Accuracy to Estimate Rates of Decline in Glomerular Filtration Rate in Renal Transplant Patients

20-year longitudinal follow-up of measured and estimated glomerular filtration rate in kidney transplant patients

BACKGROUND

The slopes of estimated glomerular filtration rate (eGFR) equations are used in the longitudinal follow-up of transplant patients. A 30% reduction in eGFR over 2 years is often used to predict the subsequent risk of mortality or end-stage renal disease. Whether, at the individual level, such changes in eGFR correspond to changes in measured GFR (mGFR) is actually unknown.

METHODS

The performance of serum creatinine-based eGFR equations was compared with mGFR during the longitudinal follow-up of 20 years in a monocentric study of 417 transplanted patients.

RESULTS

The accuracy within 30% for the eGFR equations varied between 70 and 75%. All eGFR equations showed a similar pattern, very like the mGFR time profiles. Individual changes (slopes) of mGFR or eGFR were predictive of graft loss in the next months or years, following the decline in GFR, with no evidence for a difference. However, although the tendency is the same as for mGFR, the percentage of transplant patients with a >30% GFR decrease in the last period before graft loss is significantly lower for eGFR than for mGFR, with discordant results from mGFR in ~25% of the cases.

CONCLUSIONS

All eGFR equations showed similar trends as mGFR, but eGFR predictions may not be very useful at the individual patient level.

The Practical Implications of Using Estimated GFR as the Presumed Reference Variable to Estimate Transplant Chronic Kidney Disease

We determined the proportions of matched kidney transplant isotope GFRs (iGFRs) to the estimated functions (eGFRs) calculated from Isotope Dilution Mass Spectrometry (IDMS), Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI), and Cockcroft-Gault (CG) equations. One thousand four hundred and three iGFR/eGFR pairs on 390 kidney transplant patients were compared considering the iGFR or eGFR as the reference or test variable. Conformity of iGFR to CG estimates demonstrated the least bias of 1.3±18.4 mL/min/1.73 m2 (compared to 1.5±19.4 and - 2.2±19.2 for IDMS and CKDI-EPI, P<0.05) and CKD-EPI estimates the highest precision of 4.1±41.8 (compared to 11.3±43.9 for IDMS and 5.7±37.3 for CG; P<0.05). IDMS eGFR cut off less than 60 and less than 30 mL/min/1.73m2 were correctly matched by iGFR in 79.4% and 49.1% of the times, while CKD-EPI was matched by iGFR in 83.5% and 52.5%. CG was matched in 78.3% and 53.6%. IGFR cut off levels of less than 60, and less than 30 mL/min/1.73m2 were predicted by IDMS in 83.8% and 64.0% of the times. CKD-EPI was correct in 77.8% and 59.0% and CG in 82.5% and 41.6%, respectively. Transplant eGFR results obtained by CKD-EPI or CG are likely to be more precise and less biased than IDMS.

Development and validation of a new statistical model for prognosis of long-term graft function after pediatric kidney transplantation

BACKGROUND

No adequate statistical model has been established to estimate future glomerular filtration rate (GFR) in children after kidney transplantation (KTX). Equations based on simple linear regression analysis as used in adults are not established in children.

METHODS

An optimal prognostic model of GFR was generated for 63 children at 3-7 years after KTX. The main regression model for prediction of the log-transformed GFR (logGFR) included the mean monthly change of GFR in the period 3-24 months after KTX (∆GFR), the baseline GFR at 3 months (bGFR), and an intercept. Additionally, we investigated if the inclusion of cofactors leads to more precise predictions. The model was validated by leave-one-out cross-validation for years 3-7 after KTX. Prognostic quality was determined with the mean squared error (MSE) and mean absolute error (MAE). Results were compared with the simple linear regression model used in adults.

RESULTS

The following statistical model was calculated for every prognosis year (i = 3, …, 7):[Formula: see text] [Formula: see text] Comparison of the new statistical model and the simple linear model for adults led to relevantly lower MSEs and MAEs for the new model (year 7: New model: MSE 0.1, MAE 0.3/adult model: MSE 1069, MAE 18). The benefit of inclusion of cofactors was not relevant.

CONCLUSIONS

This statistical model is able to predict long-term graft function in children with very high precision.

Reliability of GFR formulas based on serum creatinine, with special reference to the MDRD Study equation

Estimation of glomerular filtration rate (GFR) is central to the diagnosis, evaluation and management of chronic kidney disease (CKD). This review summarizes data on the performance of equations using serum creatinine to estimate GFR, particularly the Modification of Diet in Renal Disease (MDRD) Study equation. The size of studies evaluating GFR estimation equations and their level of sophistication in estimating bias, precision, validity and sensitivity to the source population have improved over the past decade. We update our review from 2006, which included 7 studies with over 500 individuals and 12 studies with 50-499 individuals with measured GFR evaluating the MDRD Study and Cockcroft-Gault equations. More recent studies include an individual level pooling analysis of 5504 participants in 10 studies which showed that creatinine calibration to reference methods improved the performance of the MDRD Study equation but increased bias for the Cockcroft-Gault equation. The MDRD Study equation had a bias of 3.0 %, interquartile range of 29.0 % and percentage of estimates within 30 % of the measured GFR value (P(30)) of 82 % for estimates below 60 mL/(min x 1.73 m(2)). Above this value, the bias was greater (8.7 %) and estimates are less useful since 30 % error is a large absolute error in GFR. Results vary across studies but are generally similar with disappointing performance in the high GFR range, which is of particular interest in early diabetic nephropathy. New equations using serum creatinine can reduce the bias present in the high GFR range but are unlikely to dramatically improve precision, suggesting a need for additional markers. Finally, algorithms are needed to tailor clinical practice based on data from GFR estimates and other participant characteristics, including the source population and level of proteinuria.