Maturation of GFR in Term-Born Neonates: An Individual Participant Data Meta-Analysis

Significance Statement

The evidence from individual studies to support the maturational pattern of GFR in healthy, term-born neonates is inconclusive. This paper describes GFR reference values in the first month of life using an individual participant data meta-analysis of reported measured GFR (mGFR) data using data from 881 neonates. GFR doubled in the first 5 days after birth, from 19.6 to 40.6 ml/min per 1.73 m², and then more gradually increased to 59.4 ml/min per 1.73 m² by 4 weeks of age. GFR was best estimated by 0.31×height (cm)/serum creatinine (mg/dl). These mGFR reference values and more accurate GFR estimations can help to identify altered GFR in term-born neonates; however, further validation of the eGFR equation is needed.

Visual Abstract

Abstract

Background

The evidence from individual studies to support the maturational pattern of GFR in healthy, term-born neonates is inconclusive. We performed an individual participant data (IPD) meta-analysis of reported measured GFR (mGFR) data, aiming to establish neonatal GFR reference values. Furthermore, we aimed to optimize neonatal creatinine-based GFR estimations.

Methods

We identified studies reporting mGFR measured by exogenous markers or creatinine clearance (CrCL) in healthy, term-born neonates. The relationship between postnatal age and clearance was investigated using cubic splines with generalized additive linear mixed models. From our reference values, we estimated an updated coefficient for the Schwartz equation (eGFR [ml/min per 1.73 m²]=(k×height [cm])/serum creatinine [mg/dl]).

Results

Forty-eight out of 1521 screened articles reported mGFR in healthy, term-born neonates, and 978 mGFR values from 881 neonates were analyzed. IPD were available for 367 neonates, and the other 514 neonates were represented by 41 aggregated data points as means/medians per group. GFR doubled in the first 5 days after birth, from 19.6 (95% CI, 14.7 to 24.6) to 40.6 (95% CI, 36.7 to 44.5) ml/min per 1.73 m², and then increased more gradually to 59.4 (95% CI, 45.9 to 72.9) ml/min per 1.73 m² by 4 weeks of age. A coefficient of 0.31 to estimate GFR best fitted the data.

Conclusions

These reference values for healthy, term-born neonates show a biphasic increase in GFR, with the largest increase between days 1 and 5. Together with the re-examined Schwartz equation, this can help identify altered GFR in term-born neonates. To enable widespread implementation of our proposed eGFR equation, validation in a large cohort of neonates is required.

Measuring GFR is an essential part of kidney function evaluation. Solid reference values are needed to be able to recognize acute and chronic kidney impairment, and to define safe and effective dosing regimens for drugs cleared by glomerular filtration.

In clinical practice, GFR is generally estimated from plasma concentrations of endogenous markers. This approach may be inaccurate, however, because endogenous markers never fulfill all criteria of a perfect GFR marker, and, soon after birth, plasma concentrations of these biomarkers partially reflect maternal concentrations.^1,2 In addition, the creatinine-based eGFR calculations, which are regularly used in neonatal clinical care, still rely on the equation proposed by Schwartz et al.³ in 1987 using a coefficient (k) of 0.45. On the basis of the clearance of renally excreted drugs, a lower k of 0.277 has been suggested for non–low-birth-weight infants,⁴ but this coefficient remains to be validated against mGFR values.

Alternatively, GFR can be measured by determining plasma or urinary clearances after administration of an exogenous marker, such as inulin, radioisotopes such as technetium-99m–diethylene triamine penta-acetic acid (^99mTc-DTPA) and chromium-51–EDTA (⁵¹Cr-EDTA), mannitol, and iohexol. Although minor biases between the clearance of these markers and renal inulin clearances exist,⁵ all exogeneous markers are considered accurate to determine GFR. They were validated against gold-standard inulin clearances, and very good agreement was demonstrated in both the adult^6–9 and pediatric population.^10–12 Thus, for all markers, their clearance is a true reflection of GFR, even in neonates. In addition, creatinine clearance (CrCL), which incorporates both plasma and urinary creatinine levels, approximates the inulin clearance in children, and only at very low GFR values (<21 ml/min per 1.73 m²) does CrCL significantly overestimate measured GFR (mGFR) due to the larger contribution of tubular secretion.¹³

Thus, preferably, reference values for neonatal GFR are determined on the basis of mGFR. In contrast to term-born neonates, for preterm neonates (born between 27 and 31 weeks of gestation) and children >1 month of age, mGFR reference values are already well described using CrCL¹⁴ or ⁵¹Cr-EDTA.¹⁵ However, for healthy, term-born neonates, knowledge regarding the maturation of GFR in the first month of life is inconclusive. Reported mean mGFR values in healthy, term-born neonates range from approximately 15 to 50 ml/min per 1.73 m2,^16–18 but these values were established in small studies, which showed greatly differing maturational patterns. Furthermore, available and often-cited review articles report GFR to be as low as 2–4 ml/min per 1.73 m².^19,20

To establish the maturational pattern of GFR and to define reference values of mGFR, we performed an individual participant data (IPD) meta-analysis (IPDMA) of reported mGFR data of healthy, term-born neonates (>37 weeks of gestation) aged 0–28 days of postnatal age. This method allows for generating evidence by assembling, checking, and reanalyzing individual-level data, and is considered the gold standard of evidence.²¹

As a secondary objective, we aimed to estimate an updated coefficient (k) for the Schwartz equation (eGFR [ml/min per 1.73 m²]=[k×height (cm)]/serum creatinine [mg/dl]) to facilitate more accurate eGFR determination in term-born neonates.

Methods

The method and results of this IPDMA are presented according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses–IPD guidelines.²¹ Four different cohorts were used for our analysis (Table 1), which served two goals. First, to establish mGFR reference values. Second, to update the coefficient in the Schwartz equation. With regards to the first goal, reference values were established using IPD and validated by ten-fold crossvalidation. We will refer to this IPD-based model as our original model. These reference values were additionally validated and verified by sensitivity analyses and by combining IPD and aggregated data. For our second goal, we determined the k-value using normative creatinine values as reported by Boer et al.¹ with height data from growth charts.²² Last, we prospectively validated our k-value in a cohort of 43 neonates with critical illness.

Table 1.

