Performance of Nuclear Magnetic Resonance-Based Estimated Glomerular Filtration Rate in a Real-World Setting

An accurate estimate of glomerular filtration rate (eGFR) is essential for proper clinical management, especially in patients with kidney dysfunction. This prospective observational study evaluated the real-world performance of the nuclear magnetic resonance (NMR)-based GFRNMR equation, which combines creatinine, cystatin C, valine, and myo-inositol with age and sex. We compared GFRNMR performance to that of the 2021 CKD-EPI creatinine and creatinine-cystatin C equations (CKD-EPI2021Cr and CKD-EPI2021CrCys), using 115 fresh routine samples of patients scheduled for urinary iothalamate clearance measurement (mGFR). Median bias to mGFR of the three eGFR equations was comparably low, ranging from 0.4 to 2.0 mL/min/1.73 m2. GFRNMR outperformed the 2021 CKD-EPI equations in terms of precision (interquartile range to mGFR of 10.5 vs. 17.9 mL/min/1.73 m2 for GFRNMR vs. CKD-EPI2021CrCys; p = 0.01) and accuracy (P15, P20, and P30 of 66.1% vs. 48.7% [p = 0.007], 80.0% vs. 60.0% [p < 0.001] and 95.7% vs. 86.1% [p = 0.006], respectively, for GFRNMR vs. CKD-EPI2021CrCys). Clinical parameters such as etiology, comorbidities, or medications did not significantly alter the performance of the three eGFR equations. Altogether, this study confirmed the utility of GFRNMR for accurate GFR estimation, and its potential value in routine clinical practice for improved medical care.


Introduction
Glomerular filtration rate (GFR) is a critical clinical parameter in routine medical care and for the diagnosis and monitoring of kidney diseases [1]. To minimize systematic errors among and between patient groups, equations for estimated GFR (eGFR) incorporate demographic variables such as sex, age, and-until recently-race, together with endogenous filtration markers such as creatinine and cystatin C. As race is a social construct [2], healthcare professionals called for a reassessment of the inclusion of race in eGFR equations. In response, the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) recently published the 2021 CKD-EPI eGFR equations that do not incorporate race [3]. Several studies demonstrated the clinical impact of implementing the new race-free equations on global patient management [4], and its benefit for kidney transplantation eligibility listing for Black patients [5][6][7][8].
The National Kidney Foundation (NKF) and the American Society of Nephrology (ASN) Task Force recently recommended the implementation of the race-free 2021 CKD-EPI creatinine equation (CKD-EPI 2021Cr ) for US adults [9,10]. They also recommended facilitating the routine use of cystatin C, because equations combining creatinine and cystatin C (e.g., CKD-EPI 2021CrCys ) are more accurate [9,10]. Finally, the Task Force encouraged the development of accurate, unbiased, precise, race-free equations integrating new endogenous filtration markers, to promote health equity [9,10]. We recently described GFR NMR , a nuclear magnetic resonance (NMR)-based equation without a race variable, combining the serum biomarkers creatinine, cystatin C, valine, and myo-inositol [11,12]. GFR NMR demonstrated lower bias, and higher accuracy and precision than the 2009 and 2012 CDK-EPI equations [11], and demonstrated analytical performance suitable for routine clinical use [12]. When compared with the 2021 creatinine CKD-EPI 2021Cr equation, GFR NMR showed comparable bias and significantly higher P15 accuracy in patients with or without a kidney transplant, and a stronger agreement with CKD staging by measured GFR (mGFR) in kidney transplant recipients [13]. Therefore, GFR NMR may fulfill the recommendation of the NKF-ASN Task Force and hold promise for an alternative, well-performing and equitable eGFR equation.
Given the importance of accurately estimating GFR in medical care, particularly for proper CKD staging and clinical decision making, there is a clear need for independent validation of GFR NMR in a real-world setting to confirm its potential use for improved patient management. Therefore, the aim of this prospective observational study was to validate GFR NMR in a setting of routine clinical practice. Blood was collected from patients scheduled for clinically indicated urinary iothalamate clearance measurement, and GFR was estimated in fresh sera using GFR NMR and the guideline-recommended CKD-EPI 2021Cr and CKD-EPI 2021CrCys race-free equations. A wide range of clinical data describing the etiology, comorbidities, and medications of the study sample were collected to investigate their potential impact on eGFR results.

