GFR estimated with creatinine rather than cystatin C is more reflective of the true risk of adverse outcomes with low GFR in kidney transplant recipients

ABSTRACT

Background

Serum cystatin C–based estimated glomerular filtration rate (eGFRcys) generally associates with clinical outcomes better than serum creatinine–based eGFR (eGFRcr) despite similar precision in estimating measured GFR (mGFR). We sought to determine whether the risk of adverse outcomes with eGFRcr or eGFRcys was via GFR alone or also via non-GFR determinants among kidney transplant recipients.

Methods

Consecutive adult kidney transplant recipients underwent a standardized GFR assessment during a routine follow-up clinic visit between 2011 and 2013. Patients were followed for graft failure or the composite outcome of cardiovascular (CV) events or mortality through 2020. The risk of these events by baseline mGFR, eGFRcr and eGFRcys was assessed unadjusted, adjusted for mGFR and adjusted for CV risk factors.

Results

There were 1135 recipients with a mean baseline mGFR of 55.6, eGFRcr of 54.8 and eGFRcys of 46.8 ml/min/1.73 m² and a median follow-up of 6 years. Each 10 ml/min/1.73 m² decrease in mGFR, eGFRcr or eGFRcys associated with graft failure [hazard ratio (HR) 1.79, 1.68 and 2.07, respectively; P < .001 for all) and CV events or mortality outcome (HR 1.28, 1.19 and 1.43, respectively; P < .001 for all). After adjusting for mGFR, eGFRcys associated with graft failure (HR 1.57, P < .001) and CV events or mortality (HR 1.49, P < .001), but eGFRcr did not associate with either. After further adjusting for CV risk factors, risk of these outcomes with lower eGFRcys was attenuated.

Conclusion

eGFRcr better represents the true relationship between GFR and outcomes after kidney transplantation because it has less non-GFR residual association. Cystatin C is better interpreted as a nonspecific prognostic biomarker than is eGFR in the kidney transplant setting.

Graphical Abstract

Graphical Abstract.

KEY LEARNING POINTS.

What is already known about this subject?

• Estimated glomerular filtration rate (eGFR) based on cystatin C (cysC) and eGFR based on serum creatinine both show similar precision versus measured GFR (mGFR).

• However, clinical outcomes are more strongly associated with lower GFR estimated with cysC than lower GFR estimated with serum creatinine. It is unclear the extent this is due to the non-GFR factors that affect cysC and serum creatinine.

• Kidney transplant recipients have a higher mGFR at the same cysC level than nontransplant patients and thus have been excluded from the derivation of commonly used eGFR equations based on cysC.

What this study adds?

• Among kidney transplant recipients, the association of graft failure or of cardiovascular (CV) events or mortality with eGFR based on serum creatinine is predominately via the same pathway as mGFR.

• However, the association of graft failure or CV events or mortality with eGFR based on cysC is strongly via the non-GFR determinants of cysC.

• The non-GFR determinants of cysC also associate with these clinical outcomes via the same pathways as established CV risk factors.

What impact this may have on practice or policy?

• There has been a push in recent years to increase the use of eGFR based on cysC in clinical practice, but this could lead to incorrect interpretation of eGFR and its prognosis in the kidney transplant setting.

• In the kidney transplant setting, eGFR based on serum creatinine rather than cysC should be used to estimate the risk of clinical outcomes with low GFR.

• CysC may still be useful in the kidney transplant setting, but as a nonspecific highly prognostic biomarker rather than as eGFR.

INTRODUCTION

Glomerular filtration rate (GFR) is widely used to assess kidney function and monitor the progression of chronic kidney disease (CKD), including in kidney transplant recipients. However, estimated GFR (eGFR) is used rather than measured GFR (mGFR) in most clinical settings. Studies have shown that creatinine-based eGFR (eGFRcr) with the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation [1] is reasonably unbiased in kidney transplant recipients [2–4]. In contrast, cystatin C (cysC)-based eGFR (eGFRcys), which was developed in nontransplant recipients, underestimates mGFR when applied to kidney or other organ transplant recipients [5–8]. Nonetheless, there remains interest in using eGFRcys in transplant recipients [9], as it is more prognostic for adverse outcomes than eGFRcr [10].