Specification of cohorts used in the different steps of our analysis

Subject	Aim	Cohort	Included Neonates (n)	Statistical Methods
GFR reference values	Determination and internal validation of mGFR reference values	(1) IPD: Individual mGFR data	367	Generalized linear mixed model to model the association between mean mGFR and postnatal age; ten-fold crossvalidation of the mean GFR reference values; sensitivity analyses to determine the robustness of our findings, by adding an interaction term between postnatal age and a covariate to the original model, for several covariates
	Verification of mGFR reference values	(2) IPD and aggregated mGFR data	881	Visual comparison between predicted values
k-Value determination	Development of new k-value	(3) Population creatinine concentrations as reported by Boer et al.1 combined with height data from CDC growth charts	Population approach	Linear regression using least-square analysis
	External validation of k-value	(4) Individual creatinine and mGFR data measured and estimated in term-born neonates as part of the HERO study	43	Determining agreement by using accuracy and precision

Protocol and Registration

The methods of our analysis and inclusion and exclusion criteria were specified in advance and registered in PROSPERO, an international prospective register of systematic reviews, in September 2020. The protocol was registered on November 28, 2020 under registration number CRD42020217880.

Eligibility Criteria

To be eligible, the following inclusion criteria were applied: publications should comprise original research articles, of which the study population contains at least one term-born neonate (aged <29 days). Case reports were excluded. Furthermore, the article should contain mGFR data determined by urinary CrCL and/or using one of the following markers: inulin, iohexol, iothalamate, mannitol, ⁵¹CR-EDTA, or ^99mTc-DTPA. These criteria were applied at the participant level, enabling us to include subjects from studies which recruited a wider population than that were specified in our review inclusion criteria.

Identifying Studies

We searched the PubMed database for literature on October 30, 2020, using the following search strategy: ((glomerular filtration rate[MeSH Terms] AND “infant, newborn”[MeSH Terms]) AND humans[MeSH]) OR ((“iohexol”[Title/Abstract] OR “iothalamate”[Title/Abstract] OR “inulin”[Title/Abstract] OR “mannitol”[Title/Abstract] OR “CR-EDTA”[Title/Abstract] OR “TC-DTPA”[Title/Abstract] OR “creatinine clearance”[Title/Abstract] OR “clearance creatinine”[Title/Abstract]) AND (“neonate”[Title/Abstract] OR “newborn”[Title/Abstract] OR “infant”[Title/Abstract])) OR (measured AND “glomerular filtration rate” AND neonate) OR ((newborn* OR new-born* OR perinat* OR neonat* OR baby OR baby* OR babies) AND measured GFR) OR ((Kidney[ti] OR renal[ti] OR measur*[ti]) AND (neonate[ti] OR neonates[ti] OR infant[ti] OR infants[ti] OR child*[ti]) AND (clearance[ti] OR glomerular filtration rate[ti])) and human[MeSH]. No date or language restrictions were applied. Search results were imported in Rayyan to accommodate systematic and fast assessment of each title and abstract.²³ Two individual reviewers (N.J.L.S. and M.J.P.v.d.B.) independently assessed the abstracts and selected relevant abstracts for further assessment. If no abstract was available, the full text of the article was requested and assessed. Four independent exclusion criteria were applied and each excluded abstract was allocated one of the following criteria: study not containing original research data, study not containing term-born neonates, study not reporting mGFR data, and study not including data on healthy neonates (only the first criterion that applied was listed). Any discrepancy of opinion regarding the assessment of the selection criteria was resolved by discussion by the two reviewers. Also, the Clinical Trials.gov database was searched to identify ongoing or unpublished studies. Of the included abstracts, the full-text versions were retrieved to decide upon final inclusion. Full texts were translated if not written in English to decide upon eligibility. References of included studies were checked to complete our database.

Data Collection Process

For each potentially relevant article, the full-text article was retrieved by checking university libraries in The Netherlands, Germany, and Belgium. If no full-text article could be retrieved, the article was excluded for further analysis. For every relevant article, IPD were extracted by the coordinating investigator. When only graphs were reported, IPD were extracted using WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/). If only aggregated data were presented in the article and the publication date of the manuscript was <25 years ago (after 1995), the corresponding author of the selected study was contacted and asked to provide the IPD (of term-born neonates only). A reminder was sent twice to these authors. If the IPD could not be retrieved, the aggregated data listed in the articles were included in our sensitivity analysis using simulated data on the basis of the reported distribution (mean±SD) and study size of the reported data. Data were only included in our analysis if we were certain they represented data derived from term-born neonates only. Also, if GFR was only reported in milliliters per minute without traceable weight, height, and/or body surface area (BSA), and the publication date before 1995, studies were excluded from our analysis. If authors did not report GFR in milliliters per minute per 1.73 m², these values were calculated on the basis of the uncorrected clearance (in milliliters per minute) and BSA. BSA was calculated using the Haycock formula (BSA [m²]=0.024265× weight^0.5378 [kg]×height^0.3964 [cm]) because this equation was demonstrated to be the most suitable for neonates.²⁴ However, if only weight was reported along with GFR in milliliters per minute, the corresponding height was derived from the Centers for Disease Control and Prevention (CDC) growth charts on the basis of the particular weight percentile.²² If sex was not specified, the boys chart was used. We used CDC growth charts because they specified height for children aged 0, 0.5, and 1 month of age, whereas the World Health Organization (WHO) charts only specify height for full months.²⁵ Also, differences between both charts are minor; for instance, when looking at the chart for boys at 0 months of age, the 50th percentile (p50) for height was listed as 49.9 cm on the WHO chart and 49.99 cm on the CDC chart.

Data Items

The following variables were retrieved from the original research studies. Study-specific data consisted of year of publication (year), total number of inclusions, total number of included term-born neonates, and GFR marker used (inulin, iohexol, mannitol, ⁵¹Cr-EDTA, ^99mTc-DTPA, or CrCL). If CrCL was presented, the analytic method of measuring creatinine was noted to allow analysis between Jaffe assays and other enzymatic creatinine assays. Participant-specific data included gestational age (weeks), postnatal age (days), weight (grams), height (centimeters), BSA (square meters), sex (boys/girls), health status (according to authors: healthy, disease without influence on kidney function, disease with potential influence on kidney function), race and ethnicity (White/Asian/Black/other), and GFR (in ml/min, ml/min per 1.73 m², or ml/min per m²).

