Glomerular filtration rate estimation in nephrotic patients: creatinine and/or cystatin-C-based equations, or none of them? Authors: Fabiola Carrara, Serenella Valaperta, Mariagrazia Alessio, Matias Trillini, Alessia Gennarini, Aura Crispino, Alessandro Villa, Tobia Peracchi, Daniela Cugini, Nadia Stucchi, Ilaria Pagani, Giuseppe Remuzzi, Piero Ruggenenti Published: 2026 ABSTRACT ABSTRACTBackgroundThe performance of serum-creatinine- and/or cystatin-C-based chronic kidney disease -Epidemiology (CKD-Epi) Collaboration and European Kidney Function Consortium (EKFC) glomerular filtration rate (GFR) estimates (creatinine-based GFR estimates (eGFRcr), cystatin-C-based GFR estimates (eGFRcys), and eGFRcr + cys, respectively) in nephrotic patients with CKD is unknown.MethodsWe retrospectively evaluated the performance of all the six aforementioned equations in 100 blood samples collected 1-year apart from 50 consenting patients with primary membranous nephropathy (MN) and proteinuria >3.5 g/24 h who had their GFR simultaneously measured with the iohexol plasma clearance technique during three intervention studies. The study had >85% power to detect a significant difference between actual and reference (0.90) concordance correlation coefficient (CCC).ResultsWith CKD-Epi and EKFC, eGFRcr values overestimated, whereas eGFRcys values underestimated (P < 0.001 for all) measured GFR (mGFR). eGFRcr + cys bias was significant (P < 0.05) in hypo-albuminemic participants. Biases significantly increased with worsening hypoalbuminemia, but independently of proteinuria. In normo-albuminemic participants, CCC did not differ from the 0.90 reference value and total deviation index (TDI) was <30 only with CKD-Epi and EKFC eGFRcr + cys equations. The error exceeded 10% in >40% of estimates. Over one year, mGFR increased by 4.8 ml/min/1.73 m2 (P = 0.008). This change was not captured by any equation.ConclusionsThe performance of GFR estimation equations is acceptable only when creatinine and cystatin-C are simultaneously used as filtration markers in nephrotic patients without hypoalbuminemia. Direct GFR measurement is needed for accurate kidney function evaluation in those with hypoalbuminemia, and whenever individual clinical decision makings (including dosing of toxic medications with renal clearance) or evaluations of treatment effects on GFR changes over time are needed in clinics or research. What was known: Serum-creatinine-based equations such as chronic kidney disease CKD-Epidemiology (CKD-Epi) and European Kidney Function Consortium (EKFC) equations have been developed for the estimation of GFR (eGFRcr) in healthy subjects or patients with CKD, but their performance is limited by the fact that creatinine is not a reliable marker of glomerular filtration. Serum-creatinine-based equations such as chronic kidney disease CKD-Epidemiology (CKD-Epi) and European Kidney Function Consortium (EKFC) equations have been developed for the estimation of GFR (eGFRcr) in healthy subjects or patients with CKD, but their performance is limited by the fact that creatinine is not a reliable marker of glomerular filtration. The limitations of these equations were reduced with the use of another endogenous marker of glomerular filtration such as cystatin-C (eGFRcys) and their performance was particularly improved by the combined use of both biomarkers in the same equation for GFR estimation (eGFRcr + cys). The limitations of these equations were reduced with the use of another endogenous marker of glomerular filtration such as cystatin-C (eGFRcys) and their performance was particularly improved by the combined use of both biomarkers in the same equation for GFR estimation (eGFRcr + cys). A recent study found that both CKD-Epi and EKFC equations can provide acceptable GFR estimates also in CKD patients with “moderate-to-severe proteinuria” without further compromising accuracy or introducing additional bias. However, whether and to what extent this applies also to patients with nephrotic syndrome is unknown. A recent study found that both CKD-Epi and EKFC equations can provide acceptable GFR estimates also in CKD patients with “moderate-to-severe proteinuria” without further compromising accuracy or introducing additional bias. However, whether and to what extent this applies also to patients with nephrotic syndrome is unknown. This study adds: The performance of both EKFC and CKD-Epi equations was sub-optimal in estimating the GFR directly measured with the iohexol plasma clearance technique. Serum-creatinine-based equations largely overestimated, whereas cystatin-C based equations largely underestimated measured GFR values and the biases of the estimates progressively increased in parallel with the worsening of hypoalbuminemia. The performance of both EKFC and CKD-Epi equations was sub-optimal in estimating the GFR directly measured with the iohexol plasma clearance technique. Serum-creatinine-based equations largely overestimated, whereas cystatin-C based equations largely underestimated measured GFR values and the biases of the estimates progressively increased in parallel with the worsening of hypoalbuminemia. In patients with nephrotic syndrome but without hypoalbuminemia, the limitations of both EKFC and CKD-Epi equations were mitigated when serum creatinine and cystatin-C levels were considered in the same equation for GFR estimation (eGFRcr + cys). In patients with nephrotic syndrome but without