Predicting length of stay using regression and Machine Learning models in Intensive Care Unit: a pilot study

Healthcare Associated Infection (HAI) is a major health problem in several departments of the hospital sector. These infections cause prolonged length of stay (LOS), complications, and increased hospital costs. In this paper, medical record data of 415 patients admitted to in the Adult and Neonatal Intensive Care Unit (ICU) where there was a high risk of contracting HAIs, were used collectively. The aim was to create models capable of predicting LOS, measured in days, considering preoperative clinical information. Multiple linear regression analysis and Machine Learning (ML) regression analysis were performed. Subsequently, the LOS was grouped by weeks and classified with the ML classification algorithms. Multiple linear regression was implemented with IBM SPSS, the coefficient of determination (R2) was equal to 0.343. A regression with ML algorithms is performed with the Knime analysis platform. The best R2 was obtained from the Random Forest (R2 =0.414) and Gradient Boosted Tree (R2 =0.382) algorithms. Regarding the classification analysis, the RF and Multi-Layer Perceptron algorithms showed accuracy respectively 49.398% and 46.988%, an error of 50.602% and 53.012%. The goal was to create a model to support physicians in evaluating the hospitalization of patients at risk of HAI in the ICU. (H. De Koning, 2006)