A DECISION SUPPORT TOOL FOR CATEGORIZING PATIENT ACUITY AT EMERGENCY DEPARTMENT

Purpose: The urgency of care at an Emergency department (ED) in Singapore is prioritized by the nurses’ triage that categorizes patients by their acuity level upon their arrival.  Ministry of Health (MOH) defines patient acuity category (PAC) 1, 2 and 3 as resuscitation, major emergency, and minor or non-emergency, respectively. The objective of this study is to develop and validate a computerized decision support tool to help nurses make a quick and objective decision on PAC upon arrival.

Method: A retrospective study using routinely collected hospital data to develop a predictive model.  All patient visits to the ED from Jan to Sep 2010 were included. Data was extracted from the hospital information system, which included the patients’ PAC, demographics, vital signs, presenting symptoms, comorbid conditions, arrival mode and mobility. The outcome categories (PAC 1, 2 and 3) were unevenly distributed and the parallel line test was not satisfied. Therefore two binary logistic regression models with resampling were applied to identify the significant predictors and to estimate the parameters of the models. Model 1 identified PAC 1 and 2 from PAC 3. Model 2 identified PAC 1 from PAC 2. Split validation was applied for Model 1 and bootstrap validation was applied for Model 2. C-statistics of the Receiver Operation Characteristic (ROC) plot was applied to assess the discrimination power of the two models.

Result: There were about 118,000 ED visits from Jan to Sep 2010.  Of these, 1.6% were PAC1, 40.5% were PAC2, and 57.6% were PAC3. Model 1 identified patients’ presenting symptoms, mobility, age, cause of injury, pulse, comorbid condition of asthma, pain score, body temperature, and blood pressure as the important predictors. Model 2 identified patients’ presenting symptoms, comorbid condition of asthma or COPD, pain score, pulse, mobility, body temperature and blood pressure as important predictors.   The model demonstrated good performance on the validation dataset with the c-statistics of ROC being 0.945 (95% CI: 0.943-0.947) for Model 1 and 0.906 (95% CI: 0.900-0.913) for Model 2.  

Conclusion: We present an internally validated ordinal regression model to assess the patients’ need for urgent care at the point of triage. The model may be deployed to support the nurses make decisions on patients’ acuity level due to its good validation performance.