There have been different criteria used to determine the most valid GFR estimating equations [11]. The first has been to identify the GFR estimating equation that has the highest correlation, accuracy and precision against mGFR [12]. The second has been to determine the GFR estimating equation that best associates with kidney disease outcomes. However, this criterion does not determine whether a GFR estimating equation is associating with outcomes via GFR alone or via GFR and non-GFR factors. Since creatinine is a metabolite and cysC is a protein reflective of active biology, non-GFR factors that affect their concentrations may also associate with the outcome of interest. However eGFR causes clinicians to interpret these biomarkers as kidney function, and non-GFR biology could distort our understanding of the relationship between kidney function and outcomes [11]. Thus a third criteria warrants consideration. Rather than determining the GFR estimating equation that best associates with outcomes, we sought to determine the GFR estimating equation that best associates with outcomes via GFR alone rather than via GFR and non-GFR factors.

Several studies have found that cysC-based eGFR associates with cardiovascular (CV) events and mortality better than eGFRcr [13–19], including in kidney transplant recipients [10]. In the general community, CV risk factors associate more strongly with lower eGFRcys than with lower eGFRcr or mGFR [18, 19]. We previously reported in kidney transplant recipients that eGFRcys differed in its association with 10 of 15 CV risk factors compared with mGFR, whereas eGFRcr differed in 5 of 15 CV risk factors compare with mGFR [7]. These studies focus on concurrent risk factors rather than outcomes. The objective of this study was to compare mGFR, eGFRcr, eGFRcys and a combined equation (eGFRcr-cys) to determine if the risk of kidney failure or the risk of CV events or mortality was via GFR alone or via both GFR and non-GFR determinants. We further assessed whether established CV risk factors accounted for residual associations with outcomes via the non-GFR determinants of each eGFR.

MATERIALS AND METHODS

Study population

The Mayo Clinic Institutional Review Board approved this study with a waiver for informed consent. The clinical and research activities being reported are consistent with the principles of the Declaration of Istanbul. Between August 2011 and April 2013, adult (age >18 years) recipients of kidney transplants performed at the Mayo Clinic (Rochester, MN, USA) who were at least 1-year post-transplant underwent a standardized assessment of kidney function, including creatinine and mGFR, during an annual follow-up clinic visit. This clinic visit defined baseline for this study. As part of a research protocol, cysC was analyzed from stored blood samples obtained at the time of mGFR testing. All kidney recipients regardless of multiple kidney transplants, ABO incompatibility, human leukocyte antigen (HLA) matching or concurrent nonkidney organ transplants were included in the study. Individuals who refused or revoked authorization of their medical records for research were excluded [20].

Baseline GFR assessment

Serum creatinine was measured using a standardized enzymatic assay on a Cobas Chemistry Analyzer (c701 or c501; Roche Diagnostics, Indianapolis, IN, USA), whereas cysC was measured using an immunoturbidimetric assay (Gentian, Moss, Norway) traceable to an international reference material. The eGFR was estimated using the CKD-EPI 2009 race inclusive equation (eGFRcr) [1] and the two cysC-based CKD-EPI 2012 GFR estimating equations (eGFRcys and eGFRcr-cys) [8]. The race-inclusive equation was used because it was used in the cohort during baseline and follow-up clinical care, only 2.2% of the participants were Black, and the race-inclusive equation is less statistically biased compared to mGFR. Nonetheless, analyses were also separately performed using the 2021 race-free CKD-EPI eGFR equations [21]. The mGFR was determined by the urinary clearance of a subcutaneous injection of iothalamate over a 45- to 75-minute period as previously described [22]. The CKD-EPI equations were derived using data that included mGFR by this method. Bladder catheterization was used as needed in patients with urinary retention on sonographic bladder scans during the clearance test. The mGFR was converted to the same units as eGFR (ml/min/1.73 m²) by multiplying by 1.73 divided by body surface area (calculated from height and weight) [23].

Baseline CV risk factors

Established risk factors for CV events or mortality or graft failure were determined from the medical records at the time of the baseline GFR assessment. These were age, sex, any prior history of CV events, diabetes mellitus, serum albumin, hemoglobin, deceased donor allograft, measured 24-hour urine protein excretion and glucocorticoid (prednisone) use. The goal was not to study these risk factors as predictors of the clinical outcomes, but to determine if eGFR associates with clinical outcomes independent of these risk factors.

