Introduction

Accurate assessment of kidney function is essential in the care of children at high risk for renal injury and chronic kidney disease (CKD), such as those receiving a transplant. Careful attention to the glomerular filtration rate (GFR) ensures proper medication dosing and allows kidney function to be followed over time. While serum creatinine is currently used to monitor renal function, cystatin C is emerging as a potentially superior alternative [1]. Cystatin C, produced by all nucleated cells, is a 13-kDa protein inhibitor of cysteine proteases. It is freely filtered by the glomerulus without tubular secretion and is generated at a constant rate, making it an ideal marker to estimate GFR [2].

Several studies have compared cystatin C to serum creatinine for estimating GFR in children. Unlike serum creatinine, which increases with growth and development, cystatin C has been shown to be independent of age, height and body composition [3, 4]. Therefore, serum cystatin C levels parallel renal maturation, attaining peak values in the neonatal period and then decreasing during the first few months of life [5]. Despite these advantages, cystatin C has not been shown to consistently perform better than creatinine in the estimation of GFR. While some studies have suggested that cystatin C is a more accurate and sensitive marker of GFR decline [6–8], others have failed to demonstrate a significant difference [9, 10].

Many cystatin C-based equations have been published, including some which include serum creatinine and other covariates to improve GFR estimation. Most recently, Schwartz et al. reported that combining cystatin C with serum creatinine, blood urea nitrogen, height and gender provided the best estimate of GFR in 349 children in the Chronic Kidney Disease in Children (CKiD) cohort [11]. However, limited data exist on the use of cystatin C to estimate GFR in pediatric transplant recipients and other children at high risk of renal dysfunction. The primary aim of this study was therefore to evaluate cystatin C-based formulas in children whose primary disease increased the risk of renal injury.

Patients and methods

This study was a retrospective, cross-sectional analysis of 142 consecutive children and young adults who received a nuclear medicine GFR (N5ucGFR) examination using 99mTc- diethylenetriaminepentaacetic acid (99mTc-DTPA) at Cincinnati Children’s Hospital Medical Center from June to August 2010. All patients undergoing NucGFR testing for clinical indications during this time period were included. Patients were excluded if they had previously received NucGFR testing during the study period to avoid repeated measurements. Patients with more than one NucGFR examination during the study period only had their first test included in the analysis. Cystatin C values were obtained as a quality improvement effort in the Division of Nephrology clinical laboratory to evaluate the accuracy of the GFR estimation.

Examination of a residual plot of cystatin C and NucGFR revealed one significant outlier from a subject with a very high cystatin C concentration (3.34 mg/L) on the day of NucGFR testing that was inconsistent with subsequent values (0.75–1.61 mg/L). Furthermore, serum creatinine levels in this 17-year-old subject were normal, ranging from 0.6 to 0.9 mg/dL, and the NucGFR was only slightly decreased (85 ml/min/1.73 m2). This patient was therefore excluded from the analysis, leaving a final cohort of 141 subjects.

NucGFR testing was performed in the outpatient or inpatient setting, and the NucGFR was calculated according to the methods of Balachandran et al [12]. A single dose of 99mTc-DTPA was administered with direct visualization and intermittent aspiration of blood while injecting to ensure delivery of the isotope intravenously. If extravasation was suspected, the injection site was evaluated with gamma camera imaging. The NucGFR was then computed using plasma 99mTc-DTPA disappearance curves obtained from four time points at approximately 120, 150, 180 and 210 min. Each plasma disappearance curve had a correlation coefficient of >0.98, indicating intercompartmental equilibration had occurred. A quadratic correction factor was then used to adjust the slope–intercept GFR to a two-compartment model according to the methods of Brochner–Mortensen [13], as recommended for pediatric patients by Blaufox et al. [14]. 99mTc-DTPA has previously demonstrated good agreement with inulin clearance [15]. After the GFR was measured, samples from each subject were pooled and stored at −80 °C until radioisotope decay occurred prior to the measurement of cystatin C on the pooled serum samples from each subject. Cystatin C was measured using particle-enhanced immunonephelometry (Siemens Healthcare Diagnostics, Deerfield, IL) in the Cincinnati Children’s Hospital Division of Nephrology clinical laboratory.

