Prediction of the length of stay in Intensive Care Units using artificial neural networks [Identificación de los determinantes de la estadía en Unidades de Cuidados Intensivos usando redes neuronales artificiales]

Abstract

Background: The prediction of the length of stay at the moment of hospital admission is of outmost importance. Many studies have used lineal models to predict this variable, but there are inherent limitations to these models. The use of non lineal models has been scarce. Aim: To develop a non lineal model to predict length of stay in intensive care units. Material and methods: Retrospective analysis of 294 patients admitted to two intensive care units in Santiago, Chile. The severity of the disease motivating the admission was nominally quantified. This and other physiological variables were included in the model. The length of stay was modeled using Artificial Neural Networks. Results: The model yielded an error of 8.7% (3-6 ± 0.4 days, CI 95%) and a correlation coefficient of 0.9 (p <0.001) for the prediction of length of stay. Using net sensitivity analysis, the model determined that gastrointestinal diseases, infections and respiratory problems were the main causes of prolongation of intensive care unit stay. Conclusions: Intensive care units should have, in the future, computer systems that gather vital information to predict length of stay.