Outcome assessment

Patients were followed via their routine clinical care, including annual follow-up visits to determine clinical outcomes that occurred after the baseline GFR assessment through 15 May 2020. Graft failure was defined as the first onset of an eGFRcr <20 ml/min/1.73 m², dialysis, or retransplant. For the assessment of eGFR associations with graft failure outcome, patients with eGFRcr <20 ml/min/1.73 m² at baseline were excluded. Besides graft failure, a separate composite outcome was also defined as a CV event or mortality. The CV event could be any of the following: acute myocardial infarction, heart failure requiring hospitalization, cardiac arrest with successful resuscitation, ischemic or hemorrhagic stroke, carotid artery disease requiring intervention, renal artery disease requiring intervention or peripheral artery disease requiring intervention (e.g. angioplasty, amputation). Mortality was all-cause, as cause of death data were not available.

Statistical analyses

Separate models assessed the risk of graft failure and the risk of a CV event or mortality with baseline GFR. Baseline GFR was modeled as both continuous (per −10 ml/min/1.73 m²) and categorical. As categorical, >60 ml/min/1.73 m² defined the reference GFR category. For the risk of graft failure, the reference category was compared with the 45–59, 30–44 and 20–29 ml/min/1.73 m² GFR categories. For assessment of the risk of a CV event or mortality, the reference category was compared with the 45–59, 30–44 and 15–29 ml/min/1.73 m² GFR categories. Fine and Gray's subdistribution hazard models were performed for the graft failure outcome such that death without graft failure was treated as competing risk event. Cox regression models were performed for the CV event or mortality outcome. Spline plots were generated for each GFR method to compare hazard ratios (HRs) to clinical outcomes. For the association of these outcomes with eGFR, models also adjusted for mGFR (continuous) alone and then further adjusted for CV risk factors. The residual (mGFR-adjusted) association of each eGFR with each outcome between persons with and without diabetes was assessed with a test of interaction. All statistical analysis were done using SAS software, version 9.3 (SAS Institute, Cary, NC, USA).

RESULTS

Baseline characteristics

There were 1135 kidney recipients with baseline mGFR, eGFRcr and eGFRcys at least 1 year after their kidney transplant surgery. Table 1 characterizes the cohort with respect to baseline CV risk factors and baseline GFR assessment. Further details on the clinical characteristics of this cohort, including the bias, accuracy and precision of each eGFR compared with mGFR, has been previously reported [7]. The median duration of follow-up was 6.2 years (25th–75th percentile 4.7–7.0).

Table 1:

Baseline characteristics of the cohort (N = 1135).

Patient characteristics	Values
Recipient age (years), mean (SD)	56.0 (14.2)
Female, n (%)	497 (43.8)
White, n (%)	1050 (92.7)
Black, n (%)	25 (2.2)
Diabetes, n (%)	395 (34.8)
Prior cardiovascular events, n (%)	126 (11.1)
Hemoglobin (g/dl), mean (SD)	13.1 (1.7)
Serum albumin (g/dl), mean (SD)	4.2 (0.3)
Deceased donor, n (%)	274 (24.1)
24-hour urine protein excretion (mg/24 h), median (range)	100 (0.4–18 175)
Prednisone use, n (%)	1029, (95.5)
Prednisone dose (mg/day), mean (SD)	5.1 (1.7)
GFR biomarkers, mean ± SD
Serum creatinine (mg/dl)	1.40 ± 0.50
Serum cystatin C (mg/l)	1.60 ± 0.50
mGFR (ml/min/1.73 m2)	55.6 ± 20.5
eGFRcra (ml/min/1.73 m2)	54.8 ± 19.8
eGFRcys (ml/min/1.73 m2)	46.8 ± 18.1
eGFRcr-cysa (ml/min/1.73 m2)	49.7 ± 18.2

^aUsing instead the CKD-EPI 2021 race-free equations the mean ± SD eGFRcr was 49.5 ± 19.7 ml/min/1.73 m² and eGFRcr-cys was 48.2 ± 17.9 ml/min/1.73 m².