GFR estimations from these measured serum cystatin C values were calculated according to equations published by Filler and Lepage, Zappitelli et al., Larsson et al., Hoek et al., Rule et al. and Le Bricon et al. [6, 8, 16–19]. In addition, GFR estimation from cystatin C and creatinine was performed using the equations published by Schwartz et al. (New CKiD equation), Zappitelli et al. and Bouvet et al. (Table 1) [8, 11, 20]. The Bouvet equation was corrected for body surface area.

Table 1 Glomerular filtration rate estimating equations
Full size table

Demographic and clinical covariates were obtained by chart review and included disease diagnosis, age, race, gender and serum creatinine. Serum creatinine values were included only if obtained on the same day of the NucGFR measurement, a criterion met by 87 subjects (62 %) from our cohort. Serum creatinine was measured locally by an isotope dilution mass spectrometry (IDMS)-traceable enzymatic assay (Fusion Chemistry Analyzer; Ortho Vitros Diagnostics, Raritan, NJ). The study was approved by the Institutional Review Board who waived the requirement for informed consent/assent as this study was a retrospective analysis of previously obtained serum samples.

All analyses were conducted using SAS statistical software (ver. 9.2; SAS Institute, Cary, NC). Descriptive statistics for continuous variables were reported as medians and interquartile ranges (IQR). For the comparisons of cystatin C-based equations to NucGFR measurements, we calculated the mean bias (mean difference between estimated GFR and NucGFR) and 95 % limits of agreement (LOA) according to the methods of Bland–Altman [21]. Sensitivity and specificity for detecting a NucGFR of <90 ml/min/1.73 m2 (to define CKD) were also evaluated. Additionally, the proportion of the estimated (e) GFR for each formula within 10 and 30 %, respectively, of the NucGFR (accuracy) were reported, as well as the proportion of variability (R 2).

Results

Patient characteristics

A total of 141 patients were included in the analysis and their demographic and clinical characteristics are summarized in Table 2. Indications for obtaining NucGFR measurements included liver transplantation, malignancy, kidney transplantation, urologic abnormalities, hematopoietic stem cell transplantation (HSCT) and other miscellaneous conditions.

Table 2 Baseline characteristics of patient cohort
Full size table

All liver transplant recipients (n = 62) were evaluated in the outpatient setting, and for 58 of these patients the post-transplant period was longer than 1 year. Only one subject showed evidence of significant liver dysfunction (increased prothrombin time and bilirubin) and was subsequently admitted 1 week after the GFR measurement for abdominal pain and ascites. Immunosuppressive medication use in the liver transplant recipients included tacrolimus (n = 51), mycophenolate mofetil (n = 12), cyclosporine (n = 6), steroids (n = 6), sirolimus (n = 4) and azathioprine (n = 1).

Of the 37 malignancy patients, 31 were evaluated as outpatients and six were admitted when the GFR measurement was performed. All of these six patients were admitted either after their initial diagnosis and/or for chemotherapy, except one who was admitted with fever. Seventeen patients were actively receiving or had completed chemotherapy within 2 months of the time of GFR measurement. Twenty-seven patients were diagnosed with a malignancy within 1 year of the GFR measurement, while ten were over 1 year removed from the initial diagnosis.

All kidney transplant recipients were evaluated as outpatients, and all but one were over 1 year post-transplant. Immunosuppressive medication use included sirolimus (n = 8), tacrolimus (n = 5), cyclosporine (n = 2), mycophenolate mofetil (n = 10), steroids (n = 5) and azathioprine (n = 3). Five of six HSCT recipients received autologous stem cell transplants for neuroblastoma. Three patients were evaluated as outpatients and three as inpatients, two of whom had received chemotherapy for relapsed disease.

The remainder of our cohort was comprised of ten patients with urologic disease (six with cloacal anomalies or anorectal malformations) and 12 patients with other diseases (nine with underlying hematologic disease and undergoing evaluation prior to HSCT). Only three of these remaining patients were hospitalized at the time of GFR measurement.

