The Comparison of Selected Data-mining techniques in ICU Mortality Risk Prediction in Imam Hossein hospital

Intensive Care Unit (ICU) is a ward that is critical to improving the health status of critical conditions. Data mining seems to be a good way to optimize the use of resources. Identifying and analyzing the risk factors associated with mortality will lead to more efficient and accurate planning of hospitalization and interventions. In this study, the prediction of mortality of patients in the intensive care unit of Imam Hossein Hospital in Tehran with data mining techniques is discussed.

Based on patient records and hospital information system, 838 patients admitted to the General intensive care unit between 2013 and 2019 in Imam Hossein Hospital in Tehran, the data is needed to collect this research. Algorithms used to classify patients include support vector machines, k nearest neighbor, decision tree, logistic regression and random forest that was reported based on the precision, accuracy, sensitivity, specificity, and roc under the curve.

The results of this study showed, identified 26 factors affecting specific data and pre-processing of data. Among five of the algorithms used in the study, logistic regression algorithm based on the level of roc curve (0.76), accuracy percentage (75.62),precision (68.39),sensitivity (38.65) and specificity (94.53) had better performance in predicting mortality compared to other techniques of study. The variables of Glucose and Partial Thromboplastin time were the most significant effects on mortality based on the logistic regression model.

Data analysis in intensive care unit patients can be an appropriate and practical tool for predicting mortality and its related factors, but according to the quality of data, results are different. And the results extracted from logistic regression can be used as a model to predict the status of mortality in the intensive care unit.