Study Design and Participants
A prospective, observational, single center (Mayo Clinic, Rochester, MN, USA) study was conducted in 120 patients ≥ 18 years old scheduled for urinary iothalamate clearance measurement between May and September 2022. During the recruiting period, all patients scheduled for mGFR as part of clinical routine were screened for eligibility. Patients under hemodialysis or under peritoneal dialysis within seven days before urinary iothalamate clearance measurement were excluded from enrolment. Some patients were recruited repeatedly in the study, and the results considered as independent. Patients' demographic and clinical data (including comorbidities and medications) were documented on the day of examination.
The aim of the study was to validate GFR NMR in the actual daily routine setting of clinical practice. The clinical reference standard was GFR measured by urinary iothalamate clearance, which is part of standard care at Mayo Clinic (Rochester, MN, USA). Fresh blood samples were collected immediately prior to mGFR measurement to estimate GFR (eGFR) using the NMR-based GFR NMR equation [11,12] and the guideline-recommended 2021 CKD-EPI equations (CKD-EPI 2021Cr and CKD-EPI 2021CrCys ) [3,9,10]. Physicians and patients were blinded to the eGFR results.
The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the respective Institutional Review Board (Mayo Clinic IRB #21-007723, dated 11 October 2021). All patients gave written informed consent prior to enrolment.

Sample Collection and Storage
A total of 120 blood samples (9 mL) were collected by venipuncture immediately before injection of iothalamate to measure GFR by urinary iothalamate clearance. Whole blood was allowed to clot for 30-120 min at room temperature and was centrifuged to collect blood serum. Fresh sera (at least 3 mL) were aliquoted (two 1 mL and two 0.5 mL aliquots) and stored at 4 • C until testing. On the day of examination, one 0.5 mL aliquot was used to measure creatinine and cystatin C levels, as described in Section 2.3. Then, one 1 mL aliquot was used within four days of blood collection for NMR measurement, as described in Section 2.3.

Laboratory Methods
GFR was measured at the Mayo Clinic Renal Testing Laboratory (Rochester, MN, USA) by urinary iothalamate clearance (non-radiolabeled) using liquid chromatography-tandem mass spectrometry, as previously reported [13,14]. The mGFR was normalized to body surface area according to the Dubois equation (body surface area = height 0.725 × weight 0.425 × 0.007184) and expressed as milliliter per minute per 1.73 m 2 body surface area (mL/min/1.73 m 2 ).
Biomarker measurements were performed on fresh refrigerated sera at Central Clinical Chemistry Laboratory, Mayo Clinic, Rochester, MN, USA. Serum creatinine was measured by enzymatic assay standardized to international reference materials [15], using Roche Cobas clinical analyzers (c701 or c501, Roche Diagnostics, Indianapolis, IN, USA). Cystatin C was measured by an immunoturbidometric assay (Gentian ASA, Moss, Norway) that was traceable to an international reference material [16], using a Roche Cobas c501 analyzer (Roche Diagnostics; Indianapolis, IN, USA).
Serum creatinine, valine, and myo-inositol were measured by NMR spectroscopy as previously described [11,13,17]. Briefly, 540 µL serum was mixed with 60 µL of Axinon ® serum additive solution and 600 µL was transferred into a 5 mm NMR tube with a barcoded cap. Samples were pre-heated at 37 • C for 7.5 min before NMR measurement in a Bruker Avance III 600 MHz, and a 5 mm PATXI probe equipped with automatic Z gradients shimming. The 1 H-NMR spectra were recorded using a spectral width of 20 ppm, with a recycling delay of 1.5 s, 16 scans, and a fixed receiver gain of 50.4. A cycling time d2 of 8 ms was used together with a corresponding T2 filter of 112 ms. The mixing time τ between two consecutive spin echoes was 400 µs. The NMR data were automatically phase-and baseline-corrected using the lactate doublet at 1.32 ppm as reference. Metabolite quantification used curve-fitted pseudo-Voigt profiles, as previously described [11,17]. In case of analysis failure, the second 1 mL aliquot was used to repeat the NMR analysis. In case of repeated failure, the sample was excluded from final analysis.