IPD Integrity

All retrieved data were checked for consistency and reliability. More specifically, postnatal age, GFR values, weight, height, and BSA were visually displayed to assess inconsistencies. For any questionable data (i.e., values outside reasonable values for neonates), the original article was reviewed to confirm or correct these data. If GFR was only reported for part of the study population, the reason for this was assessed and data were discarded if there was suspicion of selection bias of included participants. When the total study population also included premature neonates or older infants, participants were not included in our analysis if postnatal age or gestational age were not specifically mentioned.

Assessing Risk of Bias in Individual Studies

A risk assessment at the study level using the Cochrane risk-of-bias tool²⁶ was not performed because the effect of a determinant on the clinical condition of a participant was not relevant for answering our research question. Potential confounders for the relation between postnatal age and GFR, as described under Data Items, were investigated as possible sources of bias.

Specification of Outcomes and Effect Measures

There was no minimum number of required studies because we aimed to give an overview of all available studies. Reference values were reported by day using estimated means, including 95% CIs and percentiles (p10, p25, p75, and p90) specified for postnatal age.

Synthesis Methods for Development and Validation of mGFR Reference Values

The primary analysis was conducted on IPD only, i.e., cohort 1 in Table 1. First, outliers were visually assessed and only excluded from our analysis if there was a serious suspicion of data errors in the initial study. In addition, extreme values, defined as values outside 3-SD boundaries, were excluded. After visually exploring our data, the data were analyzed using generalized additive mixed models (GAMMs), using postnatal age and a smoothing term based on cubic splines with five knots of age in combination with restricted maximum likelihood (ML) to model the relationship between GFR and postnatal age between day 0 and day 28. A normal (Gaussian) distribution for GFR was assumed. The within-study clustering of participants was taken into account by adding random effects for the study-specific intercept and slope of age in an additive way to the model. We assumed that the inclusion of the two random effects described above was sufficient to also adjust for this within-study correlation. This assumption was confirmed by similarity of the predicted clearance based on data with and without the possible duplicate values. Comparisons between the fit of different models (e.g., with and without random effect for slope) were made with likelihood ratio tests and comparison of Akaike information criteria, using models fitted with ML. As a next step, we internally validated the primary model using ten-fold crossvalidation. This is an internal validation method that uses different portions of the data to train and test a model.²⁷ Our goal was to test the model’s ability to predict mean clearance and to flag potential sources of error, including overfitting. To do so, our IPD dataset was randomly split into ten parts. Development (training) of our model then took place using the first nine parts only, after which validation occurred in the remaining tenth part. This was repeated ten times, each time using a different part of the ten partitions as the test dataset, and the remaining nine parts as training data. The model fit of the ten models was then compared with our original model based on IPD by comparing the variance, as estimated by the original model, to the sum of squared residuals in the crossvalidation, and by comparing the fitted curves.

Because the SD of the clearance values did not notably differ across the 4-week period, we used 1 SD for the percentile calculations. The clearance percentiles (p10, p25, p75, and p90) were determined on the basis of the square root of the sum of the squared SD and the squared SEM, multiplied by the appropriate z-value (e.g., 1.28 for the p90), added to the mean clearance estimate of the primary analysis. Equations predicting mean clearance by postnatal age were developed by fitting a generalized additive linear model onto the mean predicted clearance values, using combinations of transformations of postnatal age until an exact fit of the predicted mean clearance was found.

Because our data shows heterogeneity that could affect our reference values, we performed sensitivity analyses to explore the effect of several covariates. We excluded certain subgroups to check whether our developed model would stand. This included all CrCL values derived from neonates <3 days of age to investigate the effect of maternal creatinine, all mannitol values, all mGFR values for which height was imputed, and possible duplicates, respectively.

Exploration of Variation in Effects—Sensitivity Analysis

Next, we assessed the statistical significance of the GFR marker used (exogenous versus CrCL), publication year (<1980 or ≥1980), health status (healthy, disease without effect on GFR, disease with possible effect on GFR), race and ethnicity, and sex on clearance pattern. We compared the models with separate spline-based curves for the different levels of the possible effect modifiers with the primary model described above, using the ML estimation procedure in combination with likelihood ratio tests. The final estimates per level of the covariate were based on models fitted with restricted ML.

In the primary analysis described above, only IPD were included. We also conducted an additional analysis to verify our results, in which the aggregated data were also included (cohorts 1 and 2). This was performed under the condition that these data originated from studies with a precise indication of reported postnatal age. Because clearance changes rapidly in the first month of life, a small postnatal age SD was required to define precise studies (if <1 week of age, SD <1 day; if >1 week of age, SD <2 days). To include the aggregated data into the model, every aggregated point of a study of size N was replaced by N simulated individual data points, which together precisely reflected the mean and SD of the original study,²⁸ because we preferred to fit the GAMM in this combined analysis on individual data to prevent ad hoc weighing of the aggregated data. Because the simulated data exactly reflect the mean and SD of the studies, the resulting model is independent of the individual specific values of the simulated data. Because aggregated data are inherently less precise than IPD, we will base our final mGFR reference values on IPD alone. We believe this will enhance the precision of our reference values.

Risk of Bias across Studies

Because only healthy neonates were included, we consider the potential differences between studies negligible. To rule out any statistical difference between results obtained from IPD versus aggregated data, the analysis was performed twice by including IPD data only and including both IPD and aggregated data. Also, to systematically assess the risk of bias, we applied the Risk of Bias In Non-Randomized Studies – of Exposures (ROBINS-E) tool to systematically assess bias.²⁹