Graft failure

There were 176 (15.8%) graft failure events. Supplementary Fig. S1 shows the unadjusted increased risk of graft failure with lower GFR by each method. In continuous GFR analysis, the risk of graft failure associated with a lower GFR was strongest for eGFRcys, then eGFRcr-cys, mGFR, and eGFRcr (Table 2). Lower eGFRcr no longer associated with graft failure after adjusting for mGFR, but lower eGFRcys and lower eGFRcr-cys both associated with graft failure after adjusting for mGFR (Fig. 1). After further adjusting for CV risk factors, lower eGFRcys still associated with graft failure, although the strength of the association was attenuated. The residual (mGFR-adjusted) association of each eGFR with graft failure did not differ between persons with or without diabetes (P > .15 for all tests of interaction). Using the 2021 race-free CKD-EPI eGFR equations [21] did not substantively change these associations (Supplementary Table S1). Additional analyses (not shown) found that 24-hour urine proteinuria was the main risk factor responsible for attenuating the risk of graft failure with mGFR-adjusted eGFRcys. The pattern was generally similar when GFR was assessed categorically (Table 3), with the following notable exception. Categorical eGFRcr 20–29 ml/min/1.73 m² associated with graft failure even after mGFR adjustment, although to a lesser extent than did eGFRcys 20–29 ml/min/1.73 m².

Table 2:

The risk of graft failure and CV event or mortality with baseline GFR.

GFR assessment (per −10 ml/min/1.73 m2)	Univariate, HR (95% CI), P-value	Adjusted for mGFR only, HR (95% CI), P-value	Adjusted for mGFR and other clinical characteristicsa, HR (95% CI), P-value
Graft failure
mGFR	1.79 (1.61–1.99), <.001	–	–
eGFRcr	1.68 (1.46–1.94), <.001	1.10 (0.92–1.31), .294	1.00 (0.84–1.20), .959
eGFRcys	2.07 (1.82–2.34), <.001	1.57 (1.29–1.9), <.001	1.31 (1.07–1.60), .009
eGFRcr-cys	1.99 (1.74–2.29), <.001	1.41 (1.13–1.76), .002	1.15 (0.91–1.44), .247
CV event or mortality
mGFR	1.28 (1.21–1.35), <.001	–	–
eGFRcr	1.19 (1.13–1.27), <.001	0.96 (0.88–1.04), .330	0.91 (0.83–1.00), .050
eGFRcys	1.43 (1.34–1.54), <.001	1.49 (1.32–1.69), <.001	1.15 (1.01–1.31), .039
eGFRcr-cys	1.32 (1.24–1.41), <.001	1.16 (1.02–1.31), .022	1.00 (0.87–1.13), .940

^aAge, gender, prior history of CV disease, diabetes, serum albumin, hemoglobin, deceased donor, proteinuria and prednisone use.

Figure 1:

Spline plot of residual risk of kidney failure (HR, y-axis) by (A) eGFRcr, (B) eGFRcys and (C) eGFRcr-cys (x-axes) relative to eGFR = 75 ml/min/1.73 m² in models adjusting for mGFR.

Table 3:

The risk of graft failure with baseline GFR categories.

GFR method	Compared with GFR ≥60 ml/min/1.73 m2	Univariate, HR (95% CI), P-value	Adjusted for mGFR only, HR (95% CI), P-value	Adjusted for mGFR and other clinical characteristicsa, HR (95% CI), P-value
mGFR	GFR 45–59 ml/min/1.73 m2	2.73 (1.63–4.56), <.001	–	–
	GFR 30–44 ml/min/1.73 m2	6.64 (4.09–10.77), <.001
	GFR 20–29 ml/min/1.73 m2	14.53 (8.7–24.26), <.001
eGFRcr	GFR 45–59 ml/min/1.7  m2	1.61 (0.97–2.67), .065	0.87 (0.51–1.47), .600	0.86 (0.49–1.48), .578
	GFR 30–44 ml/min/1.73 m2	4.46 (2.85–6.99), <.001	1.39 (0.81–2.38), .233	1.15 (0.65–2.05), .632
	GFR 20–29 ml/min/1.73 m2	13.48 (8.13–22.37), <.001	2.32 (1.17–4.61), .017	1.36 (0.63–2.92), .433
eGFRcys	GFR 45–59 ml/min/1.73 m2	4.40 (1.7–11.37), .002	2.74 (1.05–7.17), .040	2.96 (1.02–8.59), .046
	GFR 30–44 ml/min/1.73 m2	9.86 (3.98–24.45), <.001	3.84 (1.49–9.88), .005	3.42 (1.15–10.19), .027
	GFR 20–29 ml/min/1.73 m2	30.52 (12.36–75.40), <.001	6.66 (2.33–18.99), <.001	4.58 (1.39–15.13), .012
eGFRcr-cys	GFR 45–59 ml/min/1.73 m2	1.08 (0.59–1.98), .808	0.58 (0.30–1.11), .102	0.53 (0.27–1.05), .071
	GFR 30–44 ml/min/1.73 m2	3.49 (2.10–5.81), <.001	1.12 (0.56–2.23), .753	0.83 (0.39–1.78), .635
	GFR 20–29 ml/min/1.73 m2	12.76 (7.61–21.41), <.001	2.16 (0.9–5.18), .086	1.05 (0.40–2.74), .917

