Optimal strategies for modeling the reciprocal of creatinine versus time in renal transplant recipients and patients with native chronic renal disease

Abstract

Few studies have evaluated the least squares (LS) or alternative regression methods to estimate loss of renal function using the reciprocal of creatinine over time in renal transplant recipients or have compared their performances in patients with chronic renal insufficiency (CRI). We evaluated the LS and alternative methods using the proportion of explained variance, measured by R2, and prediction of the final creatinine level. The study included two groups of transplant recipients: (1) patients who developed graft failure (FAIL; n = 31) and (2) patients with an episode of biopsy-proven acute rejection with or without subsequent renal function loss (BXAR; n = 98) and a third group of individuals with CRI (n = 28). The LS method performed poorly in both transplant groups (mean R2 range, 0.35 to 0.44; 32% to 45% with final creatinine(actual) - creatinine(predicted) +/- 0.05 mg/dL), but better in the CRI group (mean R2, 0.57; 75% with final creatinine(actual) - creatinine(predicted) +/- 0.05). The best alternative was the two-phase regression line after exclusion of outliers, which provided similar results across the FAIL, BXAR, and CRI groups (mean R2, 0.71, 0.64, and 0.73; 81%, 84%, and 96% with final creatinine(actual) - creatinine(predicted) +/- 0.05, respectively; P = not significant) and had the best performance in patients with greater nadir creatinine values. The LS method is not suited to measure kidney function loss in renal transplant recipients. Use of the two-phase regression line after exclusion of outliers is a more reliable method in renal transplant recipients, especially patients with impaired baseline function, and has results similar to those of patients with CRI.