hypoalbuminemia, the limitations of both EKFC and CKD-Epi equations were mitigated when serum creatinine and cystatin-C levels were considered in the same equation for GFR estimation (eGFRcr + cys). In patients with nephrotic syndrome and hypoalbuminemia, the performance of both equations was sub-optimal even when serum creatinine and cystatin-C were used in combination. All GFR estimations missed kidney function changes over time emerging from direct GFR measurements in any patient with nephrotic syndrome independent of serum albumin levels. In patients with nephrotic syndrome and hypoalbuminemia, the performance of both equations was sub-optimal even when serum creatinine and cystatin-C were used in combination. All GFR estimations missed kidney function changes over time emerging from direct GFR measurements in any patient with nephrotic syndrome independent of serum albumin levels. Potential impact: Independent of the used endogenous markers of glomerular filtration and the considered GFR estimation equations, previous evidence of acceptable performance of GFR estimates in patients with sub-nephrotic proteinuria cannot be extended to patients with nephrotic range proteinuria, particularly in those with hypoalbuminemia. Independent of the used endogenous markers of glomerular filtration and the considered GFR estimation equations, previous evidence of acceptable performance of GFR estimates in patients with sub-nephrotic proteinuria cannot be extended to patients with nephrotic range proteinuria, particularly in those with hypoalbuminemia. The use of GFR estimates would affect monitoring of particularly frail patients, dosing of toxic medications with renal clearance and narrow therapeutic windows, and data variability and statistical power of trials, resulting in falsely negative findings that could disrupt the clinical development of innovative, potentially nephroprotective medications. The use of GFR estimates would affect monitoring of particularly frail patients, dosing of toxic medications with renal clearance and narrow therapeutic windows, and data variability and statistical power of trials, resulting in falsely negative findings that could disrupt the clinical development of innovative, potentially nephroprotective medications. The performance of GFR estimation equations is acceptable only when creatinine and cystatin-C are simultaneously used as filtration markers in nephrotic patients without hypoalbuminemia. Direct GFR measurement is needed for accurate kidney function evaluation in patients with hypoalbuminemia and whenever individual clinical decision-making (including dosing of toxic medications with renal clearance) or evaluations of treatment effects on GFR changes over time are needed in clinical practice or research. The performance of GFR estimation equations is acceptable only when creatinine and cystatin-C are simultaneously used as filtration markers in nephrotic patients without hypoalbuminemia. Direct GFR measurement is needed for accurate kidney function evaluation in patients with hypoalbuminemia and whenever individual clinical decision-making (including dosing of toxic medications with renal clearance) or evaluations of treatment effects on GFR changes over time are needed in clinical practice or research. INTRODUCTION Serum-creatinine-based equations such as the Chronic Kidney Disease Epidemiology Collaboration (CKD-Epi) and the European Kidney Function Consortium (EKFC) equations are considered gold standard for GFR estimation and CKD staging [1–6]. Unfortunately, the performance of these equations is limited by the fact that creatinine (cr) is not a reliable marker of glomerular filtration because, after glomerular ultrafiltration, it is mainly secreted at tubular levels. Tubular handling of creatinine can change in different clinical conditions and individual patients, which further affects the accuracy and reproducibility of creatinine-based GFR estimates (eGFRcr) [7]. In addition, the renal transport of creatinine can also be affected by protein tubular traffic in patients with proteinuric CKD [8], that is, in those patients who are most in need of accurate GFR evaluations [9–11]. Although the effect of tubular handling should be inherently incorporated into these equations, there is a general trend to GFR overestimation in patients with renal insufficiency [12, 13] or underestimation in those with glomerular hyperfiltration [14, 15]. In addition, muscle mass, diet, inflammation, and medications, may influence serum creatinine levels independently of kidney function [16]. In a recent, retrospective longitudinal study in 352 patients with proteinuric (24-h urinary protein excretion >1 g/24 h) CKD included in the Ramipril Efficacy In Nephropaty (REIN) 1 and 2 trials [17, 18] and longitudinally monitored with serial measurements of GFR by the Iohexol plasma clearance technique [19], Zoccali et al. [8] reported that the overall performance of CKD-Epi and EKFC equations in CKD patients with clinical proteinuria was similar to that in studies that established the CKD-Epi 2009 [1] and 2021 [3] and EKFC [5] formulas. Thus, these equations can also be applied to CKD patients with “moderate-to-severe proteinuria” [8] without further compromising accuracy or introducing additional bias. However, whether and to what extent this applies also to patients with nephrotic range proteinuria is unknown. Cystatin-C (cys-C) is a non-glycosylated, basic protein encoded