Artificial Neural Networks to Predict Postoperative Nausea and Vomiting

Abstract

The effect of postoperative nausea and vomiting (PONV) is not a dangerous effect but it causes discomfort. It is therefore an issue to reduce this effect as much as possible. The reasons for its occurrence, however, are not yet clear. Many factors seem to be candidates for facilitating or depressing this post-anaesthetic side effect ranging from individual to treatment factors. Even though some prediction of PONV can be made on statistical analyses, there is still a high level of uncertainty. In this paper, we use neural networks for predicting PONV and show how feed-forward networks can be designed as problem-specific predictors. The resulting networks outperform the statistical approaches and simple black box approaches using neural networks. The methods applied for designing the networks and encoding the data are furthermore generalizable for other medical applications.