Prediction of renal function (GFR) from cystatin C and creatinine in children: Body cell mass increases accuracy of the estimate

AIM: To derive an accurate prediction model for estimating glomerular filtration rate (GFR) in children based primarily on the endogenous renal function marker cystatin C (CysC) and body cell mass (BCM).

THEORY: Cystatin C is produced at a constant rate in all cells of the body and is excreted by glomerular filtration followed by catabolization in the tubular cells. We hypothesized that production rate is proportional to body cell mass (BCM) and inferred GFR (mL/min) to be proportional to BCM/CysC.

MATERIAL AND METHODS: GFR was determined with ⁵¹Cr-EDTA-clearance in 131 children (52 girls, 79 boys) aged 2-14 years (mean 8.8 years). GFR was 14-147 mL/min/1.73m² (mean 97 mL/min/1.73m²). BCM was estimated using bioimpedance spectroscopy (Xitron Hydra 4200). Log-transformed data on BCM/CysC, serum creatinine (SCr), body-surface-area (BSA), height x BSA/SCr, serum CysC, weight, sex, age, height, serum urea and albumin were considered possible explanatory variables using robust regression in a forward, stepwise procedure. GFR (mL/min) was the dependent variable. The accuracy and precision of the prediction model were compared to other prediction models from the literature, using leave-one-out cross validation. Local constants and coefficients were calculated for all models.

RESULTS: New prediction equation GFR (mL/min) = (BCM/CysC)^0.40 x (heightxBSA/SCr)^0.65 x 10.2 could predict 130/131 within 30% of reference GFR, and 87/131 within 10%. This was higher than all other equations, see Table I. The present equation also had the highest R² and the narrowest 95% limits of agreement.

CONCLUSION: The new equation predicts GFR with higher accuracy than other equations. Endogenous methods are, however, still not accurate enough to replace exogenous markers when GFR must be determined with high accuracy.

 

Table I. Bias, 95% limits of agreement (LOA), R², root mean squared error (RMSE) and percentages of estimates within 10% and 30% of measured GFR of log-transformed data.

Reference	Bias (%)	95% LOA (%)	R2	RMSE	Within 30% (%)	Within 10% (%)
Schwartz, 1976	-1.1*	-29.9; 26.8*	0.68	14.2*	95.4	50.4
Bökenkamp, 1998	0.07*	-32.6; 32.8*	0.57	16.61*	89.4	41.7
Filler, 2003	0.12	-29.9; 43.1	0.70	0.181	90.1	42.4
Bouvet, 2006	0.10	-23.1; 30.1	0.93	0.133	96.2	57.2
Zappitelli, 2006	0.03	-20.3; 25.5	0.88	0.115	97.7	61.8
Schwartz, 2008	0.01	-20.4; 25.7	0.88	0.156	96.9	64.1
Present work, 2010	0.02	-18.5; 22.8	0.96	0.104	99.2	66.4

*Not log-transformed