Health Technology Assessment (Jul 2024)
Accuracy of glomerular filtration rate estimation using creatinine and cystatin C for identifying and monitoring moderate chronic kidney disease: the eGFR-C study
Abstract
Background Estimation of glomerular filtration rate using equations based on creatinine is widely used to manage chronic kidney disease. In the UK, the Chronic Kidney Disease Epidemiology Collaboration creatinine equation is recommended. Other published equations using cystatin C, an alternative marker of kidney function, have not gained widespread clinical acceptance. Given higher cost of cystatin C, its clinical utility should be validated before widespread introduction into the NHS. Objectives Primary objectives were to: (1) compare accuracy of glomerular filtration rate equations at baseline and longitudinally in people with stage 3 chronic kidney disease, and test whether accuracy is affected by ethnicity, diabetes, albuminuria and other characteristics; (2) establish the reference change value for significant glomerular filtration rate changes; (3) model disease progression; and (4) explore comparative cost-effectiveness of kidney disease monitoring strategies. Design A longitudinal, prospective study was designed to: (1) assess accuracy of glomerular filtration rate equations at baseline (n = 1167) and their ability to detect change over 3 years (n = 875); (2) model disease progression predictors in 278 individuals who received additional measurements; (3) quantify glomerular filtration rate variability components (n = 20); and (4) develop a measurement model analysis to compare different monitoring strategy costs (n = 875). Setting Primary, secondary and tertiary care. Participants Adults (≥ 18 years) with stage 3 chronic kidney disease. Interventions Estimated glomerular filtration rate using the Chronic Kidney Disease Epidemiology Collaboration and Modification of Diet in Renal Disease equations. Main outcome measures Measured glomerular filtration rate was the reference against which estimating equations were compared with accuracy being expressed as P30 (percentage of values within 30% of reference) and progression (variously defined) studied as sensitivity/specificity. A regression model of disease progression was developed and differences for risk factors estimated. Biological variation components were measured and the reference change value calculated. Comparative costs of monitoring with different estimating equations modelled over 10 years were calculated. Results Accuracy (P30) of all equations was ≥ 89.5%: the combined creatinine–cystatin equation (94.9%) was superior (p 5%/year difference between mGFR and eGFR. Ability of equations to detect change was studied based on whether or not eGFR detected overall change, or decline only, in mGFR over 3 years against threshold changes variously defined as (1) > 10 ml/minute/1.73 m2; (2) > reference change value (RCV) (a > 21.5% increase or a > 17.7% decrease); (3) > 25% change; and (4) > 25% change and a change in disease stage. Sensitivity and specificity of eGFRs to identify progressive disease were evaluated. Estimated GFRs, in addition to urinary ACR, were also tested as predictors of progression and mortality. In the substudy of disease progression the change in mGFR, and the difference between mGFRs and eGFRs (bias), assessed every 12 months, were modelled over time using a longitudinal linear random coefficients regression model, to estimate average and variability in disease progression and bias. A model of disease progression based on mGFR was developed and differences in progression for risk factors estimated. In the biological variation substudy, analytical (CVA) and individual (CVI) components of variation were calculated and used to derive the RCV for significant changes in serial results for both mGFR and eGFR. Results from the main study informed a measurement model analysis. The trajectory of participants mGFR and eGFR over 10 years was used to estimate the proportion meeting the National Institute for Health and Care Excellence (NICE) definition of accelerated progression or of progression to CKD stage G4, assuming an annual testing schedule, and the number of participants expected to be incorrectly managed at each of the evaluated monitoring time points using different estimating equations. Based on the findings, the comparative costs of monitoring with GFR-estimating equations were calculated. Sample size Main study. Complete baseline data n = 1167. Three-year follow-up GFR data n = 875. Disease progression substudy. n = 278. Biological variation substudy. n = 20. Results All estimates of GFR relating to the primary study objectives were negatively biased compared to mGFR. There was no difference in median bias (ml/minute/1.73 m2) against mGFR between the MDRD (−3.7), CKD-EPIcreatinine (−2.8), CKD-EPIcystatin (−4.1) and CKD-EPIcreatinine-cystatin (−3.9) equations. Accuracy (P30) of the CKD-EPIcystatin equation (89.5%) did not differ from that of the MDRD (89.5%) and CKD-EPIcreatinine (90.2%) equations: accuracy of the CKD-EPIcreatinine-cystatin equation (94.9%) was superior (p 70% concordance. The CKD-EPIcreatinine-cystatin equation had better concordance than the other three primary study equations (p 83% in all cases), sensitivity for detecting change was < 63% in all cases. There was no clear difference in sensitivity or specificity between the four main study equations. For all equations and all thresholds, there was no clear evidence of improved performance of cystatin-containing equations