by the CST3 gene that is found in virtually all nucleated cells [20, 21]. Cys-C is generated at a mostly constant rate compared to creatinine, is freely filtered through the glomerular barrier, and fully reabsorbed and catabolized in the proximal tubule of the nephron [22]. Moreover, it is not affected by body habitus and muscle mass. Thus, cys-C appears to be a more accurate endogenous measure of kidney function than creatinine [23, 24]. However, despite improved accuracy [3, 25–28], the eGFRcys equation showed a tendency to underestimate directly measured GFR (mGFR). Although other investigators have found the EKFC with cys-C to be essentially unbiased, particularly in external validation cohorts [6], eGFRcys has not been specifically evaluated in patients with nephrotic syndrome. To mitigate the systematic overestimation of eGFRcr and the underestimation of eGFRcys equations, combined equations (eGFRcr + cys) have been developed to balance these two opposite trends. This reduced bias and improved the accuracy of GFR estimations, particularly in patients with impaired kidney function [24, 29, 30]. Again, however, the performance of eGFRcr + cys equations in estimating GFR in nephrotic patients is unknown. To address this issue, in the present study, we took advantage of a cohort of 50 patients with primary membranous nephropathy, nephrotic-range proteinuria, with or without hypoalbuminemia, who had their GFR measured by the iohexol plasma clearance technique [19] and concomitantly estimated by the use of eGFRcr-, eGFRcys-, and eGFRcr + cys-based CKD-Epi and EKFC equations in two visits one year apart. MATERIALS AND METHODS Study population This retrospective, observational study (registered at: NCT06797609) included 50 subjects with primary membranous nephropathy who had been enrolled in three previous clinical trials (registered at: NCT04095156, NCT04893096, and NCT05050214, respectively) aimed to assess the efficacy of first- and second-generation anti-CD20 monoclonal antibodies (rituximab and obinutuzumab) or an anti-CD38 monoclonal antibody (felzartamab) therapy in inducing remission of nephrotic syndrome (NS) during 1-year follow-up, that were performed and coordinated at the “Aldo e Cele Daccò” Clinical Research Center for Rare Diseases of the Istituto di Ricerche Farmacologiche Mario Negri IRCCS at Villa Camozzi (Bergamo, Italy) in cooperation with the Unit of Nephrology of the Azienda-Socio-Sanitaria-Territoriale (ASST) Papa Giovanni XXIII (Bergamo, Italy). A total of 100 serum samples, 50 at baseline and 50 at 1 year, were collected for cr and cys-C measurement from patients with NS who had provided consent for sample storage. Samples were maintained at the certified biobank of the Clinical Research Center, UNI EN ISO 9001:2015; certification no. 6121–Centro di Risorse Biologiche Mario Negri—Biobanca Malattie Rare e Malattie Renali (CRB). Both cr and cys-C aliquots were stored at −80°C to preserve their integrity for extended periods [31]. The study protocol was revised and approved by the Local Ethical Committee (Comitato Etico Lombardia 6 n. approvazione 2025–3.11/112 data 11 March 2025). All participants provided written informed consent in compliance with the Declaration of Helsinki. The study was conducted according to good clinical practice guidelines. Analytical methods Cr and cys-C levels were measured on the Siemens Atellica CH analyzer (Siemens Healthcare s.r.l, Milan, Italy). The coefficient of variation (CV), reflecting measurement reproducibility, was ≤5% and ≤1,7 for Cr and cys-C, respectively. The Atellica CH assay for both cr and cys-C was aligned to International Federation of Clinical Chemistry (IFCC) recommendations. GFR was estimated with the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI 2009) formula [26] and EKFC [32] based on serum cr, cys-C, or a combination of both markers. Proteinuria and albuminemia were quantified using an automated AU480 chemistry analyzer (Beckman Coulter, Milan, Italy), following the manufacturer’s standard protocols. Measured glomerular filtration rate On the morning of renal function assessment, each patient received an intravenous injection of 5 ml of iohexol solution (Omnipaque 300, GE Healthcare, Milan, Italy). Blood samples were collected at 120, 150, 180, 210, and 240 min after injection in subjects with eGFR >40 ml/min/1.73 m2 and at 120, 180, 240, 300, 360, 420, and 480 min in subjects with eGFR ≤40 ml/min/1.73 m2. Postinjection plasma iohexol concentrations were quantified at the laboratory of the Clinical Research Center using high-performance liquid chromatography (HPLC, Agilent Technologies, Milan, Italy) as previously described [19]. The laboratory participated in an external quality assessment (EQA) scheme administered by the accredited provider Equalis, in which plasma samples containing predetermined concentrations of iohexol were distributed to participating laboratories for assessment of analytical performance. Internal quality control measures were also performed routinely [33]. Pharmacokinetic analysis was performed using a one-compartment open model and the resulting clearance value was subsequently corrected using the Bröchner–Mortensen equation [34] and normalized to a standard body surface area of 1.73 m² [35]. On the morning of each iohexol clearance test, demographic and anthropometric variables required for the CKD-EPI [26] and EKFC [32] equations were also recorded. Statistical