^aAge, gender, prior history of CV disease, diabetes, serum albumin, hemoglobin, deceased donor, proteinuria and prednisone use.

CV event or death

There were 352 (31.0%) CV event or mortality outcomes; of these, 182 were CV events and 170 were deaths. Supplementary Fig. S2 shows the unadjusted increased risk of CV event or mortality with lower GFR by each method. In continuous GFR analysis, the risk of CV event or mortality associated with a lower GFR was strongest with eGFRcys, then eGFRcr-cys, mGFR and eGFRcr (Table 2). Lower eGFRcr no longer associated with CV event or mortality after adjusting for mGFR, but lower eGFRcys and eGFRcr-cys both associated with CV event or mortality after adjusting for mGFR (Fig. 2). After further adjusting for CV risk factors, lower eGFRcys still associated with CV event or mortality, although the strength of association was attenuated. The residual (mGFR-adjusted) association of each eGFR with CV event or mortality did not differ between persons with or without diabetes (P > .25 for all tests of interaction). Using the 2021 race-free CKD-EPI eGFR equations [21] did not substantively change these associations (Supplementary Table S1). Additional analyses (not shown) found that age, hypoalbuminemia, and diabetes were the main risk factors responsible for attenuating the risk of CV event or mortality with mGFR-adjusted eGFRcys. The pattern was generally similar when GFR was assessed categorically (Table 4). Notably, eGFRcr 45–59 ml/min/1.73 m² associated with a lower risk of CV event or mortality after both mGFR and CV risk factor adjustment.

Figure 2:

Spline plot of residual risk of CV event or mortality (HR, y-axis) by (A) eGFRcr, (B) eGFRcys and (C) eGFRcr-cys (x-axes) relative to eGFR = 75 ml/min/1.73 m² in models adjusting for mGFR.

Table 4:

The risk of CV event or mortality with baseline GFR categories.

GFR method	Compared with GFR ≥60 ml/min/1.73 m2	Univariate, HR (95% CI), P-value	Adjusted for mGFR only, HR (95% CI), P-value	Adjusted for mGFR and other clinical characteristicsa, HR (95% CI), P-value
mGFR	GFR 45–59 ml/min/1.73 m2	1.33 (0.99–1.77), .054	–	–
	GFR 30–44 ml/min/1.73 m2	2.11 (1.59–2.80), <.001
	GFR 15–29 ml/min/1.73 m2	4.04 (2.98–5.47), <.001
eGFRcr	GFR 45–59 ml/min/1.73 m2	1.19 (0.89–1.58), .233	0.82 (0.6–1.12), .213	0.71 (0.52–0.98), .037
	GFR 30–44 ml/min/1.73 m2	1.83 (1.39–2.41), <.001	0.95 (0.67–1.35), .767	0.74 (0.51–1.08), .120
	GFR 15–29 ml/min/1.73 m2	2.95 (2.11–4.12), <.001	1.03 (0.63–1.68), .899	0.70 (0.41–1.19), .188
eGFRcys	GFR 45–59 ml/min/1.73 m2	2.22 (1.49–3.31), <.001	2.18 (1.44–3.31), <.001	1.50 (0.99–2.28), .058
	GFR 30–44 ml/min/1.73 m2	2.68 (1.83–3.93), <.001	2.58 (1.63–4.08), <.001	1.51 (0.95–2.42), .084
	GFR 15–29 ml/min/1.73 m2	6.44 (4.4–9.44), <.001	6.06 (3.46–10.61), <.001	2.39 (1.32–4.33), .004
eGFRcr-cys	GFR 45–59 ml/min/1.73 m2	1.20 (0.86–1.67), .278	0.95 (0.65–1.37), .768	0.78 (0.54–1.13), .182
	GFR 30–44 ml/min/1.73 m2	1.97 (1.46–2.67), <.001	1.31 (0.86–1.99), .214	0.88 (0.57–136), .570
	GFR 15–29 ml/min/1.73 m2	3.84 (2.77–5.30), <.001	1.98 (1.12–3.51), .019	1.06 (0.57–1.95), .862

