Prediction of Length-of-stay at Intensive Care Unit (ICU) Using Machine Learning based on MIMIC-III Database

The length-of-stay (LOS) is critical for patient care and accommodation in the intensive care unit (ICU). In this work, we developed a framework to predict the LOS using the Medical Information Mart for Intensive Care (MIMIC-III) database. We extracted six features from individual patients and submitted them to the regressors model and examined how well these features could predict LOS. We considered four prediction regimes; extreme gradient boosting (XGBoost), support vector regressor, random forest, and voting regressor. Our analysis reveals that XGBoost yields the best result among other regressors with R
²
0.86 and root mean square error (RMSE) 1.2. Remarkably, our results show that ICD9 (9
^th
International classification of diseases code), saline intake per hour, and drug rates are the top three critical features for predicting the LOS.