analysis Agreement between each eGFR equation and mGFR was evaluated using linear mixed-effects models [36]. These models included the eGFR equation as a fixed effect and a random intercept for each subject to account for repeated measurements and for subjects contributing observations across subgroups. This approach was adopted to appropriately compare mean differences between subgroups while accounting for within-subject correlation. The primary outcome was the concordance correlation coefficient (CCC), with a value ≥0.90 considered indicative of optimal agreement. Secondary outcomes were bias, defined as the mean difference between eGFR and mGFR (eGFR—mGFR), and the total deviation index (TDI) at the 90th percentile. TDI was a measure that captured a large proportion of data within a boundary for allowed differences between 2 measurements [37]. An acceptable level of agreement was predefined as a TDI <30%. Variance components estimated from the mixed-effects models were used to compute the CCC and TDI, along with conditional R², reflecting the proportion of variance explained by both fixed and random effects [36–39]. The proportions of eGFR values within ±10% and ±30% of mGFR (P10 and P30) were computed as descriptive measures and displayed using target plots to visualize the distribution of estimation errors across equations. The reproducibility of the iohexol plasma clearance was 6.28% [40]. Pairwise comparisons between equations were performed using bootstrap resampling (1000 replicates), with subjects resampled with replacement to preserve within-subject correlation. For each bootstrap sample, mixed-effects models were refitted and agreement metrics recalculated; empirical P-values were derived from the bootstrap distribution of paired differences. Bland–Altman plots and scatter plots were used for graphical assessment of agreement [41]. Data were summarized as mean ± standard deviation or number (%), as appropriate. Differences between baseline and 1-year values were assessed using paired t-tests or Wilcoxon signed-rank tests, as appropriate. The normal distribution of data was evaluated using the Shapiro–Wilk test. All analyses were repeated within predefined subgroups based on serum albumin levels (<3.5 and ≥3.5 g/dl). Statistical analyses were performed using R (version 4.1) with the lme4 package for mixed-effects modeling, SAS version 9.4, and GraphPad Prism version 10 for graphical outputs. All tests were two-sided, with a significance level of α = 0.05. Sample size A post-hoc power analysis was conducted using bootstrap-based simulations to assess the ability to detect deviations of the CCC from a predefined threshold of 0.90 across varying effective sample sizes. A total of 50 patients would provide the study with ∼85% power to detect a statistically significant deviation of different eGFR equations from the 0.90 threshold. RESULTS A total of 50 adult Caucasians (38 males) with biopsy-proven MN, slightly reduced GFR, and proteinuria >3.5 g/24 h were included. At inclusion, 43 (86%) had concomitant hypoalbuminemia, defined as serum albumin level <3.5 g/dl (Table 1). All of them had previously been exposed to several immunosuppressive medications, including steroids, alkylating agents, and calcineurin inhibitors in different combinations and different anti-CD20 and anti-CD38 monoclonal antibodies. Measured GFR increased significantly (P < 0.01) from baseline to 1 year, whereas no eGFR estimate changed appreciably during the same observation period (Fig. 1 and Table 2). Serum albumin concentration significantly increased (P < 0.001), and the number (percentage) of hypo-albuminemic participants significantly (P < 0.0001) decreased to 14 (28%). Other changes, including the significant (P < 0.01) increase in hemoglobin concentration, paralleled amelioration of the nephrotic syndrome (Table 2). Changes in mGFR between baseline and 1-year evaluation and concomitant changes in different eGFRs obtained with different serum-creatinine- and/or cystatin-C-based CKD-Epi and EKFC equations. Only changes in mGFR were statistically significant. Values are presented as mean ± standard error of the mean (SEM). Table 1:Baseline characteristics of the patients.Overalln = 50 Demographic characteristics Age (years)58.0 ± 10.6Male sex, n (%)38 (76.0)Caucasian, n (%)50 (100.0)Body weight (kg)79.58 ± 13.67Height (cm)171.20 ± 8.53Body-mass index (kg/m2)27.09 ± 4.23 Clinical characteristics Systolic BP (mmHg)127.92 ± 11.89Diastolic BP (mmHg)76.90 ± 7.40Mean BP (mmHg)93.91 ± 8.07 Laboratory parameters Serum Creatinine (mg/dl)1.14 ± 0.41Serum Cystatin C (mg/l)1.38 ± 0.42Serum Albumin (g/dl)2.82 ± 0.57Serum Albumin <3.5 g/dl—n (%)43 (86%)Hemoglobin (g/dl)13.04 ± 1.43 Urine parameters Proteins (g/24 h)5.91 [4.48–7.83]P/C (mg/g)3910 [3039–5541] Kidney function parameters GFR (ml/min/1.73 m2)62.4 ± 17.1CKD-EPI (ml/min/1.73 m2)eGFRcr73.0 ± 23.6eGFRcys57.4 ± 20.9eGFRcr + cys64.8 ± 22.3EKFC (ml/min/1.73 m2)eGFRcr70.3 ± 21.7eGFRcys58.5 ± 18.5eGFRcr + cys64.4 ± 17.1 Previous Immunosuppressive treatments Steroid therapy23 (46)Calcineurin inhibitors11 (22)Alkylating agents15 (30)Rituximab39 (78)Felzartamab1 (2)Ofatumumab8 (16.0)Daratumumab1 (2) Treatment at baseline Monoclonal antibody Rituximab22 (44.0)Obinutuzumab19 (38.0)Felzartamab9 (18.0) Concomitant treatment ACEi32 (64)ARBs31 (62)ACEi + ARBs42 (84)SGLT2i2 (4)MRA1 (2)SGLT2i + MRA + GLP-13 (6)Statins23 (46)Data are mean ± SD or median with interquartile