^aAge, gender, prior history of CV disease, diabetes, serum albumin, hemoglobin, deceased donor, proteinuria and prednisone use.

DISCUSSION

This study clarifies the interpretation of eGFR as a valid assessment of the risk of GFR-related clinical outcomes among kidney transplant recipients. First, eGFRcr showed less residual association with outcomes independent of mGFR than did eGFRcys or eGFRcr-cys. Second, lower eGFRcys showed strong residual associations with clinical outcomes in this population independent of mGFR, and CV risk factors only partially explained these residual associations. These data suggest that among kidney transplant recipients, eGFRcr is more accurate than eGFRcys for determining associations with clinical outcomes due to low GFR when mGFR is not available.

These analyses highlight the importance of having thoughtful criteria when determining which eGFR is ‘best’ for prognostic purposes. If the goal is to optimally predict clinical outcomes, then eGFRcys is superior to eGFRcr for this purpose. However, use of the term ‘eGFRcys’ in this setting is misleading, as the prediction of the clinical outcomes is via non-GFR biology. In this study, eGFRcys associated with CV events or mortality to a similar extent with and without adjusting for mGFR. This implies that the non-GFR determinants of cysC contribute more to CV events or mortality than does GFR. CysC can certainly be used as a nonspecific prognostic biomarker for clinical outcomes, but the term ‘eGFRcys’ is misleading in this setting. If the goal is to understand associations with clinical outcomes via only true reductions in GFR, then eGFRcr is superior to eGFRcys. The eGFRcr had less association with clinical outcomes via non-GFR pathways than did eGFRcys. Notably, eGFRcr was weaker than mGFR for associations with clinical outcomes, but this should be expected from model error with eGFR (i.e. eGFR is always an imperfect statistical model for estimating mGFR). Much of the non-GFR biology of eGFRcr is creatinine generation from muscle mass [24]. Higher muscle mass may explain why eGFRcr 45–59 ml/min/1.73 m² was modestly more protective than eGFRcr ≥60 ml/min/1.73 m² for CV event or mortality in the fully adjusted model. However, there was no evidence of eGFRcr 45–59 ml/min/1.73 m² being protective for CV event or mortality in analysis adjusting for mGFR alone. Conversely, lower eGFRcys, whether continuous or categorial, was consistently associated with a higher risk of adverse outcomes adjusting for mGFR alone. Thus lower eGFRcr more consistently represented the true risks with lower mGFR than did lower eGFRcys.

There are several potential mechanisms that can explain why eGFRcys associated with clinical outcomes in kidney recipients via non-GFR biology. Prior studies have found that after adjusting for mGFR, cysC is significantly associated with older age, obesity, diabetes, prior history of CV events, proteinuria and hypertriglyceridemia in transplant recipients [7, 19, 25–27]. CysC is also involved in inflammation, atherosclerosis and modulation of cardiac remodeling [28, 29] and elevated cysC levels correlate with CV disease and all-cause and CV mortality in the general population [30–32]. Kidney transplantation itself is a risk factor for these outcomes [33, 34] and kidney transplant recipients have higher cysC levels than those without a transplant at the same mGFR [5, 35]. We found that much of the non-GFR association of eGFRcys with CV events or mortality was via diabetes and hypoalbuminemia (a marker of inflammation). Indeed, several studies found that cysC correlates with diabetes and markers of inflammation independent of GFR [36–38]. Lower eGFRcys still associated with clinical outcomes after further adjusting for many CV risk factors. This may be due to nontraditional CV risk factors (e.g. dimethylarginine and insulin resistance) that associate with cystatin C via non-GFR pathways [18].

