Bayesian Learning of the Gas Exchange Properties of the Lung for Prediction of Arterial Oxygen Saturation

This paper describes how real-time Bayesian learning of physiological model parameters is used to predict arterial oxygen saturation at the bedside. The efficacy of using these predictions as a decision support tool in a system for estimating gas exchange parameters of the lung (ALPE) was tested retrospectively. For the predictions to offer effective decision support they need to be accurate and safe. These qualities were tested for two patient groups, using two different test strategies for each group. The prediction accuracy when used in combination with the predictions’ safety margin was found to be adequate in all the test cases. Thus the method described can be used as the basis for effective model-based decision support in ALPE.