Use of learning approaches to predict clinical deterioration in patients based on various variables: A review of the literature

Abstract

Machine learning can be considered as the current gold standard for predicting deterioration in Intensive Care Unit patients and is in extensive use throughout the world in different fields. As confirmed by many studies, preventing the occurrence of the onset of deterioration in a sufficient time window is a priority in healthcare centers. Also, the significance of enhancing the quality of hospital care and the reduction of adverse outcomes is of great importance. Notably, it is hypothesized that by exploiting recent technologies, models built upon dynamic variables (e.g. vital signs, lab tests, and demographic variables) could reinforce the predictive ability of models aimed at detection of in clinical deterioration with high accuracy, sensitivity and specificity. This manuscript summarises the techniques and approaches proposed in the literature for predicting deterioration and compares the performance and limitations of various approaches grouped based on their application. While several approaches can attain promising results, there is still room for additional improvement, especially in pre-processing and modeling enhancement steps where most methods do not take the necessary steps for ensuring a high-performance result. In this manuscript, the most effective machine learning models, as well as deep learning models, for predicting deterioration of patients are discussed in hopes of assisting the readers with ascertaining the best possible solutions for this problem.