Prediction of the progression of chronic renal failure in children: are current models accurate?

Mathematical models have been used since 1976 in an attempt to predict the progression of chronic renal failure. These models have used the serum creatinine level as either a reciprocal or logarithmic plot against time. The accuracy of such predictions are characterized by either a correlation coefficient (r value) or prediction error (time at which dialysis was predicted minus time at which dialysis actually occurred). Retrospectively, we analyzed 37 children who progressed to end-stage renal disease, grouped them by their respective primary diseases, and calculated both r values and prediction error for each individual and group. Although r values for each group (with the exception of patients less than 1 year of age) were comparable, prediction errors were widely disparate among the various groupings. Individuals within each of the groups had disparate values even though identical r values were present. These observations indicate that predictive models using serum creatinine levels are of limited clinical use and are least useful for patients younger than 1 year of age.