Development and Validation of k-Value

To support future estimation of GFR using serum creatinine concentrations, we determined the optimal coefficient (k) to use in the Schwartz equation (eGFR=[k×height (cm)]/serum creatinine [mg/dl]) by least-square analysis, comparing our mGFR data to eGFR data as calculated by the Schwartz equation.³⁰ For this approach, two new cohorts of neonates were used, one for development (cohort 3) and one for validation (cohort 4) (Table 1). First, development took place using normative enzymatic serum creatinine values, as reported by Boer et al.,¹ and median height, from CDC growth charts,²² as input for the eGFR calculations. Boer et al.¹ applied stringent exclusion criteria to ensure the selection of neonates with normal kidney function. Next, mGFR from our reference values was plotted with eGFR as a function of postnatal age using these normative, age-specific creatinine and height values. Different coefficients, ranging from 0.300 to 0.3500 (increments of 0.001) served as input for these calculations. A least-square analysis was then applied to determine the smallest sum of squares. After the determination of the most suitable k-value, data were validated. Because performing GFR measurements in healthy newborns is not ethical because it requires vascular access, we externally validated our findings in a cohort of neonates with critical illness. The methods of this single-center, prospective study were previously described in detail (unpublished data). In short, after obtaining signed informed consent from the guardians, we included 43 term-born neonates with critical illness who had an indwelling central venous or arterial line. After intravenous administration of iohexol, blood samples were drawn for analysis of iohexol concentrations at 2, 5, and 7 hours after administration. Two-point blood sampling at 2 and 5 hours after administration is a validated method for mGFR determination in children.³¹ To enhance accuracy, we added another sampling point at 7 hours after administration for neonates with a bodyweight of at least 3.5 kg. mGFR was then calculated on the basis of the ratio between the administered iohexol dose and the area under the plasma concentration time curve, corrected for BSA. Next, to assess agreement between eGFR based on our developed k-value and mGFR, bias and accuracy were calculated. First, to evaluate the difference between eGFR and mGFR, GFR data were analyzed by calculating the difference between eGFR and mGFR per patient and determining the median of this difference (bias). Comparison of eGFR and mGFR values on a group level was performed using the Wilcoxon signed-rank test for paired data. Accuracy was calculated as the percentage of patients having a similar eGFR when compared with mGFR. Similarity was a priori considered as individual eGFR values that differed <30% from mGFR values.^32,33 To enable comparison of our k-value with existing k-values used in neonates by Schwartz et al. (k=0.45)³ and Zhang et al. (k=0.277),⁴ we included these values in our analysis.

Data were analyzed using IBM SPSS statistical software (version 25.0.0.1) and the statistical software R (version 4.0.3).³⁴ Aggregated data were expanded using the R package SimComp (version 3.3),³⁵ and the GAMMs were fitted using the R package mgcv (version 1.8-33).³⁶ Graphics were created using the R package ggplot2 (version 3.3.3).³⁷

Results

Study Selection

We identified 1521 original research articles via PubMed that fulfilled selection criteria for our main analysis (Figure 1). There were no ongoing or unpublished studies listed at ClinicalTrials.gov.

Figure 1.

Flowchart of studies included in our IPDMA. In total, 881 neonates with 978 GFR measurements were included.

Title and abstract screening resulted in the selection of 105 possibly relevant articles, of which 30 contained mGFR data in healthy neonates. Reference checking revealed another 21 relevant articles.

IPD Integrity

Data from these 51 studies were assessed for reliability. After visual inspection of the data, the study from Mannan et al.,³⁸ including 53 neonates, was excluded, because BSA-corrected mGFR values were <8 ml/min per 1.73 m² and not in line with the reported serum creatinine levels in the same article. Two studies with aggregated data, including 30 and 33 neonates, were excluded due to imprecise reporting of postnatal age.^39,40

Thus, 48 articles were available for analysis. Three individual mGFR values from three different studies were defined as extreme values, and these values were, therefore, excluded from our analysis (Figure 1).

Study Characteristics and IPD Obtained

Thirty studies reported 375 mGFR measurements for 367 neonates. The other 18 studies presented aggregated data only. Because clearance was specified for multiple age groups in some studies, these studies presented aggregated data by 41 separate data points. Together, these summarizing data included 603 mGFR measurements in 514 neonates. Through combining individual and aggregated data, 978 mGFR measurements from 881 neonates were available. Only four studies specifically mentioned race, which was reported as White (n=2) or other (n=2). Thirteen studies reported sex. Most studies (76%) included neonates described as being healthy. Some studies did not report health status (3%), and a few summarized the clinical status of included neonates as having no potential influence on renal function (6%), or having a possible effect (15%). From all included IPD (n=367), we estimated BSA for 28 neonates only (7.6%). For these neonates, BSA was calculated on the basis of imputed heights and reported weights. Risk of bias according to the ROBINS-E tool was considered low, and the direction of bias was possibly toward underestimation of our mGFR data (Supplemental Table 1). Studies had been published in the period 1948–2014 and included one up to 82 neonates. There was a skewed age distribution toward the younger neonates, with 48% of all mGFR measurements involving neonates <3 days of age. Twenty two of all studies (46%) used inulin infusions to calculate GFR, three (6%) used mannitol, and one (2%) used ⁵¹Cr-EDTA. Twenty-two studies (46%) used CrCL, with either Jaffe assays or modified Jaffe assays, and no enzymatic creatinine assays were reported. Characteristics of each study—including author, country of origin, method of reporting GFR, year of publication, specifics of the included population with health status, and GFR marker used—are listed in Table 2.

Table 2.

