Prediction of single-pool Kt/v based on clinical and hemodialysis variables using multilinear regression, tree-based modeling, and artificial neural networks.

The impact of clinical and other variables on single-pool Kt/V (spKt/V) is unclear. The goal of this study was to identify clinical and hemodialysis treatment related predictors of spKt/V and use multilinear regression (LM), tree-based modeling (TBM), and artificial neural networks (ANN) to predict actual spKt/V. When 602 hemodialysis records were analyzed, spKt/V correlated with urea reduction ratio (URR) (r=0.91) and weakly with other variables. When URR was excluded, both LM and TBM identified normalized protein equivalent of total nitrogen appearance (nPNA), prehemodialysis (HD) and post-HD weights, blood flow rate, and dialyzer surface area as predictors of spKt/V. LM identified sex, height, dialyzer ultrafiltration coefficient (Kuf), and duration of dialysis, while TBM identified the dialysis nurse code. Prediction algorithms were developed from a "training" dataset, and validated on a separate ("testing") dataset. Correlation coefficients of predicted spKt/V with measured spKt/V with and without nPNA respectively were 0.745 and 0.679 for LM, 0.6 and 0.512 for TBM, and 0.634 for ANN, which performed better without using nPNA.