Determining Factors That Predict Technique Survival on Peritoneal Dialysis: Application of Regression and Artificial Neural Network Methods

Abstract

Background/Aims: Peritoneal dialysis (PD) rates continue to decline worldwide in spite of the increasing number of patients with end-stage renal disease. PD technique failure has been cited as one of the reasons for this decline. The purpose of this study was to compare the factors that predict technique survival using artificial neural network (ANN) and logistic and Cox regression methods. Methods: We used high-quality, prospectively collected data from the United Kingdom Renal Registry and created both ANN and regression models to predict technique survival. Incident PD patients in the UK from 1999 to 2004 were included in the analysis. Technique failure was defined as a change in modality to hemodialysis for a period >30 days. Results: Removal of dialysis center code had a significant effect on the fit and/or predictive performance of all three types of models. In contrast, the effect of demographic data, comorbidity, physical examination and laboratory data varied according to the type of model. Conclusions: PD center significantly impacts PD technique survival. Other putative predictive factors had marginal and/or variable effects. The presence of comorbid conditions and a high body mass index is not consistently associated with increased PD technique failure.