Comparison of cystatin C-based equations in different disease categories

Cystatin C-based equations were compared to NucGFR measurements in patients with different underlying diseases (Table 3). Of previously published equations, the Rule formula performed best, accurately estimating GFR in all patients except liver transplant recipients, in whom there was about an 8 ml/min/1.73 m2 underestimation of GFR. The Zappitelli, Larsson, Hoek, and Le Bricon equations also performed reasonably well, although the GFR was significantly overestimated in urologic and oncology patients. The Filler equation consistently overestimated renal function in all patient categories. Among those equations including cystatin C and creatinine (Table 4), the New CKiD formula significantly outperformed the Bouvet and Zappitelli equations, which overestimated GFR in most patient categories.

Table 3 Comparison of cystatin C equations to nuclear medicine glomerular filtration rate (NucGFR) among different disease categories
Full size table
Table 4 Comparison of cystatin C and creatinine estimating equations to nuclear medicine glomerular filtration rate (NucGFR) among different disease categories
Full size table

Comparison of cystatin C-based equations and creatinine-based equations in the entire cohort

The performance of each formula based on mean bias, 95 % LOA, accuracy, sensitivity and specificity for estimating NucGFR and predicting renal insufficiency (NucGFR <90 ml/min/1.73 m2) in the entire cohort is summarized in Table 5. The only established formulas that did not have a significant mean bias were the Rule and Hoek formulas. The Zappitelli (including only cystatin C) and New CKiD equations also performed well, with mean biases of only 5.9 and 4.2 ml/min/1.73 m2, respectively. All other formulas more significantly overestimated GFR, especially the Filler equation and creatinine-based GFR formulas derived by Bouvet and Zappitelli (mean biases ranging from 15.8 to 27.5 ml/min/1.73 m2).

Table 5 Overall performance of glomerular filtration rate (GFR) estimating equations
Full size table

Discussion

We evaluated the accuracy and performance of cystatin C equations in a pediatric cohort at high risk of renal injury. The majority of our subjects did not have significant underlying renal disease at the time of assessment, as evidenced by the cohort median GFR of >90 ml/min/1.73 m2. We found that the Rule, Hoek, Zappitelli and New CKiD equations provided reasonably accurate assessments of GFR in our patients. In contrast, the Filler equation and the other creatinine-based equations (Bouvet and Zappitelli combined formula) significantly overestimated the GFR in our subjects.

The limitations of using only creatinine to estimate GFR are well documented and include variability due to gender, age, tubular secretion and muscle mass [22]. To improve the estimation of kidney function, other investigators have developed cystatin C-based equations in children and adults with known kidney disease [6, 8, 11, 16–20]. Many of these adult studies have rigorously assessed GFR estimating equations in liver and kidney transplant recipients, as well as patients with malignancy (Table 6). However, the formulas published to date have not been analyzed or validated in a large group of pediatric patients with similar high-risk conditions.

Table 6 Summary of literature of cystatin C-based estimating equations in transplant and other high-risk patients
Full size table

In the pediatric liver transplant population, Samyn et al. measured renal function in 62 children but did not report cystatin C-based GFR estimation [23]. Berding et al. reported on 48 pediatric liver transplant recipients and found reasonable performance of cystatin C, but assessed only the Filler equation [24]. Several additional pediatric studies have included a variable proportion of transplant recipients, including those of Grubb et al. [7] (unclear how many transplant patients), Filler et al. [6] (only 5.4 % received a kidney transplant), Bouvet et al. [20] (48 % of cohort were kidney transplant recipients) and Zappitelli et al. [8] (27 % kidney transplant, 7 % ‘other’ transplants). Finally, we identified one pediatric study comparing the Filler cystatin C equation to the new Schwartz formula in 68 children with malignancy [25].

