A Machine Learning-Based Surrogate Model of a Hemodialyzer for the Prediction of the Urea Dialyzer Clearance

Abstract

End-stage renal diseases are often treated with hemodialysis. To prescribe the dialysis dose, an index named Kt/V is commonly measured. This index incorporates the effect of different contributions, among which the amount of urea (the solute usually considered as marker in hemodialysis) removed by the dialyzer during the treatment. This work aims at providing a fast and accurate surrogate model of a one-dimensional first-principles model of a dialyzer to predict the value of the dialyzer clearance, considering both diffusive and convective transport. This surrogate model may be used clinically for therapy optimization and intradialytic troubleshooting.