To assure adequate and efficient treatment in the emergency department (ED) despite increasing patient numbers, early risk stratification might be helpful for directing resource allocation.
To determine whether routine clinical data can predict in-hospital mortality in nonsurgical ED patients and to specifically identify the best predictive parameters.
This retrospective cohort study investigated 34,333 nonsurgical adult patients who attended one of the two participating EDs in Berlin, Germany, within 1 year. Routine clinical data were analysed for their potential to predict in-hospital mortality using logistic regression as well as classification and regression tree (CART) analysis. A validation dataset contained 35,646 patients of the following year.
In-hospital mortality was 1.8 % (634/34,333). C-reactive protein (CRP) and red cell distribution width (RDW) were the best predictors of mortality. A model with nine predictors (CRP, RDW, age, potassium, sodium, WBC, platelets, RBC and creatinine) achieved an area under the receiver operating characteristic curve (AUROC) of 0.870 (95 % confidence interval, CI: 0.857–0.883). A three-marker model (CRP, RDW, age) resulted in an AUROC of 0.866 (95 % CI: 0.853–0.878). In the independent validation dataset the AUROC for this three-marker model was 0.837 (95 % CI: 0.825–0.850). CART analysis corroborated the importance of CRP and RDW, and a clinical algorithm for risk stratification was developed (Emergency Processes in Clinical Structures, EPICS score).
Two different statistical procedures and independent validation revealed similar results, suggesting a combination of CRP and RDW as a score (EPICS score) for early identification of high-risk patients. This might be particularly helpful in overcrowded situations and where resources are limited. The suggested score should be validated and potentially adapted to diverse ED settings and patient populations in international multicentre trials.
