A review of modeling methods for predicting in-hospital mortality of patients in intensive care unit

Severity scoring models, which can be used to predict the probability of death for patients in intensive care units (ICUs) have been developed for nearly 30 years and are clearly the recognized risk stratified tools and ICU quality metrics available today. However, the accuracy of these scoring models remains far from perfect when it was applied in a new ICU setting. Therefore, implementing locally customized prognostic models that show better predictive performance rather than adopting standard severity scoring models seem to be more worthwhile in the future as the electronic medical records getting popular. In this review, we aim to outline the essential procedures and concepts for the development of reliable models in the field of in-hospital mortality prediction of ICU patients. First, we present the common variables or predictors necessary to construct mortality prediction models. Then, we elaborate the following modeling methodologies: logistic regression (LR), artificial neural network (ANN), decision tree (DT), and support vector machine (SVM). LR underlies most of the illness scoring models, while the others are extensions of traditional statistical approaches and show promise as more useful tools in the analysis of large, and heterogeneous data. Finally, we describe the common approaches and measures used to evaluate model performance.