Studies reporting mGFR values in term-born neonates

Author(s), Year	Neonates Included (n)	Measurements Included (n)	Age Range (d)	Health Status	GFR Marker Used	Method of Presenting GFR	Individual or Aggregated Data (n data points for aggregated data)	City and Country of Origin
Al-Dahhan et al., 198353	6	6	5–6	Healthy	CrCLa	ml/min per 1.73 m2	Individual	London, United Kingdom
Aperia et al., 197254	5	5	3–15	Healthy	Inulin	ml/min per 1.73 m2	Individual	Stockholm, Sweden
Aperia et al., 197555	9	9	3–21	Disease without effect on GFR	Inulin	ml/min per 1.73 m2	Individual	Stockholm, Sweden
Aperia et al., 198156	15	15	2–5	Healthy	Inulin	ml/min per 1.73 m2	Individual	Stockholm, Sweden
Arant, 197857	26	26	1	Healthy	CrCLa	ml/min + weight	Aggregated (1)	Memphis, United States
Boehm et al., 198458	44	44	2–19	Healthy	Mannitol	ml/min per 1.73 m2	Individual	Leipzig, Germany
Broberger, 197359	15	15	3–14	Not mentioned	Inulin	ml/min per 1.73 m2	Individual	Stockholm, Sweden
Broberger and Aperia, 197960	14	14	4–5	Healthy	Inulin	ml/min per 1.73 m2	Aggregated (2)	Stockholm, Sweden
Brodehl and Gellissen, 196861	5	5	16–26	Disease without effect on GFR	Inulin	ml/min per 1.73 m2	Individual	Bonn, Germany
Brodehl et al., 198262	8	8	14–27	Disease without effect on GFR	Inulin	ml/min per 1.73 m2	Individual	Hannover, Germany
Bueva and Guignard, 199442	28	46	1–16	Healthy	CrCLb	ml/min per 1.73 m2	Aggregated (3)	Lausanne, Switzerland
Casimiro Pantoja et al., 197763	7	7	4–7	Healthy	Inulin	ml/min per 1.73 m2	Individual	Unclear, Mexican journal
Coulthard, 198564	5	5	2–6	Healthy	Inulin	ml/min + weight	Individual	Newcastle upon Tyne, United Kingdom
Dean and McCance, 194765	8	8	2–8	Disease with possible effect on GFR	Inulin	ml/min per 1.73 m2	Individual	Cambridge, United Kingdom
Edelmann et al., 196066	13	13	1–28	Not mentioned	Inulin	ml/min per 1.73 m2	Individual	New York City, United States
Fawer et al., 197918	13	13	1–19	Healthy	Inulin	ml/min per m2	Individual	Lausanne, Switzerland
Fawer et al., 197967	31	31	1–9	Disease with possible effect on GFR	Inulin	ml/min per m2	Aggregated (2)	Lausanne, Switzerland
Friederiszick, 195468	1	1	21	Healthy	Inulin	ml/min per 1.73 m2	Individual	Mainz, Germany
Gekle et al., 196769	3	3	14–21	Healthy	Inulin	ml/min per 1.73 m2	Individual	Würzburg, Germany
Godard et al., 197970	19	19	1–8	Healthy	CrCLa	ml/min per 1.73 m2	Aggregated (2)	Geneva, Switzerland
Gordjani et al., 198871	39	39	1–9	Healthy	CrCLb	ml/min per 1.73 m2	Aggregated (4)	Marburg, Switzerland
Gubhaju et al., 201443	20	31	3–28	Healthy	CrCLc	ml/min per m2	Aggregated (2)	Brisbane, Australia
Guignard et al., 197517	10	10	1–19	Disease without effect on GFR	Inulin	ml/min per m2	Individual	Lausanne, Switzerland
Haimi-Cohen et al., 199772	37	37	3	Disease with possible effect on GFR/healthy	CrCLb	ml/min per 1.73 m2	Aggregated (2)	Tel Aviv, Israel
Kapoor et al., 198873	20	20	1	Healthy	CrCLb	ml/min per 1.73 m2	Aggregated (1)	Lucknow, India
Leake et al., 197641	6	6	7–28	Healthy	Inulin	ml/min per 1.73 m2	Individual	Torrance, United States
Leake and Trygstad, 197774	14	14	0–3	Disease with possible effect on GFR/healthy	Inulin	ml/min + weight	Individual	Torrance, United States
Leititis et al., 198744	82	82	1–9	Healthy	CrCLc	ml/min per 1.73 m2	Aggregated (4)	Marburg, Germany
Maruyama, 196575	7	7	4–27	Healthy	CrCLa	ml/min per 1.73 m2	Individual	City unknown, Japan
McCrory et al., 195276	14	14	5–19	Disease with possible effect on GFR/healthy	Inulin	ml/min + BSA	Individual	New York City, United States
Oh et al., 196677	66	66	0–5	Healthy	Inulin	ml/min + weight and height	Individual	Stockholm, Sweden
Olavarria et al., 198678	20	20	1	Healthy	CrCLb	ml/min per 1.73 m2	Aggregated (1)	Valdivia, Chile
Nair et al., 198779	48	48	1–14	Healthy	CrCLb	ml/min per 1.73 m2	Aggregated (3)	Hyderabad, India
Passwell et al., 197480	15	15	3–4	Healthy	CrCLc	ml/min + weight	Individual	Tel-Hashomer, Israel
Peters et al., 199481	6	6	9–28	Disease without effect on GFR	51Cr-EDTA	ml/min per 1.73 m2	Individual	London, United Kingdom
Richmond et al., 195182	4	4	3–24	Healthy	Inulin	ml/min per 1.73 m2	Individual	Chicago, United States
Robinson et al., 199083	2	2	1	Healthy	Inulin	ml/min + weight	Individual	Iowa city, United States
Rubin et al., 194916	11	11	2–22	Healthy	Mannitol	ml/min per 1.73 m2	Individual	Buffalo, United States
Sawa et al., 197484	30	60	1–4	Healthy	CrCLa	ml/min per 1.73 m2	Aggregated (2)	Warsaw, Poland
Schwartz et al., 198430	9	9	5	Healthy	CrCLa	ml/min per 1.73 m2	Aggregated (1)	New York City, United States
Sertel and Scopes, 197385	17	25	1–6	Healthy	CrCLb	ml/min per 1.73 m2	Individual	London, United Kingdom
Siegel and Oh, 197686	21	21	1	Healthy	CrCLa	ml/min per 1.73 m2	Individual	Torrance, United States
Stapleton, 198387	15	15	1	Healthy	CrCLa	ml/min + weight	Aggregated (1)	Memphis, United States
Sulyok and Guignard, 199088	22	22	7	Healthy	CrCLb	ml/min per 1.73 m2	Aggregated (2)	Lausanne, Switzerland
Vio et al., 198789	14	14	1	Healthy	CrCLb	ml/min per 1.73 m2	Aggregated (1)	Valdivia, Chile
Watanabe et al., 198990	10	70	1–7	Healthy	CrCLa	ml/min + weight	Aggregated (7)	Osaka, Japan
West et al., 194891	4	4	8–25	Disease without effect on GFR	Mannitol	ml/min per 1.73 m2	Individual	New York City, United States
Winberg, 195992	13	13	4–12	Disease without effect on GFR	CrCLa	ml/min per 1.73 m2	Individual	Stockholm, Sweden
Total (N=48)	881	978	0–28

^aCompensated Jaffe method used for creatinine determination.