Statistical Analysis
Sample size was estimated based on existing GFR NMR P20 accuracy results using the MedCalc Statistical Software version 12.7.7, according to Machin et al. [18]. Based on an estimated error not exceeding 15% absolute, with a two-sided alpha level of 5% and a power of 90%, a minimum of 106 patients was required for the study. Including a safety margin of 15% for possible dropouts, a total of 120 patients was planned for enrolment.
Performance evaluation (signed median bias, precision, accuracy, and precision intervals) was conducted in all enrolled patients with a valid GFR NMR result. Subgroup analyses were also conducted to evaluate the impact of disease, comorbidities, or medication on the performance of each eGFR equation.
All statistical evaluations were performed within R 4.0.2 [19]. Data structures were handled with data.frame [20], data.table [21], and archivist [22] packages. Bootstrap procedures were implemented via the boot package [23,24]. Visualization was performed with ggplot2 [25]. In bias, precision and accuracy analyses, the respective 95% confidence intervals (CI) were calculated using the bootstrap method. In all analyses, p-values ≤ 0.05 were considered statistically significant.
Bias was calculated as 'eGFR-mGFR' and expressed as median signed bias to mGFR. Pairwise significance levels between bias distributions were assessed via the Wilcoxon-signed rank test [26,27] with p-value correction for multiple testing according to Benjamini-Hochberg [28,29].
Precision was assessed by the interquartile range (IQR) of the difference to mGFR. Significance of differences was assessed via the bootstrap method.
Distribution of mGFR at any given eGFR was assessed by fitting a quantile regression model for quantiles 2.5th, 10th, 25th, 50th, 75th, 90th, and 97.5th (one model for each quantile value, for a total of 7 quantile regression models). Quantile regression differs from the ordinary least squares (OLS) regression in that OLS regression estimates the conditional mean, whereas quantile regression estimates the conditional quantile of interest (e.g., 75th quantile or 50th quantile) [31][32][33][34][35]. For each of the studied equations (GFR NMR , CKD-EPI 2021Cr and CKD-EPI 2021CrCys ), mGFR distribution was calculated at given eGFR thresholds (45, 60 and 90 mL/min/1.73 m 2 ). These thresholds were chosen as they represent GFR decision values for CKD staging [36]. The 95% prediction interval (PI) of mGFR for a given eGFR threshold (e.g., 30 mL/min/1.73 m 2 ) was defined as the predicted mGFR by the 97.5th quantile model at eGFR equals 30 mL/min/1.73 m 2 minus the predicted mGFR by the 2.5th quantile model at eGFR equals 30 mL/min/1.73 m 2 . This 95% PI is thus expected to include approximately 95% of the mGFR values from patients with a given eGFR.

Patient Characteristics
A total of 120 sera were collected as part of routine clinical practice from patients whose mGFR was determined by urinary iothalamate clearance ( Figure 1). GFR was estimated using three race-free eGFR equations: NMR-based GFR NMR [11,12], guideline-recommended CKD-EPI 2021Cr , and CKD-EPI 2021CrCys [3]. Of the 120 tested samples, 115 with a valid GFR NMR result were included in the analysis ( Figure 1). with ggplot2 [25]. In bias, precision and accuracy analyses, the respective 95% confidence intervals (CI) were calculated using the bootstrap method. In all analyses, p-values ≤ 0.05 were considered statistically significant.
Bias was calculated as 'eGFR-mGFR' and expressed as median signed bias to mGFR. Pairwise significance levels between bias distributions were assessed via the Wilcoxonsigned rank test [26,27] with p-value correction for multiple testing according to Benjamini-Hochberg [28,29].
Precision was assessed by the interquartile range (IQR) of the difference to mGFR. Significance of differences was assessed via the bootstrap method.
Distribution of mGFR at any given eGFR was assessed by fitting a quantile regression model for quantiles 2.5th, 10th, 25th, 50th, 75th, 90th, and 97.5th (one model for each quantile value, for a total of 7 quantile regression models). Quantile regression differs from the ordinary least squares (OLS) regression in that OLS regression estimates the conditional mean, whereas quantile regression estimates the conditional quantile of interest (e.g., 75th quantile or 50th quantile) [31][32][33][34][35]. For each of the studied equations (GFRNMR, CKD-EPI2021Cr and CKD-EPI2021CrCys), mGFR distribution was calculated at given eGFR thresholds (45, 60 and 90 mL/min/1.73 m 2 ). These thresholds were chosen as they represent GFR decision values for CKD staging [36]. The 95% prediction interval (PI) of mGFR for a given eGFR threshold (e.g., 30 mL/min/1.73 m 2 ) was defined as the predicted mGFR by the 97.5th quantile model at eGFR equals 30 mL/min/1.73 m 2 minus the predicted mGFR by the 2.5th quantile model at eGFR equals 30 mL/min/1.73 m 2 . This 95% PI is thus expected to include approximately 95% of the mGFR values from patients with a given eGFR.
Demographic and clinical characteristics of the study population (n = 115) are shown in 1 Percentages refer to documented characteristics in the study population, not taking into account patients with missing characteristics (the number of patients with missing characteristics is indicated in brackets); 2 CKD staging based on mGFR and according to the Kidney Disease: Improving Global Outcomes (KDIGO) guideline [36]; 3 Includes patients with single kidney transplantation, combined kidney-liver or kidney-pancreas transplantation; 4 Expressed as mL/min/1.73 m 2 of body-surface area. Abbreviations: ACE, angiotensin-converting enzyme; CKD, chronic kidney disease; CKD-EPI, Chronic Kidney Disease-Epidemiology Collaboration; CKD-EPI 2021Cr , 2021 creatinine eGFR equation without race [3]; CKD-EPI 2021CrCys , 2021 creatinine-cystatin C eGFR equation without race [3]; eGFR, estimated GFR; GFR, glomerular filtration rate; GFR NMR , NMR-based eGFR equation [11,12]; mGFR, measured GFR; N, number of samples; SD, standard deviation.