range (square brackets) or numbers and percentages (round brackets).BP, blood pressure. Baseline characteristics of the patients. Data are mean ± SD or median with interquartile range (square brackets) or numbers and percentages (round brackets). BP, blood pressure. Table 2:Baseline and 1-year characteristics of the patients.Baselinen = 501 yearn = 50 Demographic characteristics Body weight (kg)79.58 ± 13.6779.68 ± 14.35Body-mass index (kg/m2)27.09 ± 4.2327.16 ± 4.58 Clinical characteristics Systolic BP (mmHg)127.92 ± 11.89126.16 ± 12.06Diastolic BP (mmHg)76.90 ± 7.4075.78 ± 8.51Mean BP (mmHg)93.91 ± 8.0792.57 ± 9.07 Laboratory parameters Serum Creatinine (mg/dl)1.14 ± 0.411.08 ± 0.41Serum Cystatin C (mg/l)1.38 ± 0.421.31 ± 0.39Serum Albumin (g/dl)2.82 ± 0.573.68 ± 0.48**Serum Albumin <3.5 g/dl—n (%)43 (86%)14 (28)***Hemoglobin (g/dl)13.04 ± 1.4313.53 ± 1.58* Urine parameters Proteins (g/24 h)5.91 [4.48–7.83]1.67 [0.51–3.53]**P/C (mg/g)3910 [3039–5541]579 [237–1587]** Kidney function parameters mGFR (ml/min/1.73 m2)62.4 ± 17.167.2 ± 20.0*CKD-EPI (ml/min/1.73 m2)eGFRcr73.0 ± 23.675.6 ± 21.9eGFRcys57.4 ± 20.960.9 ± 22.2eGFRcr + cys64.8 ± 22.368.2 ± 22.5EKFC (ml/min/1.73 m2)eGFRcr70.3 ± 21.772.5 ± 20.0eGFRcys58.5 ± 18.561.3 ± 18.9eGFRcr + cys64.4 ± 17.166.9 ± 19.0Data are mean ± SD or median with interquartile range (square brackets). .Paired-t test and Wilcoxon signed-rank test used for normally and non-normally distributed variables as necessary to calculate baseline versus 1 year *P < 0.01, **P < 0.001. McNemar test used for proportions ***P < 0.0001 versus baseline versus 1 year. Baseline and 1-year characteristics of the patients. Data are mean ± SD or median with interquartile range (square brackets). . Paired-t test and Wilcoxon signed-rank test used for normally and non-normally distributed variables as necessary to calculate baseline versus 1 year *P < 0.01, **P < 0.001. McNemar test used for proportions ***P < 0.0001 versus baseline versus 1 year. All GFR estimates obtained with different equations, different tracers, and with their combination significantly correlated with mGFRs. The intercepts of both CKD-Epi and EKFC cr + cys combined equations showed the lowest deviations from the lines of identity (Fig. S1, Panels C and F) and the narrowest confidence intervals of the Bland and Altman plots (Fig. S2, Panels C and F). Bias CKD-Epi eGFRcr significantly (P < 0.001) overestimated mGFR by 9.64 ml/min/1.73 m2 in the whole study group and the overestimation was almost double in hypo-albuminemic than in normo-albuminemic patients. Conversely, CKD-Epi eGFRcys significantly (P < 0.001) underestimated mGFR by −5.52 ml/min/1.73 m² in the whole study group, and the underestimation was almost double in normo-albuminemic than in the hypo-albuminemic patients. The bias of the CKD-Epi eGFRcr + cys equation was not significant in the whole study group and was virtually zero in normo-albuminemic patients, whereas the combined equation slightly overestimated mGFR in hypo-albuminemic patients (Fig. 2, Panel A). Biases of different EKFC equations in the whole study groups, and in hypo- and normo-albuminemic patients considered separately where quite similar to those of CKD-Epi eGFR equations (Fig. 3, Panel A). Overall, the lowest bias was observed with either CKD-Epi and EKFC scr + cys combined equations in all considered groups (Fig. 2, Panel A and Fig. 3, Panel A). Forest plot of bias (Panel A), CCC (Panel B), TDI (Panel C), and P30 (Panel D) for CKD-Epi equation in the whole study group (Overall), and in hypo-albuminemic (serum albumin <3.5 g/dl) and normo-albuminemic (serum albumin ≥3.5 g/dl) patients considered separately. Values are presented as mean and 95% bootstrap confidence interval (CI). Forest plot of bias (Panel A), CCC (Panel B), TDI (Panel C), and P30 (Panel D) for EKFC equation in the whole study group (overall), and in hypo-albuminemic (serum albumin <3.5 g/dl) and normo-albuminemic (serum albumin ≥3.5 g/dl) patients considered separately. Values are presented as mean and 95% bootstrap confidence interval (CI). Notably, with all equations, the bias was negatively correlated with serum albumin levels; thus, the largest biases were observed in association with the lowest albumin levels, whereas the biases ameliorated in parallel with amelioration of hypoalbuminemia, and the lowest biases were observed in association with normoalbuminemia (Fig. 4, Panels A, B, and C). Conversely, for all the serum-creatinine- and/or cys-C-based equations with considered markers, the bias was independent of the severity of proteinuria (Fig. 4, Panels D, E, and F). Superimposable findings were observed with the EKFC eGFR equations (Fig. 5). Relationship between bias and serum albumin (Panels A, B. and C) or 24-h proteinuria (Panels D, E, and F) for CKD-Epi GFR estimates. With all equations, bias significantly correlated with serum albumin levels (larger bias was associated with more severe hypoalbuminemia), whereas there was no correlation with any considered equation between bias and severity of proteinuria. Relationship between bias and serum albumin (Panels A, B, and C) or 24-h proteinuria (Panels D, E, and F) for EKFC GFR estimates. With all equations, bias significantly correlated with serum albumin levels (larger bias was associated with more severe hypoalbuminemia), whereas there was no correlation with any considered equation between bias and severity of proteinuria. Concordance