^bJaffe method used for creatinine determination.

^cUnkown method used for creatinine determination.

Results of Syntheses: mGFR Results

Our model was based on a linear part combined with a smooth part, categorized by four trajectories (Supplemental Figure 1). The smooth term for postnatal age was statistically significant (P<0.001). The following coefficients were estimated for the intercept, linear slope, and the four smooth terms, respectively: intercept, 27.79 (SEM, 2.58); age, 1.14 (SEM, 0.30); s(age).1, 7.39 (SEM, 1.72); s(age).2, 9.99 (SEM, 2.22); s(age).3, 3.02 (SEM, 4.93); s(age).4, 0.00 (SEM, 0.00).

GFR increases rapidly during the first 5 days of life, and then rises more gradually until the fourth week of life (Figure 2, Supplemental Table 2). On average, GFR increases in the first 5 days after birth from a mean value of 19.6 (95% CI, 14.7 to 24.6) ml/min per 1.73 m² at the first day to 40.6 (95% CI, 36.7 to 44.5) ml/min per 1.73 m² at day 5, followed by a more gradual increase to 59.4 (95% CI, 45.9 to 72.9) ml/min per 1.73 m² at the end of the fourth week. Ten-fold crossvalidation confirmed the robustness of our model (Supplemental Figures 2 and 3). Furthermore, results from all sensitivity analyses were similar to the results of the original model based on IPD alone (Table 3, Supplemental Figures 4–7).

Figure 2.

The development of GFR in the first month of life in term-born neonates. Solid black line represents median mGFR; the darker blue shading indicates area between p25 and p75; the lighter blue shading between p10 and p90.

Table 3.

Results of sensitivity analyses

Aim	Excluded Value or Comparison Group	Results
To explore effect of heterogeneity	Excluded values	Visual representation in Supplemental Figures 4–7
	All values measured with mannitol
	All CrCL values derived from neonates <3 days of age
	All mGFR values for which height was imputed
	Possible duplicates (day 6 data from Sertel and Scopes85 study)
To assess differences between groups	Compared groups
	mGFR as measured by exogenous markers versus CrCL	Supplemental Figure 8 (P=0.28)
	Health status	Supplemental Figure 9 (P=0.74)
	Sex	Supplemental Figure 10 (P=0.29)
	Year of publication	Supplemental Figure 11 (P=0.37)
To verify mGFR reference values on aggregated data	IPD mGFR values versus IPD + aggregated mGFR values	Supplemental Figure 12

Abbreviations:.

GFR measured with CrCL did not significantly differ from GFR measured with the exogenous markers (inulin, mannitol, and ⁵¹Cr-EDTA; P=0.28; Supplemental Figure 8). On the basis of mean GFR, until 7 days of age, GFR measured by CrCL seemed lower compared with GFR measured by exogenous markers. After 7 days of age, this relation is exactly opposite: mean GFR as measured by CrCL seemed higher than median GFR based on exogenous markers. Also, health status, sex, and year of publication did not significantly influence the relationship between postnatal age and clearance (P=0.74, P=0.29, and P=0.37, respectively; Supplemental Figures 9–11). Due to the small number of studies reporting race, the effect on GFR could not be assessed. Verification of the GFR maturational pattern by aggregated data confirmed our results (Supplemental Figure 12).

The predicted mean clearance between 0 and 28 days of postnatal age was well described by an equation using postnatal age (Equations 1–3), in which age is postnatal age in days.

• Equation 1. Predicted mean clearance (CL; ml/min per 1.73 m²) for neonates <6 days of age:

  $$\begin{array}{c}CL=41.72397-0.31740\times \mathit{age}+12.77424\times \mathit{age}.c^2\\ +9.73373\times \mathit{age}.c^3-0.46823\times \ln \left(\mathit{age}\right)\end{array}$$
• Equation 2. Predicted mean clearance (CL; ml/min per 1.73 m²) for neonates ≥6 days of age:

  $$\begin{array}{c}CL=41.72397-0.31740\times \mathit{age}+12.77424\times \mathit{age}.c^2\\ +9.73373\times \mathit{age}.c^3-0.46823\times \ln \left(\mathit{age}\right)\\ +(-26.03480-10.06836\times \mathit{age}.c^2-9.98987\\ \times \mathit{age}.c^3+13.64182\times \text{ln}(\mathit{age})\end{array}$$
• Equation 3. Centered age (age.c) in days to be used in Equations 1 and 2:

  $$\mathit{Age}.c=(\mathit{age}-14)/7$$

Optimal Coefficient Selection for Updated Schwartz Formula

On the basis of least-square analysis, eGFR in our population was best described by using a k of 0.308, which was rounded to 0.31, resulting in a very close approximation of mGFR. The coefficient of 0.45, as proposed by Schwartz et al.,³ significantly overestimated GFR, whereas the coefficient of Zhang et al., 0.277,⁴ underestimated GFR (Figure 3).

Figure 3.

Estimation of GFR by using different coefficients (k) in the Schwartz equation (eGFR [ml/min per 1.73 m²]=[k×height (cm)]/serum creatinine [mg/dl]). mGFR (green line) represents our data, normative serum creatinine levels are indicated in orange, and eGFR was calculated on the basis of the Schwartz equation using three different coefficients. These estimations of GFR are indicated by the dashed lines.

These findings were validated in term-born neonates with critical illness (Table 4). Using a k of 0.31 demonstrated the highest accuracy (74.4%) and lowest bias (0.0 [interquartile range (IQR), 6.4–5.2] ml/min per 1.73 m²) when compared with the Schwartz and Zhang coefficient. Applying the Schwartz coefficient led to overestimation of GFR, with an accuracy of 44.2% and a bias of 9.9 (IQR, 1.9–20.0) ml/min per 1.73 m², whereas, for the Zhang coefficient, mGFR was underestimated (accuracy, 69.8%; bias −2.9 [IQR, −8.3 to 1.3] ml/min per 1.73 m²).

Table 4.

