Length of Stay Prediction in Acute Intensive Care Unit in Cardiothoracic Surgery Patients
Nafiseh Mollaei, Ana Rita M. Londral, Cátia Cepeda, Salomé Azevedo, Jorge Pinheiro Santos, Pedro Coelho, José Fragata, Hugo Filipe Silveira Gamboa · 2021
The goal of this study was to apply machine learning (ML) methods to predict the Length of Stay in an Intensive Care Unit (LOS-ICU) based on preoperative factors. To optimize the capacity of the ICU in surgery department, the prediction of a long stay (more than 2 days) can support the clinical decision making on accepting or delaying a patient intervention, considering the ICU occupancy. A database with records from 7364 patients that were operated in the Cardiothoracic surgery department of a public Portuguese hospital was used as the base of ML algorithms training. Regarding the risk of the patients to be in the group of long LOS-ICU, we compared five machine learning algorithms including Gradient Boosting, Random Forest, Support Vector Machine (SVM), Adaboost and Logistic Regression. We studied the classifier performance to adjust the sensitivity of a long stay classification, in order to reduce the potential of long LOS-ICU classification being miss classified as a short LOS-ICU.