correlation coefficient CCC was significantly lower than 0.90 with all considered equations in the whole study group (Fig. 2, Panel B and Fig. 3, Panel B). The worst performance was observed for both CKD-Epi and EKFC eGFRcr equations. Again, the less poor CCC was observed with cr + cys equations. CCC was significantly <0.90 with all equations in both normo- and hypo-albuminemic patients considered separately, with the only exceptions for both CKD-Epi and EKFC cr + cys combined equations (Fig. 2, Panel B and Fig. 3, Panel B and Table S1). Total deviation index TDI tended to be worse in all considered groups with CKD-Epi than with EKFC equations (Fig. 2, Panel C and Fig. 3, Panel C). TDI was close to 30 with both cr + cys combined equations in the whole study group and in hypo-albuminemic patients. Both CKD-Epi and EKFC cr + cys equations were significantly lower than 30 in hypo-albuminemic patients (Fig. 2, Panel C and Fig. 3, Panel C). P30–P10 Regardless of the equation (CKD-EPI or EKFC) and filtration marker used (creatinine, cys-C, or both), P30 values ranged from 84% to 96% (Fig. 2 Panel D and Fig. 3, Panel D), indicating that the majority of eGFR estimates were within 30% of the mGFR. The highest P30 values were observed with equations combining creatinine and cys-C (Fig. 6). However, the eGFR values with <10% error were always less than half of all measurements. With both combined equations, the percentage of eGFRs with <10% errors ∼60%. Thus, whatever the used equation, the error exceeded 10% in at least 40% of estimates (Fig. 6). Target plot illustrating the mean % error (accuracy) of eGFR equations compared with mGFR. The plots show the mean % error of eGFR versus mGFR and the proportion (n,%) of estimates falling within the ±10% (P10) and ±30% error range (P30). The circle identifies the determinations that fall in P10. Panels A–C refer to the CKD-Epi equation (Panel A: eGFRcr, Panel B: eGFRcys, Panel C: eGFRcr + cys), and Panels D–F to the EKFC equation (Panel D: eGFRcr, Panel E: eGFRcys, Panel F: eGFRcr + cys). Points are randomly distributed within the plots primarily for visualization purposes to avoid overlap between observations. The relevant information is the radial distance from the center: points located within the inner circle correspond to eGFR values within ±10% of mGFR, whereas points outside the circle indicate larger deviations. DISCUSSION In this longitudinal observational cohort-study, we found that in patients with CKD and the nephrotic syndrome: The performance of both EKFC and CKD-Epi equations was similarly sub-optimal. The performance of both EKFC and CKD-Epi equations was similarly sub-optimal. Serum-creatinine-based equations largely overestimated, and cys-C-based equations largely underestimated actual GFR. The biases of the estimates progressively increased in parallel with the worsening of hypoalbuminemia. Serum-creatinine-based equations largely overestimated, and cys-C-based equations largely underestimated actual GFR. The biases of the estimates progressively increased in parallel with the worsening of hypoalbuminemia. The performance of CKD-Epi and EKFC creatinine and cys-C combined equations was acceptable in nephrotic patients without hypoalbuminemia. The performance of CKD-Epi and EKFC creatinine and cys-C combined equations was acceptable in nephrotic patients without hypoalbuminemia. The performance of any considered equation was sub-optimal in all nephrotic patients with hypoalbuminemia. The performance of any considered equation was sub-optimal in all nephrotic patients with hypoalbuminemia. All GFR estimation equations failed to capture kidney function changes over time that emerged from direct GFR measurements. All GFR estimation equations failed to capture kidney function changes over time that emerged from direct GFR measurements. Notably, over 1 year, the significant increase in GFR was paralleled by a significant and clinically relevant amelioration of hypoalbuminemia and reduction in proteinuria. Considering that with both equations GFR overestimation decreased in parallel with the increase in serum albumin (independent of changes in proteinuria), it is conceivable that the increase in serum albumin levels observed at 1 year of follow-up compared to baseline translated into an artifactual decrease in (over) estimated GFR that offset the concomitant actual increase in directly measured GFR. This could explain why all equations not only poorly estimated GFR values, but also did not capture the increase in measured GFR observed at 1 year compared to baseline. Importantly, this interpretation is unlikely to be explained by limitations in the reference method itself when applied to nephrotic participants. Pharmacokinetic studies have consistently shown that iohexol exhibits negligible protein binding, with only ∼1.5% of circulating iohexol bound to plasma proteins [42]. Therefore, the substantial changes in serum albumin concentration observed during follow-up unlikely materially affected iohexol clearance measurements, supporting the conclusion that the observed discordance between measured and estimated GFR changes primarily reflected limitations of creatinine- or cys-C-based estimation equations in nephrotic patients, particularly in those with hypoalbuminemia. Thus, in hypo-albuminemic nephrotic patients, treatment-induced amelioration of kidney function could be largely missed with GFR estimation equations based on