Introduction

Echo-state networks are part of a group of reservoir computing methods and are basically a form of recurrent artificial neural networks. These methods can perform classification tasks on time-series data. The recurrent artificial neural network of an echo-state network has an echo-state characteristic. This echo state functions as a fading memory: samples that have been introduced into the network in a further past are faded away. The echo-state approach for the training of recurrent neural networks was first described by Jaeger [1]. In clinical medicine, until this moment, no original research articles have been published to examine the use of echo-state networks.

Methods

The present study examines the possibility of using an echo-state network for prediction of dialysis in the ICU. Therefore, diuresis values and creatinine levels of the first 3 days after ICU admission were collected from 830 patients admitted to the ICU between 31 May 2003 and 17 November 2007. The outcome parameter was the performance by the echo-state network in predicting the need for dialysis between day 5 and day 10 of ICU admission. Patients with an ICU length of stay <10 days or patients that received dialysis in the first 5 days of ICU admission were excluded. Performance by the echo-state network was then compared by means of the area under the receiver operating characteristic curve (AUC) with results obtained by two other time-series analysis methods by means of a support vector machine and a naive Bayes algorithm.

Results

The AUCs in the three developed echo-state networks were 0.822, 0.818, and 0.817. There were no statistically significant differences at the 0.05 level with the results obtained by the support vector machine and the naive Bayes algorithms.

Conclusion

This proof of concept study is the first to evaluate the performance of echo-state networks in predicting the need for dialysis in an ICU population. The AUCs of the echo-state networks were good and comparable with the performance of other classification algorithms.