Agreement between methods

Method	GFR (ml/min per 1.73 m2), Median (IQR)	Difference between eGFR and mGFR (ml/min per 1.73 m2), Median (IQR)	P Valuea	Accuracy30 (%)	Accuracy10 (%)
Iohexol-based mGFR	29.2 (22.3–35.0)	—	—	—	—
eGFR with k=0.31	26.3 (18.9–36.5)	0.0 (−6.4 to 5.2)	0.85	74.4	32.6
eGFR with k=0.45 (Schwartz et al.3)	37.3 (26.9–51.9)	9.9 (1.9–20.0)	0.00	44.2	20.9
eGFR with k=0.277 (Zhang et al.4)	23.5 (16.9–32.7)	−2.9 (−8.3 to 1.3)	0.01	69.8	32.6

Median GFR and median bias with corresponding IQR are displayed in ml/min per 1.73 m². Accuracy30, percentage of eGFR values within 30% of mGFR values; Accuracy10, percentage of eGFR values within 10% of mGFR values.

^aComparison of mGFR and eGFR using the Wilcoxon signed-rank test.

Discussion

Summary and Strength of Evidence

This IPDMA provides robust reference values for GFR in term-born neonates and delineates the development of GFR within the first month of life. Evidence from previous trials suggests that mean values in term-born neonates may be as low as 2–4 ml/min per 1.73 m²,¹⁹ and as high as 117 ml/min per 1.73 m²,⁴¹ and findings on the exact maturational pattern were inconclusive. The results from our IPDMA provide a sound and novel insight, i.e., there is a sharp increase in average GFR in the first 5 days (even doubling from 19.6 to 40.6 ml/min per 1.73 m²), followed by a more gradual increase (to 59.4 ml/min per 1.73 m² at day 28). Also, we propose equations to predict mGFR on the basis of postnatal age and offer a new coefficient of 0.31 to calculate eGFR on the basis of height and serum creatinine levels for use in daily clinical care.

The clear results from our IPDMA are different from the published conclusions about GFR in term-born neonates, which previously suggest a longer phase of rapid increase in GFR until 7 or 14 days of age,^19,20 or an absence of any rapid rise.^17,18,42,43 Those conclusions were formulated on the basis of findings in small cohorts. Yet, they are frequently referenced and rereferenced in review articles—thereby enhancing the influence of these conclusions. The largest individual study reporting mGFR in neonates included 82 neonates, but the study did not aim to delineate the maturational pattern of GFR and only reported mGFR values as a byproduct.⁴⁴ Only few studies have focused on the maturational pattern of GFR, and those had small sample sizes between 10 and 28 neonates.

Conducting a large, prospective trial in healthy newborns to conclusively delineate the maturation of GFR will not be possible in the current ethical framework, which puts strict boundaries on nontherapeutic research in healthy minors. This underscores the importance of summarizing results of past findings. To the best of our knowledge, summarizing and analyzing all available mGFR data from healthy, term-born neonates, going back to data published as early as 1948, has never been performed before.

Our data showing GFR values at 4 weeks postnatally (59.4 ml/min per 1.73 m²) are in line with the reference values (61.7 [SD, 14.3] ml/min per 1.73m²) at the fourth week of life reported by Piepsz and colleagues,¹⁵ who studied children from 0.1 year of age onwards. Interestingly, normative mGFR data, as measured by CrCL in preterm and very preterm neonates, show a near linear relation between postnatal age and GFR when corrected for BSA (ml/min per 1.73 m²).¹⁴ This suggests a different pattern of GFR maturation in term- versus preterm-born neonates.¹⁴

We propose the following formula to estimate GFR on the basis of enzymatic serum creatinine levels and height in healthy, term-born neonates: eGFR (ml/min per 1.73 m²)=(0.31×height [cm])/serum creatinine (mg/dl). Our k-value is different from proposed k-values in older children with CKD⁴⁵ and in children with different states of hydration.⁴⁶ This highlights the importance of k-value determination in different subpopulations.

Our demonstrated accuracy of 74% is in line with previously reported accuracy for eGFR in older children.³³ In 349 children with CKD, aged 1–16 years of age, accuracy was 79% when using the same equation, but with a coefficient of 0.413. Accuracy improved to 88% when cystatin C, BUN, and sex were also included in the eGFR equation. Whether accuracy in neonates would also increase when incorporating these parameters remains to be investigated. Although there is a strong and validated correlation between eGFR and mGFR in all neonates, including the youngest ones, rapid clearance of maternal creatinine occurs, preventing stable conditions. Because the Schwartz equation was specifically developed for stable conditions, uncertainty with regards to eGFR determination remains in the youngest neonates.

Strengths and Limitations

This meta-analysis provides a comprehensive and up-to-date summary of the development of GFR in the first month of life, on the basis of a large number of neonates. Many studies have reported mGFR values in the neonatal period; however, none of these studies summarized data from previous studies and reliable reference values were thus still missing.

We were unable to retrieve all IPD from the relevant papers, which may have affected the accuracy of our results. However, the results from the sensitivity analysis, based on all data, including the aggregated data, were similar to the results from IPD alone. Therefore, we believe our results are accurate. Furthermore, not all studies reported GFR using milliliters per minute per 1.73 m². For studies reporting GFR in milliliters per minute and only reporting weight, we derived height from the CDC charts for growth and calculated BSA accordingly. This approach may have resulted in less accurate determination of GFR. However, this applied to 28 neonates only. Also, the difference in height between two adjacent percentiles (for example p75 versus p90) is small and leads to negligible differences in BSA-corrected GFR. For instance, inulin clearance was reported to be 2.23 ml/min for a 1-day old male neonate weighing 2495 g. This weight is approximately similar to p5 (2527 g) and, therefore, the p5 for height (45.6 cm) was imputed, leading to a GFR of 21.4 ml/min per 1.73 m². When the adjacent percentile (p10) was used to impute height (46.6 cm) instead, this would have led to a GFR of 21.2 ml/min per 1.73 m². This is a difference of only 0.2 ml/min per 1.73 m², which we believe is negligible and within our margin of accuracy.