Recently, Bacchetta et al. published a well-designed analysis validating cystatin C-based equations in 252 children at high risk of renal injury [26]. However, in contrast to our cohort, most of these children had primary renal disease, and only 26 subjects (10 %) received a non-renal organ transplant. The authors concluded that the Le Bricon, Larsson, Rule and both Zappitelli formulas provided the most accurate agreement with inulin clearance, whereas the Filler equation overestimated GFR and the Bouvet formula underestimated GFR. Similarly, in our study, the Rule and Zappitelli equations performed reasonably well, although the Larsson and Le Bricon formulas significantly overestimated GFR (mean bias of 10.2 and 8.0 ml/min/1.73 m2, respectively).

Equations including cystatin C and creatinine have demonstrated improved accuracy and precision compared to those including only cystatin C [8, 20]. The Bouvet and combined Zappitelli formulas, however, markedly overestimated GFR in our cohort. Systematic biases evident in these formulas may be secondary to variation in creatinine assay methods or endogenous creatinine production between study populations. For example, creatinine-based GFR estimations have overestimated renal function in pediatric liver transplant recipients and children with malignancy [25, 27], possibly due to low serum creatinine from muscle wasting. Additionally, Bouvet et al. used a non-compensated Jaffe method for creatinine measurement, whereas all others (including the present study) used an enzymatic assay. Smaller creatinine values yielded by enzymatic techniques likely contributed to the overestimation of GFR by the Bouvet formula. These limitations of creatinine notwithstanding, the New CKiD formula performed quite well, demonstrating the best 30 and 10 % accuracy among all formulas. Our study is the first, therefore, to validate its use in a cohort composed primarily of children with malignancy or solid organ transplantation.

The strengths of our analysis include the use of an established nuclear medicine technique as the gold standard and measuring cystatin C on the same sample as the NucGFR, thereby reducing intra-patient variability which is often a concern with these measurements [28]. Furthermore, we used an accurate nephelometric assay for the measurement of cystatin C, which has been shown to be more reliable than other techniques, such as those based on turbidimetry [1]. In addition, we used an IDMS-traceable, enzymatic assay to measure serum creatinine, similar to those used by Zappitelli et al. and the 2009 CKiD study. Our patient population had a wide range of GFR (17–147 ml/min/1.73 m2), whereas prior studies have been limited by including patients with narrower ranges of renal function (Table 6). Finally, we included all patients having a formal GFR measurement at our institution, reducing selection bias. As these tests were ordered for clinical indications, our cohort represented a heterogeneous group of patients at high risk for kidney dysfunction.

Several limitations of our study deserve consideration. First, our cohort was composed of many liver transplant recipients and patients with a history of malignancy. In these subjects, variations in endogenous cystatin C production may have affected the performance of cystatin C-based estimations. For example, elevated cystatin C levels have been demonstrated in patients with active liver disease (cirrhosis, hepatitis) and leukemia [29–31], potentially underestimating GFR when cystatin-C based equations are used. However, cystatin-C based equations have not demonstrated a consistent bias in clinically stable liver transplant recipients and patients with malignancy, although variability in performance has been demonstrated among individual formulas [24, 25, 32–34]. Second, we used 99mTc-DTPA plasma disappearance curves as our ‘gold standard’ of GFR measurement, thereby differing from previous studies that measured GFR by iohexol or iothalamate clearance. However, GFR estimation using 99mTc-DTPA has demonstrated close agreement with iohexol, iothalamate and inulin [15, 35, 36]. Lastly, our last plasma sample was drawn at 210 min post-injection, whereas other studies have used longer intervals of up to 300 min [37]. Shorter sampling intervals can cause an overestimation of GFR in patients with severe renal dysfunction. However, as only one patient in our cohort had a GFR of <30 ml/min/1.73 m2, this effect was likely to be insignificant.

In summary, our results demonstrate that the Rule, Hoek and Zappitelli cystatin C-based equations and the New CKiD formula provided a good estimation of the GFR in children at high risk of developing renal insufficiency, such as after transplant. While these GFR estimating equations are unlikely to replace gold standard methods in all clinical situations, they offer reasonable accuracy, improved cost and patient convenience compared to formal isotope techniques.