Proposal of Predictive Model of Mortality in the Emergency Room in Addition to the Usual Triage System in a Teaching Hospital in Bogotá, Colombia

Abstract

Triage allows priorization of patients according to their medical urgency. Multiple triage systems have been developed in the world. We propose a statistic model using triage to create an alert system related to mortality rates that could be used as a screening during triage.  

A prospective cohort of 6438 adults who came into Hospital Universitario San Ignacio’s emergency room between 01/03/2018 and 28/02/2019 was used. The data was divided into “training” and “testing”. A bivariate logistic regression between triage and mortality using “training” data was done. Afterwards a multivariate logistic regression was reduced, along with the previous information. In order to find the set point, an Area Under the Curve (AUC) was calculated using the “testing data”. The efficiency was evaluated using measures of association.  

Three different AUC models were created. “triage” showed an AUC-0.82, “reduced” an AUC-0.90 and “age+systolic” an AUC-0.87 not exhibiting significant difference. The reduced model was chosen, presenting sensitivity of 0.869, specificity of 0.842, PPV 0.062, NPV 0.998.

A set point was chosen according to significant variables, and thus finding a higher mortality rate in those classified as triage 1-2, over 58 years old, presenting with SAP under 117mmHg. 

Our final model could be used as additional screening for patients within the same triage classification as an alert system for mortality.

Keywords: Emergency Medical Services, Emergency Medicine, Triage, Mortality, Health Status Indicators