from 2 hospitals [Diana, Princess of hood ratio increased (YH: 2.77 v. 2.99 v. related Organ Failure Assessment Wales Hospital and Scunthorpe General 3.06; NH: 3.18 v. 3.32 v. 3.45) and the (qSOFA) score at identifying patients Hospital] in the Northern Lincolnshire positive predictive value increased (YH: with sepsis. We investigated the extent and Goole National Health Service 11.44% v. 12.24% v. 12.49%; NH: 22.75% to which the accuracy of the NEWS is Foundation Trust [NH] for external v. 23.55% v. 24.21%). enhanced by developing and comparing model validation). We used a validated 3 computer-aided NEWS (cNEWS) mod-Canadian method for defining sepsis INTERPRETATION: From the 3 cNEWS els (M0 = NEWS alone, M1 = M0 + age + from administrative hospital data. models, model M2 is the most accu-sex, M2 = M1 + subcomponents of NEWS rate. Given that it places no additional + diastolic blood pressure) to predict the RESULTS: The prevalence of sepsis was burden of data collection on clinicians risk of sepsis. lower in YH (4.5%, 1596/35 807) than in and can be automated, it may now be NH (8.5%, 2983/35 161). The C statistic carefully introduced and evaluated in METHODS: We included all emergency increased across models (YH: M0 0.705, hospitals with sufficient informatics medical admissions of patients 16 years M1 0.763, M2 0.777; NH: M0 0.708, M1 infrastructure.