Longitudinal analysis of performance of estimated glomerular filtration rate as renal function declines in chronic kidney disease.

BACKGROUND: Numerous studies have assessed the accuracy of equations estimating glomerular filtration rate (eGFR) from serum creatinine in individuals with chronic kidney disease (CKD) in cross-sectional studies. Limited literature exists, however, on the consistency of performance of these equations in longitudinal studies as renal function declines.

METHODS: Radionucleotide-measured GFR from 155 predialysis patients with stage 3-5 CKD was compared with eGFR derived from four equations [6-variable Modification of Diet in Renal Disease (6-MDRD), 4-variable MDRD (4-MDRD), Cockcroft-Gault (CG) and Cockcroft-Gault equations corrected for body surface area (CGC)] at baseline, 12 and 24 months. Bias (difference between eGFR and measured GFR) was used as a measure of performance. Restricted Maximum Likelihood (REML) models were used to identify variables potentially affecting the performance of estimating equations across time.

RESULTS: Mean measured GFR (+/-SD) at baseline, 12 and 24 months was 25.9 +/- 10.7, 23.1 +/- 10.6 and 20.3 +/- 10.1 mL/min/1.73 m(2), respectively. There was a statistically significant negative association between bias and GFR for all four estimates (range: -0.76 to -0.71, P < 0.001 for all), indicating worsening underestimation and overestimation at higher and lower GFR, respectively. This negative association significantly reduced over the 24 months (P < 0.001); however, this was largely due to persistent underestimation of eGFR from individuals with GFR >50 mL/min/1.73 m(2). For those with a baseline GFR <50 mL/min/1.73 m(2), the change in bias for any of the four equations over 24 months was
CONCLUSION: GFR estimates are inexpensive and show an acceptable longitudinal performance for monitoring CKD patients with GFR <50 mL/min/1.73 m(2). Inaccuracies appear more substantial above this level of GFR, and care with interpretation is required.