Prior studies have also suggested the non-GFR determinants of cystatin C inflate the risk of adverse outcomes in native kidney disease patients. A study in CKD patients found a stronger association of cystatin C with all-cause mortality and CV mortality compared with the association of serum creatinine and mGFR with these outcomes, even after adjusting for CV risk factors [14]. In the same cohort with longer follow-up, higher serum creatinine did not associate with kidney failure or CV mortality after adjusting for mGFR and CV risk factors [39]. In contrast, the relationship of higher cysC with all-cause mortality and CV mortality was stronger after adjusting for mGFR and remained significant after adjusting for CV factors, findings very consistent with our study. In the African American Study of Kidney Disease and Hypertension trial, higher cysC was more strongly associated with the risk of end-stage kidney disease or mortality than higher serum creatinine or lower eGFRcr after adjusting for mGFR [17]. The increased risk of mortality and end-stage kidney disease with higher cysC even after adjusting for mGFR has been reported in studies in Sweden [40] and among the indigenous Americans from the Pima tribe [41]. Cardiovascular risk factors were similar to those in our study and included diabetes, age, hypertension and urine protein assessment [41]. While these studies often used creatinine and cysC alone rather eGFRcr or eGFRcys, the conceptual framework is the same as eGFR is a rescaled biomarker. These studies support the notion that cysC (or eGFRcys) reflects a higher risk of CV outcomes, mortality, and kidney failure due to non-GFR biology.

There are several potential limitations to these analyses to consider. First, we only had a single assessment eGFR and mGFR available. It is possible that if the non-GFR determinants of eGFRcys did not change over time within a patient, then changes in eGFRcys would reasonably approximate the risk of outcomes with changes in mGFR. Further studies are needed to assess this. Second, race has more impact on the performance of creatinine- than cysC-based equations [42], but we only had 2.2% Black patients and could not meaningfully assess this impact. Third, residual confounding may have occurred such that eGFR associates with clinical outcomes independent of mGFR because of measurement error with mGFR. We have previously reported that the test–retest coefficient of variance for our method of mGFR was 8.2% [19]. Even so, it is implausible that residual confounding alone would lead to residual associations with eGFRcys but not eGFRcr. Fourth, we adjusted for a limited set of CV risk factors that we selected a priori. It is possible that additional risk factors could further account for the residual association of eGFR with outcomes. Finally, the onset of graft failure was determined by eGFRcr, not eGFRcys. But this would be expected to bias associations to be stronger with eGFRcr than with eGFRcys, the opposite of what was observed in these data.

In conclusion, the association of adverse clinical outcomes with lower eGFRcr among kidney recipients reasonably approximates the true risk with lower GFR. However, the association of these same outcomes is inflated with lower eGFRcys due to the non-GFR determinants of cysC. As such, cysC should not be interpreted as ‘eGFR’ in the kidney transplant setting but as a nonspecific but highly prognostic biomarker. CysC may be particularly useful in predictive models for outcomes when the biological basis for the prediction is not important.

FUNDING

This project was supported in part by the National Institute of Allergy and Infectious Diseases of the National Institutes of Health [NIH; award no. K23AI143882 (principal investigator E.F.B.)]. The funding source had no role in study design; data collection, analysis or interpretation; writing the report; or the decision to submit the report for publication. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH. There are no other relevant disclosures by the other authors.

AUTHORS’ CONTRIBUTIONS

A.D.R. and M.T.K. designed the study. M.T.K., M.R.H. and L.G. collected the data. N.Z. and R.J.B. analyzed the data. M.T.K., L.G., M.R.H., E.F.B. and A.D.R. drafted the manuscript. All authors contributed to revisions and approved the final version of the manuscript.

DATA AVAILABILITY STATEMENT

The authors are open to collaborations, but these clinical data cannot be shared without a data use agreement and institutional review board approval. Research proposals for use of these data should be submitted to the corresponding author.

CONFLICT OF INTEREST STATEMENT

The authors have nothing to disclose.