In this study Dascena evaluates a mortality prediction algorithm. The findings are published in Annals of Medicine and Surgery.
Accurate predictions of patient mortality risk could help with medical intervention, patient management, and the allocation of limited and costly ICU resources. However, existing clinical decision support (CDS) systems have little predictive value in the clinical setting, as seen in their suboptimal specificity and sensitivity when applied to patient mortality predictions. A new machine learning approach, AutoTriage, utilizes eight widely used patient vital signs and performs multi-dimensional analysis of the correlations and trends between these measurements to produce accurate predictions about patient mortality. When applied to 12-hour mortality prediction, AutoTriage achieved an area under the receiver operating characteristic curve (AUROC) of 0.88 with 81% specificity at 80% sensitivity. These results show AutoTriage performs better than the most commonly used CDS systems that include Modified Early Warning Score (MEWS), Sepsis-Related Organ Failure Assessment (SOFA), and Simplified Acute Physiology Score (SAPS II).
Read the paper here.