endogenous markers of glomerular filtration. On the other hand, the use of sub-optimal GFR estimates would lead to wide data variability and systematic biases that would inevitably affect the power of statistical analyses. Therefore, the strong limitations of these equations could result in falsely negative studies that could even preclude the clinical development of innovative medications that could modify the course of this severe clinical condition. On the other hand, the wide and unpredictable fluctuations of the GFR estimates result in poor precision and accuracy, which did not allow us to introduce correction factors to address the albumin-dependent GFR biases. Thus, previous evidence of acceptable performance of GFR estimates in patients with sub-nephrotic proteinuria [8] cannot be extended to patients with nephrotic range proteinuria, particularly in those with hypoalbuminemia. Seminal study showed that in the nephrotic syndrome, creatinine clearance largely exceeds actual GFR—even more than in the “average” CKD patient—because of an increased creatinine tubular secretion [13] that exceeds the compensatory tubular over-secretion observed in patients with renal insufficiency [43]. A subsequent survey by Branten et al., [44] showed that the Modification for Diet in Renal Diseases study (MDRD) equation overestimated the GFR in 42 patients with nephrotic syndrome. However, it did not evaluate the performance of CKD-Epi and EKFC equations, that have been recently introduced in clinical practice to overcome the several limitations of the MDRD equations. Subsequently, discrepancies between GFR estimated using the MDRD formulas compared with the CKD-EPI formula and cys-C- and β-2-microglobulin-based equations taken as surrogates of GFR were registered in a series of one hundred forty-two patients with glomerular diseases [45]. In this study, however, the GFR was not directly measured. Our data showed that nephrotic syndrome was associated with an increase also in cys-C-based GFR estimates. Indeed, the nephrotic syndrome is characterized by marked extracellular and intravascular volume expansion due to kidney sodium and water retention, a process that in more severe cases may result in – or contribute to – hypoalbuminemia, with further hemodilution and further GFR overestimation [46–48]. Thus, on one hand, hemodilution contributes to serum creatinine levels reduction with further amplification of GFR overestimation, whereas, on the other hand, it counterbalances GFRcys-C underestimation resulting from increased cys-C levels associated with inflammation, metabolic disturbances, corticosteroid exposure, and thyroid dysfunction [30]. However, hemodilution alone did not fully explain the observed discrepancy between mGFR and eGFR in patients with nephrotic syndrome. While acute changes in plasma volume may transiently lower serum concentrations of creatinine and cys-C, chronic volume overload is expected to result in a new steady state in which biomarker concentrations are largely restored through adjustments in filtered load and biomarker kinetics. Therefore, other additional mechanisms may contribute to the discrepancy between eGFR and mGFR in nephrotic syndrome. In particular, reduced muscle mass and decreased creatinine generation may affect creatinine-based estimates, whereas alterations in the production, distribution, or non-renal handling of both filtration markers may also play a role. Explorative analysis EKFC equations have been shown to perform better than the CKD-Epi equations in particular populations at the extremes of age, particularly among younger adults (<30 years) and the elderly (>75–80 years) [49]. However, our cohort was largely composed of middle-aged individuals, with a median (IQR) age of 60 (51–66) years with an age ranging between 23 and 78 years. Thus, age extremes were not represented and this conceivably explained why all analyses showed similar suboptimal performance of both equations, not only for the primary outcome (CCC), but also for bias, P10, TDI, and Bland–Altman that consistently converged to indicate similarly sub-optimal performance of both equations. Limitations and strengths This was a retrospective analysis of data recorded for other purposes in a relatively small cohort of Caucasian patients, all affected by primary membranous nephropathy. Thus, whether and to what extent our present findings can be generalized to other ethnicities or to nephrotic patients with different underlying glomerular diseases is matter of further investigation. We used the 3.5 g/dl cut-off value to discriminate between normoalbuminemia (≥3.5 g/dl) and hypoalbuminemia (<3.5 g/dl). Importantly, this universally accepted threshold has well-established clinical and prognostic implications in affected patients because the more severe complications of the nephrotic syndrome including thrombophilia and thromboembolic complications, severe dyslipidemia, accelerated renal function loss, increased risk of fluid overload and heart failure, fatal and non-fatal major cardiovascular events are particularly associated with the presence of hypoalbuminemia (even independently of severity of proteinuria). One could argue that a continuous analysis would theoretically provide a more refined characterization of the relationship between albumin levels and eGFR formula performance. However, there is a marked difference in the variability of serum albumin levels in hypo-albuminemic compared