Comparing Automated Acuity and SOFA Scores for Predicting ICU Mortality

First Author: Ludmila Brochini


Introduction: This study aims to perform a comparative validation of the Philips Automated Acuity (AA) score and the Sequential Organ Failure Assessment (SOFA) score in predicting in-ICU mortality, using data from a retrospective cohort of tele critical care patients collected over a 12-year period.


Methods: AA is a rule-based score used in ICUs to estimate patient acuity for triage and monitoring, based on cardiovascular, respiratory, infection, neurologic, renal, and hematologic data. The AA score ranges from 0 to 30 and combines components derived from vital signs, medication records, lab results, and physical exams, updating every 5 minutes during the patient's stay. This study evaluated the average AA and SOFA scores during the first 24 hours of ICU stay in relation to mortality. The dataset consisted of 53,729 patients from 58 ICUs in 9 US hospitals, in the period from 2017 to 2025, with a 4.2% ICU mortality rate. Statistical analysis was done using Mann Whitney U test, Spearman correlation, and the Area Under the Receiver Operating Curve (AUROC). Confidence intervals were calculated using two-sided bias-corrected and accelerated bootstrapping at a 95% confidence level.


Results: The average scores on the first day for both AA and SOFA showed a distinct distribution between the populations of survivors and non-survivors (Mann-Whitney p-value < 0.001). For both AA and SOFA, average mortality rate per score point was highly correlated with each score, achieving a Spearman correlation of 0.99. Mortality rates ranged between 0.06% and 66.7% from lowest to highest AA scores and between 1.2% and 71.4% from lowest to highest SOFA scores. At the individual patient level, AA presented AUROC confidence interval of [0.86, 0.9], while SOFA presented AUROC in [0.78, 0.83], both for predicting in-ICU mortality.


Conclusions: Both AA and SOFA scores effectively indicated rising mortality rates with increasing scores, but AA showed superior performance in correlating individual scores with mortality through AUROC measurements. This suggests AA may be valuable for assessing patient acuity and supporting clinical decisions in critical care.