Prediction Intervals
For each of the studied equations (GFRNMR, CKD-EPI2021Cr and CKD-EPI2021CrCys), we assessed the distribution of mGFR at eGFR thresholds used to define CKD stages, by determining the mGFR 95% prediction intervals (PI). The 95% PI is expected to include approximately 95% (2.5th to 97.5th percentiles) of the mGFR values from patients with a given eGFR. The width of the 95% PI has direct clinical implications on the precision of CKD staging [37]. Due to the small number (n = 3) of patients with GFR < 30 mL/min/1.73 m 2 in our study population (Table 1, CKD stages G4 and G5), we considered only the eGFR thresholds of 45, 60, and 90 mL/min/1.73 m 2 .

Prediction Intervals
For each of the studied equations (GFR NMR , CKD-EPI 2021Cr and CKD-EPI 2021CrCys ), we assessed the distribution of mGFR at eGFR thresholds used to define CKD stages, by determining the mGFR 95% prediction intervals (PI). The 95% PI is expected to include approximately 95% (2.5th to 97.5th percentiles) of the mGFR values from patients with a given eGFR. The width of the 95% PI has direct clinical implications on the precision of CKD staging [37]. Due to the small number (n = 3) of patients with GFR < 30 mL/min/1.73 m 2 in our study population (Table 1, CKD stages G4 and G5), we considered only the eGFR thresholds of 45, 60, and 90 mL/min/1.73 m 2 .

Impact of Disease, Comorbidities and Medication on the Performance of eGFR Equations
To evaluate the impact of clinical parameters on the performance of the three eGFR equations, subgroup analyses were performed according to the documented etiology, comorbidities, and administered medications. This analysis was limited to groups of > 25 samples and to conditions related to kidney function or dysfunction (Table 1). Bias distribution for each equation was compared between subgroups of patients with vs. without CKD, kidney transplantation, hypertension, or hyperlipidemia, and those treated vs. untreated with immunosuppressive agents, corticosteroids, or beta blockers (Table 3). The whiskers represent the 95% prediction intervals (PI) between mGFR and eGFR (2.5th to 97.5th percentiles; p2.5, p97.5). The symbols (triangles, circles, and squares) indicate the percentiles of mGFR at a given eGFR (p10 to p90, p25 to p75, and p50 or median, respectively). At a given eGFR, 50% of mGFR values range from the 25th to 75th percentiles, 80% from the 10th to 90th percentiles, and 95% from the 2.5th to 97.5th percentiles (defined as 95% PI). Abbreviations: CKD-EPI 2021Cr , 2021 creatinine eGFR equation without race [3]; CKD-EPI 2021CrCys , 2021 creatinine-cystatin C eGFR equation without race [3]; eGFR, estimated GFR; GFR, glomerular filtration rate; GFR NMR , NMRbased eGFR equation [11,12]; mGFR, measured GFR; px, percentile.

Impact of Disease, Comorbidities and Medication on the Performance of eGFR Equations
To evaluate the impact of clinical parameters on the performance of the three eGFR equations, subgroup analyses were performed according to the documented etiology, comorbidities, and administered medications. This analysis was limited to groups of >25 samples and to conditions related to kidney function or dysfunction (Table 1). Bias distribution for each equation was compared between subgroups of patients with vs. without CKD, kidney transplantation, hypertension, or hyperlipidemia, and those treated vs. untreated with immunosuppressive agents, corticosteroids, or beta blockers (Table 3).