compared to their matched creatinine-only equation. In the substudy of disease progression, modelling data showed a strong association between albuminuria status and rate of progression in mGFR and CKD-EPIcreatinine eGFR, with those with albuminuria having faster progression (steeper decline). Higher baseline mGFR values were associated with faster progression rate for mGFR. African-Caribbean ethnicity increased (slower decline) and South Asian ethnicity decreased (faster decline) the estimate of progression slope for mGFR and CKD-EPIcreatinine GFR. However, recruitment of ethnic minority participants in particular to the substudy fell short of target, limiting the strength of any conclusions. Within-subject biological variation of mGFR was 6.7%, with similar, although in some cases significantly lower, biological variation of eGFR (5.0, 5.3, 5.3 and 5.0% for MDRD, CKD-EPIcreatinine, CKD-EPIcystatin and CKD-EPIcreatinine-cystatin equations, respectively). Derived RCVs (%, positive/negative) were 21.5/−17.7 (mGFR), 15.1/−13.1 (MDRD), 15.9/−13.7 (CKD-EPIcreatinine), 15.9/−13.8 (CKD-EPIcystatin) and 15.1/−13.1 (CKD-EPIcreatinine-cystatin). We observed 62 deaths during the 3-year follow-up period. The study was not powered for hard end points. However, in agreement with earlier studies, regression models including each GFR-estimating equation separately demonstrated mortality was associated with lower eGFR, increasing age and male gender. An association with categorical albuminuria was not observed. There was no evidence of superiority of CKD-EPI equations, including the cystatin C-containing equations, as predictors of death compared to the MDRD equation. A measurement model analysis found no evidence to suggest that any of the estimating equations were superior for identifying CKD progression based on NICE-defined clinical end points. The average incremental per patient costs (compared to MDRD) of monitoring over a 10-year period using the cystatin C-based equations were estimated (CKD-EPIcystatin £42.20; CKD-EPIcreatinine-cystatin £43.32). Conclusions Most GFR equations achieved acceptable accuracy as judged by P30. There was little difference between the equations in accuracy, with evidence of superior accuracy for the CKD-EPIcreatinine-cystatin equation. Across several important characteristics (age, gender, diabetes, albuminuria, BMI, GFR level) we found no difference in accuracy of GFR-estimating equations. In relation to GFR estimation in African-Caribbean individuals, there was evidence to suggest caution should be exercised before advocating simple removal of the black race factor from the original CKD-EPI equations. In the longitudinal study, the CKD-EPIcreatinine-cystatin displayed slightly better concordance with mGFR than the other main study equations when tracking patients, but all study equations underestimated the mGFR decline. The sensitivity of GFR equations to detect clinically relevant threshold changes in mGFR, either overall or when considering decline in GFR only, was ≤ 63% for all equations. This is of concern given that such thresholds, including the NICE definition of accelerated progression and the change recognised as being true as determined by biological variation (RCV), were studied. Overall, data comparing the accuracy of different GFR-estimating equations demonstrated no notable benefit of using a cystatin C-containing equation in detecting GFR change. The measurement model underpinning the health economic analysis focused on the comparative accuracy of the estimating equations to detect accelerated progression. The analysis estimated accuracy over a longer trajectory than the main study and factored in measurement error, but found no clear benefit of using a cystatin C-based estimating equation. There was therefore no evidence to suggest that adding cystatin C measurement to current GFR monitoring protocols would be cost-effective. The disease progression modelling of the substudy data noted faster progression associated with higher baseline GFR and albuminuria; the latter consistent with other studies. Any conclusions relating to the influence of ethnicity were tempered by poor recruitment of ethnic minority individuals. The biological variability data have implications for monitoring of patients with CKD and clinical ability to understand CKD progression, both in clinical practice and research. The information presented provides an evidence base allowing clinicians to have meaningful discussions with their patients about the implications of changes in their GFR results. Inclusion of cystatin C in GFR-estimating equations was associated with marginal improvements in accuracy, but no clear advantages in terms of detecting GFR change over time. Problems of standardisation of cystatin C assays remain, despite the introduction of an international standard. The use of cystatin C increases the economic cost of CKD monitoring with little apparent gain. These data do not support the use of cystatin C for the routine monitoring of GFR in people with stage 3 CKD. Further research is warranted to investigate specific patient groups that may benefit from cystatin C use. Trial registration This trial is registered as ISRCTN42955626 www.controlled-trials.com/ISRCTN42955626 (accessed 26 July 2023). Funding This award was funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment programme (NIHR award ref: 11/103/01) and is published in full in Health Technology Assessment; Vol. 28, No. 35. See the NIHR Funding and Awards website for further award information.
Keywords
