Background: The field of Emergency Critical Care (ECC) is rapidly expanding. In order to assess the efficacy of ECC interventions, we need a reliable illness severity score that can be calculated based on variables available in the ED. In this study, we assessed the ability of a modified SOFA score, calculated using patient variables available at the time of critical care admission from the ED (eccSOFA), to predict in-hospital mortality.
Methods: This was a retrospective cohort study using electronic health record data from an academic medical center ED. All adult patients with a critical care admission order placed in the ED from 10/24/2013 to 9/30/2016 were included. eccSOFA scores were calculated using the worst of SOFA variables recorded from ED arrival up to 1 hour after critical care admission order. Any variables not available within this time frame were assigned a score of zero. To determine the discriminatory ability of the eccSOFA score with regard to in-hospital mortality, we generated an ROC curve and calculated the area under this curve (AUROC). We also established mortality estimates for 3 eccSOFA ranges for subsequent assessment of calibration.
Results: Of 3,912 patients, 57.5% were male, the median age was 63, and 11.4% died in hospital. 67% of patients had all eccSOFA variables available, with the most common missing variables being GCS and PaO2/FiO2. Overall, the AUROC for eccSOFA as a predictor of in-hospital mortality was 0.77 (95% CI 0.74 – 0.79). The proportion of patients in each eccSOFA category were as follows: eccSOFA 0-3, 51.7%; eccSOFA 4-6, 29.2%; and eccSOFA ≥ 7, 19.1%. The mortality for each of these eccSOFA categories was 3.5%, 13.5%, and 29.6%, respectively.
Conclusion: As a predictor of in-hospital mortality, the eccSOFA score has good discriminatory ability, with AUROC roughly comparable to other commonly used illness severity scores. The advantage of eccSOFA is that it can be calculated based on variables that are commonly available at the time of critical care admission order. Assessing the calibration of our absolute risk estimates will require additional studies in other settings.