Discussion
The performance of three eGFR equations (GFR NMR , CKD-EPI 2021Cr , and CKD-EPI 2021CrCys ) was compared using fresh samples of patients with clinically ordered urinary iothalamate clearance mGFR. The three equations showed an overall low median bias, however the bias distributions differed in subtle but potentially important ways. Bias distribution for GFR NMR was centered around its median of 2.0 mL/min/1.73 m 2 , while that of the CKD-EPI equations was broad and bimodal, indicating heterogeneity in performance. This heterogenicity may be reflective of the different confounders for creatinine and cystatin C, which can be in opposite directions. On the other hand, GFR NMR outperformed the 2021 CKD-EPI equations in terms of precision (interquartile range to mGFR of 10.5 mL/min/1.73 m 2 ) and accuracy (P15, P20, and P30 of 66.1%, 80.0%, and 95.7%, respectively).
Similar to the 2021 CKD-EPI equations, GFR NMR is a race-free equation, and like CKD-EPI 2021CrCys , GFR NMR integrates both creatinine and cystatin C as biomarkers in addition to age and sex. The improved precision and accuracy of GFR NMR compared to CKD-EPI 2021CrCys supports the benefit of adding myo-inositol and valine to the eGFR equation. These results further support published reports and recommendations on the need for adequate biomarker constellations to improve GFR estimation [9][10][11]17,[38][39][40].
The GFR NMR performance evaluation reported here confirms in an independent cohort the findings of our first validation study [11]. Both studies, employing an iothalamate-based mGFR reference standard and including patients with comparable demographics (Mayo Clinic, Rochester, NY, USA), confirmed the results obtained for GFR NMR as to precision ( . This finding further supports the suitability of GFR NMR for use in clinical routine settings, particularly for e.g., the accurate classification of CKD stages, assessment of renal impairment and renal drug dose adjustment. For the first time, we were able to systematically investigate the impact of a variety of etiologies, comorbidities, and medications on GFR NMR results. Although clinical data were collected on a wide range of conditions, the statistical analysis had to be restricted to groups of > 25 samples and to univariate statistics to ensure result integrity. While in most comparisons, no statistically significant differences in bias distributions were found between groups, we observed a significant difference in bias for GFR NMR between beta-blocker treatment groups (median bias [95% CI] of 0.0 [−3.0; 4.0] vs. 2.0 [0.0; 4.0]; p = 0.05). In the kidney, adrenergic receptors mediate vasoconstriction, renin secretion, and vasodilation, respectively [41][42][43]. Blockade of beta-receptors may therefore affect renal blood flow and possibly GFR through intrarenal effects [41][42][43]. Interestingly, all investigated eGFR equations tended to underestimate mGFR in the beta-blocker treatment group compared to the control group, an observation consistent with a decreased renal perfusion. However, the univariate nature of our evaluation due to the small sample size prevented us from correcting the observed effect sizes for important covariates, such as age, sex, race, other medications, or CKD etiology. Significantly larger patient populations for such an analysis is warranted.
Limitations of our study include its small population size (n = 115), its single center nature, the high proportion (88.7%) of solid-organ transplant recipients, the small proportion (<3%) of non-White participants, and the small number (n = 3) of patients with low (<30 mL/min/1.73 m 2 ) mGFR. The latter is critical because low mGFR thresholds represent essential clinical decision points, such as to eligibility to kidney transplantation listing (at GFR < 20 mL/min/1.73 m 2 in the US). The weak representation of non-White participants and of patients with low mGFR, and the strong representation of certain medical conditions (such as solid-organ transplantation) constitute the drawback of a real-world single-center study. Future real-world studies should be conducted at multiple centers to cover more ethnicities, more patients with severe CKD, and more diverse medical conditions. Strengths of our study include its real-world setting, the subgroup analysis according to comorbidities and medication, and the use of a single standardized method for mGFR determination (urinary iothalamate clearance). This is important because the use of mixed mGFR determination methods might introduce a bias in eGFR performance interpretation [3,11,14,44,45]. In fact, recent studies evaluating GFR NMR exclusively relied on urinary iothalamate clearance (this study and [13]).
Overall, our results add real-world eGFR validation data to the growing body of evidence confirming the suitability of GFR NMR to meet the NKF-ASN Task Force recommendations, and hold promise for an alternative, well-performing, and equitable eGFR equation.

Conclusions
This real-world study demonstrated a superior performance of GFR NMR compared to the CKD-EPI 2021Cr and CKD-EPI 2021CrCys equations on routine samples. This study thus validates the use of GFR NMR for accurate estimation of GFR in routine clinical practice, potentially improving patient management. Funding: This study received no external funding and was sponsored by Numares AG. The APC was funded by Numares AG.

Institutional Review Board Statement:
The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board of Mayo Clinic (IRB #21-007723 dated 11 October 2021).

Informed Consent Statement:
Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available within the article.