Modeling for the prediction of urea concentration and hemodialysis adequacy assessment by artificial neural network

Abstract

In the prediction of hemodialysis adequacy, nephrologists have proposed several models to predict the hemodialysis adequacy. The most favorite models are the Formal Urea Kinetic Model (Formal UKM) and Daugirdas natural logarithm equation. The Formal UKM and Daugirdas equation are the method which Kidney Disease Outcome Quality Initiative (K/DOQI) recommends to be used in the hemodialysis adequacy assessment because the Formal UKM gives high accuracy while the Daugisdas equation is not complicate and easy to be used. This research proposes an alternative way to model hemodialysis adequacy by Artificial Neural Network (ANN). This network model is developed based on two hidden layers. The network model is selected after training, testing and validation process by considering the least mean square error (MSE). The neural network model structure is 8-7-8-1. Simulation results show that though unseen data are given to test the neural network model, the neural network model can still provide good prediction of the Formal UKM with correlation coefficient equal to 0.955.