In this study Dascena demonstrates the use of a patient stability prediction algorithm. The findings are published in BMJ Open.
Unplanned readmissions to the ICU are often accompanied by many undesirable outcomes, such as increased variance in care, increased length of stay, and increased mortality. This makes resource planning more difficult in the already challenging ICU setting. A machine learning algorithm was developed to predict readmission to the ICU, and its performance was compared to the Stability and Workload Index for Transfer (SWIFT) score which is designed to assist with this task. Using transfer learning techniques, the algorithm was able to predict unplanned ICU readmission with an area under the receiver operating characteristic curve (AUROC) of 0.71, besting the 0.61 AUROC of the SWIFT score. This study showcases how machine learning algorithms may assist clinicians with decisions about patient discharge and resource allocation.
Read the paper here.