GFR Prediction From Cystatin C and Creatinine in Children: Effect of Including Body Cell Mass

BACKGROUND: Aiming to develop a more accurate cystatin C-based model for estimation of glomerular filtration rate (GFR) in children, we hypothesized that inclusion of body cell mass (BCM) would increase the accuracy of the GFR estimate in comparison to a well-established GFR reference method. STUDY DESIGN: Diagnostic test accuracy study. SETTINGS & PARTICIPANTS: 119 children (mean age, 8.8; range, 2.3-14.9 years) referred for GFR measurement by chromium 51 ethylenediaminetetraacetic acid ((51)Cr-EDTA) clearance (mean GFR, 98; range, 13.7-147.4 mL/min/1.73 m(2)). INDEX TEST: GFR estimations by the 2 prediction models resulting from theoretical considerations corroborated by forward stepwise variable selection: GFR (mL/min) = 0.542 × (BCM/SCysC)(0.40) × (height × BSA/SCr)(0.65) and GFR (mL/min) = 0.426 × (weight/SCysC)(0.39) × (height × BSA/SCr)(0.64), where SCysC is serum cystatin C level, BSA is body surface area, and SCr is serum creatinine level. The accuracy and precision of these models were compared with 7 previously published prediction models using random subsampling cross-validation. Local constants and coefficients were calculated for all models. Root mean square error, R(2), and percentage of predictions within ±10% and ±30% of the reference GFR were calculated for all models. Based on 1,000 runs of the cross-validation procedure, median values and 2.5th and 97.5th quantiles of the validation parameters were calculated. REFERENCE TEST: GFR measurement by (51)Cr-EDTA clearance. RESULTS: The BCM model predicted 98% within ±30% of reference GFR and 66% within ±10%, which was higher than for any other model. The weight model predicted 97.5% within ±30% of reference GFR and 62% within ±10%. The BCM model had the highest R(2) and the smallest root mean square error. LIMITATIONS: Included only 9 children with GFR