eccSOFA: SOFA illness severity score adapted to predict in-hospital mortality in emergency critical care patients.

BACKGROUND

Boarding of ICU patients in the ED is increasing. Illness severity scores may help emergency physicians stratify risk to guide earlier transfer to the ICU and assess pre-ICU interventions by adjusting for baseline mortality risk. Most existing illness severity scores are based on data that is not available at the time of the hospital admission decision or cannot be extracted from the electronic health record (EHR). We adapted the SOFA score to create a new illness severity score (eccSOFA) that can be calculated at the time of ICU admission order entry in the ED using EHR data. We evaluated this score in a cohort of emergency critical care (ECC) patients at a single academic center over a period of 3 years.

METHODS

This was a retrospective cohort study using EHR data to assess predictive accuracy of eccSOFA for estimating in-hospital mortality risk. The patient population included all adult patients who had a critical care admission order entered while in the ED of an academic medical center between 10/24/2013 and 9/30/2016. eccSOFA's discriminatory ability for in-hospital mortality was assessed using ROC curves.

RESULTS

Of the 3912 patients whose in-hospital mortality risk was estimated, 2260 (57.8%) were in the low-risk group (scores 0-3), 1203 (30.8%) in the intermediate-risk group (scores 4-7), and 449 (11.5%) in the high-risk group (scores 8+). In-hospital mortality for the low-, intermediate, and high-risk groups was 4.2% (95%CI: 3.4-5.1), 15.5% (95% CI 13.5-17.6), and 37.9% (95% CI 33.4-42.3) respectively. The AUROC was 0.78 (95%CI: 0.75-0.80) for the integer score and 0.75 (95% CI: 0.72-0.77) for the categorical eccSOFA.

CONCLUSIONS

As a predictor of in-hospital mortality, eccSOFA can be calculated based on variables that are commonly available at the time of critical care admission order entry in the ED and has discriminatory ability that is comparable to other commonly used illness severity scores. Future studies should assess the calibration of our absolute risk predictions.