BACKGROUND/AIMS: To predict risk of end-stage renal disease (ESRD) in individual patients with chronic nephropathy. METHODS: Sequential use of univariate analyses and Cox regression to identify risk factors, artificial neural network to quantify their relative importance and Bayesian analysis to address uncertainty of relationships and incorporate ESRD prevalence information in 344 patients with chronic nephropathy enrolled in the Ramipril Efficacy In Nephropathy study. RESULTS: Serum creatinine (SC), 24-hour urinary protein excretion (UPE) and calcium-phosphorus (Ca*P) product were, in this order, the strongest time-adjusted ESRD predictors. Individual risk of ESRD ranged from near zero when SC and UPE were or =2.41 mg/dl, > or =3 g/24 h and > or =32.64 mg2/dl2, respectively. Receiver operating characteristic curves showed that within lowest, middle and highest tertiles of basal SC (0.90-1.65, 1.66-2.40 and 2.41-6.30 mg/dl, respectively) the model accurately predicted ESRD (AUC = 0.80, 0.72 and 0.65; p = 0.0003, 0.0001 and 0.0022, respectively), quality of life or treatment costs. CONCLUSION: Integrated use of regression analysis and probabilistic models allows computation of individual risk of progression to ESRD and related utilities. This may help in optimizing care and costs in nephrology and other medical areas and designing trials in high-risk patients.
|Journal||Nephron - Clinical Practice|
|Publication status||Published - 2003|