Effect of cause and time of dropout on the residual GFR: A comparative analysis of the decline of GFR on dialysis

Effect of cause and time of dropout on the residual GFR: A comparative analysis of the decline of GFR on dialysis. The decline of residual renal function (RRF) on dialysis has been reported to be slower in peritoneal dialysis (PD) then hemodialysis (HD). However, some clinicians have questioned whether this reported difference might not be caused by selection bias. In particular, if continuous ambulatory PD (CAPD) delivers only marginally adequate therapy as some clinicians speculate, then perhaps those patients on CAPD with low glomerular filtration rate (GFR) are purposefully switched to HD. If true, transferring CAPD patients with low GFR to HD could create a selection bias that very well may account for the differences in GFR between PD and HD. This is particularly problematic if one then censors patients at the time of transfer from PD to HD from analysis (that is, patients are no longer followed in the study once they have switched treatment modalities). When this occurs, the data are said to be informatively censored, a term used by statisticians to describe any kind of systematic bias associated with censored or incomplete data. In particular, informative censoring occurs when patients who die or transfer to another modality very early have an associated lower starting GFR or higher rate of decline of GFR than patients who either complete the study or who die or transfer much later. If patient dropout is indeed related to the rate of decline in GFR and if this relationship differs between PD and HD but is ignored in the analysis, then the results of such analysis may be biased. This article analyzes the decline in GFR among 141 incident dialysis patients (39 HD and 102 PD) undergoing either HD or PD at the University of Missouri-Columbia. The decline in GFR was modeled as a nonlinear function of time, taking into account the possibility that missing values of GFR may be associated with patient dropout (death, transfer to another modality, or transplantation). To safeguard against this possibility, we utilized a conditional nonlinear mixed-effects model. The model was used to fit and compare each patient's GFR data to time adjusting for the patient's treatment modality (HD vs. PD), cause of dropout (death, transfer, transplant, lost to follow-up/study ended), and time to dropout. The model allowed a comparison of the starting GFR and the rate of decline in GFR between PD and HD adjusting for these three factors. The results of our analysis suggest t