We did not observe a statistically significant difference in GFR when measured using CrCL as opposed to exogenous markers, although all studies reporting CrCL used creatinine determination according to Jaffe or compensated Jaffe. Due to the contribution of noncreatinine chromogens in this assay, overestimation of creatinine levels occurs and, consequently, lower GFR levels are reported.⁴⁷ Visual inspection of our results suggested overestimation of mGFR by CrCL in neonates >7 days of age, which could represent the maturation of tubular creatinine secretion, as suggested by previous studies. In (preterm) neonates, underestimation of inulin clearance by CrCL was observed.⁴⁸ The observed difference between CrCL and mGFR was relatively small (25%) and authors attributed this underestimation to the fact that creatinine was measured by resin adsorption and overestimation of creatinine levels occurs. Still, this finding is also in accordance with a rodent study, in which the relationship between CrCL and mGFR based on inulin was investigated.⁴⁹ An underestimation in neonatal rabbits and an overestimation in adult rabbits was observed when GFR was determined by CrCL. We believe our results also shed light on the maturation of tubular secretion and confirm that overestimation of GFR by CrCL starts at around 7 days of age. But, to draw firm conclusions about the maturation of tubular secretion, studies with a larger sample size are required to confirm this pattern in human neonates.

Because overestimation of GFR by CrCL has been reported in adults, it is remarkable that we did not observe a statistically significant difference between GFR, as measured by exogenous markers, and CrCL. A potential explanation may be found in the limited tubular secretion of creatinine in the first days of life, as discussed above. This may be comparable to CrCL while blocking the tubular handling of creatinine with cimetidine, which has been shown to result in a much better agreement with mGFR.⁵⁰ In addition, because differences are small, our statistical calculations might have lacked power to detect true differences, although our cohort was large.

We linked normative serum creatinine values to our mGFR data because individual creatinine levels have not been reported in studies reporting mGFR. Therefore, our updated coefficient of 0.31 was determined by using a least-square analysis. Preferably, when re-examining eGFR equations, analysis takes place at the participant level by model development and validation in a large cohort.⁴⁵ Due to missing creatinine values in healthy neonates, this was not possible. Also, our cohort of 43 neonates with critical illness was too small for such an approach. Therefore, our proposed GFR equation in term-born neonates needs to be confirmed in a large cohort of neonates by comparing mGFR values with eGFR using the 0.31 coefficient.

Last, precision and accuracy of eGFR equations can be optimized by including serum cystatin C and BUN levels, next to creatinine, in the equation, as demonstrated in a large cohort of children with CKD of older age.³³ By doing so, the unexplained variability in eGFR decreased and 88% of all eGFR values were within 30% of mGFR values. The added value of cystatin C and BUN in neonatal eGFR equations, however, remains to be investigated.

In conclusion, our mGFR reference values and updated coefficient for the Schwartz formula for term-born neonates will help to delineate normal kidney function from pathologic patterns and to identify neonates with altered GFR who possibly need additional investigation and monitoring to adjust drug dosing and fluid management. Although renal drug clearance is the sum of glomerular filtration and tubular handling, quantifying tubular handling remains difficult and dosing regimens for renally cleared drugs rely solely on GFR. For many drugs, GFR-adjusted dosing has been shown to optimize drug target concentrations for drugs cleared by the kidneys.⁵¹ Because clearance in neonates shows extensive interindividual variability,⁵² having more accurate GFR estimates helps to tailor drug dosage in a vulnerable population. Also, in clinical research in which kidney function is assessed with mGFR, these reference values will aid to diagnose AKI or augmented kidney clearance. To enable widespread implementation of our proposed coefficient, further validation of our proposed eGFR equation in large cohort of neonates would be needed.

Disclosures

S.N. de Wildt reports having consultancy agreements with AM-Pharma and Khondrion; serving as a director of the Dutch Knowledge Center Pharmacotherapy for Children Foundation and Dutch Pediatric Formulary BV; and receiving research funding from IMI2 conect4children. M.F. Schreuder reports serving as chair of the European Society for Paediatric Nephrology 2021 Scientific Committee and on the editorial boards of Nephrology Dialysis Transplantation and Pediatric Nephrology; and having consultancy agreements with Travere Therapeutics. G.J. Schwartz reports receiving research funding from the National Institute of Diabetes and Digestive and Kidney Diseases. All remaining authors have nothing to disclose.

Funding

The research was funded by the Radboud Universitair Medisch Centrum.

Author Contributions

S.N. de Wildt, J. IntHout, M.F. Schreuder, and G.J. Schwartz provided supervision; S.N. de Wildt, J. IntHout, M.F. Schreuder, G.J. Schwartz, and M.J.P. van der Burgh reviewed and edited the manuscript; S.N. de Wildt, J. IntHout, and N.J.L. Smeets were responsible for visualization; S.N. de Wildt, M.F. Schreuder, and N.J.L. Smeets conceptualized the study; J. IntHout and N.J.L. Smeets were responsible for formal analysis; N.J.L. Smeets wrote the original draft and was responsible for data curation; and N.J.L. Smeets and M.J.P. van der Burgh were responsible for investigation and methodology.

Data Sharing Statement

We published our entire dataset at the Data Archiving and Networked Services, which is The Netherlands Institute for Permanent Access to Digital Research Resources. Our data can be found at https://doi.org/10.17026/dans-22f-fx6t.

Supplemental Material

This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2021101326/-/DCSupplemental.

Supplemental Figure 1. Spline basis functions for cubic splines with five knots as used in our model, and model results.

Supplemental Figure 2. Cross validation of mean GFR reference values.

Supplemental Figure 3. Predicted mGFR values by original model versus predicted values by cross-validation.

Supplemental Figure 4. Sensitivity analysis without mannitol values.

Supplemental Figure 5. Sensitivity analysis without CrCL values from neonates younger than three days of age.

Supplemental Figure 6. Sensitivity analysis without mGFR values for which height was imputed.

Supplemental Figure 7. Sensitivity analysis without potential duplicates.

Supplemental Figure 8. The development of mGFR in the first month of life in term-born neonates as measured by exogenous markers versus creatinine clearance.

Supplemental Figure 9. The development of mGFR in the first month of life in term-born neonates by health status.

Supplemental Figure 10. The development of mGFR in the first month of life in term-born neonates by sex.

Supplemental Figure 11. The development of mGFR in the first month of life in term-born neonates by year of publication.

Supplemental Figure 12. Additional evaluation of GFR maturational pattern by using aggregated data.

Supplemental Table 1. Systematic assessment of risk of bias according to ROBINS-E tool in which exposure was defined as postnatal age.

Supplemental Table 2. Reference values for mGFR in the first month of life.