Support Vector Machine for Determining Dose of Dialysis

Calculation of dose of haemodialysis using blood-based modelling is subject to controversies as it is based on unrealistic assumptions. This paper proposes the use of dialysate-based modelling by SVMs to calculate the delivered dose of dialysis. The authors use the solute removal index (SRI), which is correlated to the amount of urea removed, for calculating the dose. The SVM model was trained to recognise the evolution of weight, blood urea nitrogen concentration and solute removal index with respect to time and then used to predict the solute removal index. When the estimated SRI values were compared to the actual SRI values determined by the standard method, the prediction errors were small. This paper is the first demonstration that SVM regression can predict delivered dose of haemodialysis with a clinically acceptable accuracy. The result is an effective technique that will offer the physician a better guide to the monitoring and prescription of haemodialysis therapy thereby reducing the mortality rate among patients.