to normo-albuminemic patients. In hypo-albuminemic patients, the compensatory albumin synthetic liver potential has already reached a plateau and is overwhelmed by large albumin losses in urine. Thus, further increases in urinary losses cannot be counterbalanced by further increase in liver synthesis. Thus, in this context, larger is albuminuria and more severe is hypoalbuminemia. This thus not applies to normo-albuminemic patients. In these patients, liver synthetic activity is modulated on the basis of urinary losses in order to maintain a steady state of “normal” serum albumin levels. Thus, different levels of albuminuria are associated with relatively stable albumin levels, consistently in these patients decreased urinary levels are never associated with “hyperalbuminemia”. Thus, in hypo-albuminemic patients, serum albumin levels inversely correlate with the amount of albuminuria, whereas in normo-albuminemic patients, serum albumin levels are relatively independent of the extent of urinary albumin losses. Consequently, correlation analyses considering altogether data from patients with or without hypoalbuminemia would be biased by combined evaluation of different populations and would not reflect the pathophysiology of the nephrotic syndrome. These could result in false negative findings. On the contrary, the binary approach yields a more reliable result because more precisely reflects the pathophysiology of the nephrotic syndrome. Thus, in this setting, the binary approach yields a more reliable result, though it does not account for variation across the full albumin range, because more precisely reflects the pathophysiology of the nephrotic syndrome. The small sample is explained by the extreme rarity of the considered condition (a rare treatment-resistant form of a rare disease), but is relatively high considering also the single-center nature of the study. On the other hand, the study included a cohort of well-characterized patients prospectively followed with serial, standardized evaluations in the context of prospective longitudinal studies. Moreover, for the first time, the study evaluated the performance of CKD-Epi and EKFC equations, by considering simultaneously also the two most frequently used endogenous markers of glomerular filtration, such as creatinine and cys-C. Also, the performance of all aforementioned eGFR equations was evaluated against GFR values directly measured with a gold standard procedure, considering the potential interactions not only with the severity of proteinuria but also with the concomitant presence and severity of hypoalbuminemia. Finally, in addition to bias and CCC, the study considered a series of other parameters of accuracy and precision, such as TDI, P10, and correlation and Bland–Altman analyses. We considered also P30 values because it is universally accepted in clinics and research, although in our opinion, a 30% error is a too permissive criterion to define acceptable performance for the estimation of such a relevant clinical variable. Consistently, such a large error would be unacceptable for any other clinical variable such as blood pressure or blood glucose and even anthropometric parameters such as height, weight, and body mass index (BMI). Altogether, these considerations consistently converged to indicate the poor performance of all considered GFR estimation equations in this clinical context. This information could help improve clinical patient monitoring and appropriate design of future observational or intervention trials in this context. Moreover, accurate kidney function evaluation is crucial when toxic drugs with (prevalent) renal clearance and narrow therapeutic windows must be dosed, such as in neoplastic patients. Membranous nephropathy and, more in general, nephrotic syndrome and even hypoalbuminemia can be observed in association with cancer [50]. Moreover, in these often critically ill patients, cachexia and reduced muscle mass may contribute to further reduction in serum creatinine levels, with enhanced GFR overestimation and consequent drug overdosing and toxicity. Thus, we claim that, when feasible, GFR estimation equations should be replaced with direct GFR measurements in these so delicate cases. CONCLUSION Our findings show that, in patients with nephrotic syndrome, current creatinine- and cys-C-based GFR estimation equations have important limitations. However, due to the retrospective observational nature of the study, these observations should be considered hypothesis generating. Moreover, they apply to a selected population of nephrotic patients with primary membranous nephropathy and therefore warrant validation in larger, independent cohorts before being generalized to the broader nephrotic syndrome population, including patients with other primary or secondary glomerular diseases. Within the context of our study, the performance of GFR estimation equations was acceptable only when creatinine and cys-C were simultaneously considered as filtration markers in nephrotic patients without hypoalbuminemia. If feasible, direct GFR measurement is recommended to adequately assess kidney function in nephrotic patients with hypoalbuminemia and whenever individual clinical decision makings (including dosing of toxic medications with renal clearance) or evaluation of treatment effects on GFR changes over time are needed in clinics or research, irrespectively of serum albumin levels. Supplementary Material