Modeling the effect of acute inflammation on gas exchange

Abstract

12 prediction tasks were considered, consisting of development and recovery from hyperinflammatory states and kidney dysfunction, with predictions ranging from 1 to 4 days in advance. Four different machine learning algorithms were used: decision trees, first-order random forests, Naive Bayes, and tree-augmented Naive Bayes. Results: Criteria for discrimination and calibration were area under the receiver operator characteristic curve of at least 80% and a Hosmer-Lemeshow H-statistic P value greater than .05. Except for the prediction of development of inflammation, all prediction tasks regarding development satisfied the required criteria. Although recovery from kidney dysfunction was predicted up to 4 days in advance, none of the predictions of recovery from hyperinflammatory states satisfied the criteria completely. Table 1 shows results of a subset of prediction tasks. Conclusions: For the ICU database studied and the predictive tasks considered, standard machine learning techniques result in predictive models with good performances according to the selected criteria.