A Comparison of Human and Machine-based Predictions of Successful Weaning from Mechanical Ventilation

Abstract

When both are restricted to the same limited set of patient data, appropriately trained neural networks can be as effective as human experts in predicting whether weaning from mechanical ventilation will be successful.

Improvement in the Prediction of Ventilator Weaning Outcomes by an Artificial Neural Network in a Medical ICU

“…[5][6][7][8][9] The prediction accuracy rate and area under the ROC curve of the ANN model in this study were similar to those in other studies even when we took breathing pattern variables into consideration. 7,22 Different patient populations may have different key factors that determine who may be liberated from the mechanical ventilator. Mueller et al 23 observed that gestational age, arterial blood gas, and ventilator settings were critical factors when deciding whether newborns could be successfully extubated.…”

“…Therefore, sensitivity analysis or cross-validation has been used to explain how ANN models work. 20 Gottschalk et al 22 were the first to report that an ANN model that had been trained by using 4 variables (ie, tidal volume [V T ], minute ventilation, breathing frequency, and maximum inspiratory pressure [P I max ]) recorded during an SBT in ICU subjects could be as effective as experts in predicting whether patients could be successfully weaned from mechanical ventilation. Mueller et al 23 demonstrated that their ANN model outperformed clinical expertise and multiple logistic regression in predicting extubation outcomes in premature newborns.…”

Data mining of patients on weaning trials from mechanical ventilation using cluster analysis and neural networks

“…In our case, linear and non-linear feed-forward neural networks have been used, together with feature selection methods, to classify patients who presented success or failure in the weaning process [9,10]. Patients were classified in three groups, taking into account clinical criteria based on T-tube test: group S, 88 patients whose T-tube test was overcome successfully; group F, 38 patients who failed the test and therefore could not be extubated; and Group R, 23 patients who passed the T-tube test and were disconnected from mechanical ventilation, but they had to be reintubated before 48 hours.…”

The Role of Signs or Symptoms

“…A number of decision algorithms have been proposed (e.g., Giraldo et al 2006;Tehrani 2007). Gottschalk et al (2000), however, report a study in which they trained a neural network (which I will consider a machine) by repeatedly exposing the network to various respiratory parameters, and then compared the ability of the network to successfully predict the outcome of a weaning effort. The neural network's predictions were compared to those of a clinician.…”

“…The researchers also varied the number of indicators that the neural network and the clinician used when making their decisions. They concluded that, "When both are restricted to the same limited set of patient data, appropriately trained neural networks can be as effective as human experts in predicting whether weaning from mechanical ventilation will be successful" (Gottschalk et al 2000;p. 160).…”