Continuous age-dependent centile curves and preliminary reference intervals for renal biomarkers and electrolytes in healthy Kenyan children under five years

Abstract

Renal biomarkers change rapidly in early childhood, but pediatric reference intervals used in African laboratories are often extrapolated from non-African populations and discrete age partitions. This study derived continuous age-dependent centile curves and preliminary reference intervals for body mass index (BMI), serum creatinine, urea, and electrolytes in healthy Kenyan children aged 0 to 60 months. In this cross-sectional study, 176 children (90 males, 86 females) were systematically recruited at Kitui County Referral Hospital between June 2024 and February 2025. Children underwent clinical screening, World Health Organization–standard anthropometry, and morning fasting venous blood sampling. Creatinine and urea were measured using isotope-dilution mass spectrometry-traceable enzymatic assays, and electrolytes by direct ion-selective electrode. Centiles (2.5th, 50th, 97.5th) were estimated non-parametrically with 90% bootstrap confidence intervals, and continuous age trends modeled using generalized additive models for location, scale, and shape or quantile regression. BMI showed modest age-related variation, with median values rising from 15.8 kg/m² (2.5th–97.5th: 11.4–20.2) at 0 to 11 months to 16.5 kg/m² (10.1–23.0) at 48 to 60 months, with dispersion narrowing after age 2. Median creatinine rose steeply in infancy and more gradually thereafter, from 17.4 to 31.2 µmol/L. Urea increased gradually from a median of 2.59 to 3.63 mmol/L, with overlapping centile limits across adjacent strata. Electrolytes were tightly regulated and largely age-invariant: sodium ranged from 135 to 147 mmol/L (median ~141), potassium 3.6 to 5.7 mmol/L (median declining from 4.9 to 4.5), and chloride 100 to 111 mmol/L (median ~105–106). Sex-specific partitions were supported only in early life (BMI in infancy, creatinine in toddlers); correlations between anthropometric z-scores and biomarkers were weak (|ρ| < 0.15; P > .10). These continuous centile curves provide the first preliminary age-dependent reference estimates for renal biomarkers in Kenyan children under 5, showing rapid early renal maturation alongside stable electrolyte homeostasis. As this cross-sectional design cannot capture within-subject variation, larger multicenter and longitudinal studies are needed to confirm these trajectories, refine intervals, and develop age-dependent z-score calculators for clinical use.

1. Introduction

Pediatric diagnostic accuracy depends fundamentally on the use of robust, population‑specific reference intervals.^[1] These intervals enable clinicians to distinguish physiological variation from pathology, especially in early childhood, which is characterized by rapid physiological change and increased vulnerability to disease.^[2] Despite their critical importance, locally validated ‘intervals for renal biomarkers remain largely absent in many low‑ and middle‑income countries, including Kenya.^[3] At Kitui County Referral Hospital (KCRH), a major healthcare provider in eastern Kenya, clinical assessments frequently rely on reference values adapted from adult cohorts or non‑African pediatric populations, often sourced from reagent inserts or global reference compendia.^[4] Such generalized application overlooks key inter‑population differences in genetics, environmental exposures, dietary patterns and disease epidemiology, risking diagnostic misclassification.

Several additional factors underscore the need for region‑specific pediatric reference data rather than wholesale adoption of external norms.^[5] Kidney development is non‑linear and influenced by both prenatal and postnatal factors, with glomerular filtration rate (GFR) rising steeply in the first weeks of life owing to nephron recruitment and maturation of renal blood flow, followed by incremental gains in muscle mass and urea production through early childhood.^[6] Maternal creatinine crosses the placenta and contributes to neonatal serum concentrations for up to 2 weeks postpartum,^[6] meaning that birth cohort studies must account for this passive transfer. Besides, environmental conditions in sub‑Saharan Africa, including high ambient temperatures, recurrent gastrointestinal infections, herbal medication use and dietary patterns low in animal protein, may alter renal biomarker distributions relative to North American or European cohorts.^[4,6] Altitude also influences oxygen tension and erythropoietin levels, indirectly affecting creatinine generation and hydration status.^[7] Moreover, nutritional status and growth faltering are prevalent in many low- and middle-income country contexts and can modulate biomarkers through changes in lean body mass and nitrogen metabolism. The World Health Organization (WHO) reports that roughly a quarter of Kenyan children under 5 are stunted, with notable regional variation.^[5] Early malnutrition and low birth weight reduce nephron endowment and predispose to kidney disease later in life, underscoring the need to consider growth indicators when interpreting renal biomarkers.^[8]

Prior reference interval initiatives such as the Canadian Laboratory Initiative on Pediatric Reference Intervals (CALIPER) and the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) have produced comprehensive datasets from largely Caucasian and Asian children, employing rigorous pre‑analytical control and robust statistical methods.^[9,10] However, comparability across assays, populations and laboratories remains problematic due to differences in analytical methods (e.g. Jaffe vs enzymatic creatinine), lot‑to‑lot variation and physiologic diversity. Few studies have focused on African cohorts, and those that do often have limited sample sizes or exclude infants. Gitaka et al established hematological and biochemical reference values for Kenyan children aged 4 weeks to 17 months,^[3] but sample sizes per age group were modest and did not provide continuous centile curves. Suwanrungroj et al recently reported age‑ and gender‑specific creatinine ranges for Thai children using a cross‑sectional design,^[11] highlighting the feasibility of such studies but underscoring the scarcity of data from sub‑Saharan Africa. Building on this work, we sought to establish continuous-age centile curves and preliminary reference intervals for body mass index (BMI), creatinine, urea, and key electrolytes in apparently healthy Kenyan children aged 0 to 60 months attending KCRH, while adhering to Clinical and Laboratory Standards Institute (CLSI) EP28‑A3c recommendations^[12] and acknowledging the challenges inherent in de novo reference interval establishment. In alignment with international best practices articulated by the IFCC, the establishment of localized reference data will not only enhance diagnostic accuracy but also strengthen laboratory standardization, support workforce training, and facilitate Kenya’s integration into multinational research initiatives aimed at improving pediatric kidney health in sub-Saharan Africa and advancing equitable healthcare delivery.

2. Materials and Methods

2.1. Study area

This study was conducted at KCRH, the principal tertiary care facility in Kitui County, Eastern Kenya. KCRH serves as the referral center for over 40 primary and secondary health facilities and is located approximately 160 km east of Nairobi (1°22′30.29″ S, 37°59′42.77″ E). The hospital provides preventive, diagnostic, curative and rehabilitative services and records between 6000 and 8000 outpatient consultations and approximately 1200 inpatient admissions monthly. Maternal and child health services comprise a substantial proportion of their caseload. Its clinical chemistry, hematology and microbiology laboratories are well‑equipped and participate in both internal and external quality control (QC) schemes. Outreach services are delivered via mobile health teams targeting underserved rural communities. All laboratory investigations were performed in the Clinical Chemistry Laboratory at KCRH.

2.2. Study design

We employed a descriptive cross‑sectional study design, consistent with best practices for establishing reference intervals in pediatric populations^[13] and adhered to the Standards for Reporting of Diagnostic Accuracy Studies guidelines. Data was collected at a single time point from stratified cohorts of apparently healthy children aged below 5 years. This design enabled efficient characterization of renal biomarker distributions across age without requiring longitudinal follow‑up.

2.3. Seasonality assessment

Kitui County experiences a bimodal rainfall pattern with “long rains” (March–May) and “short rains” (October–December), interspersed with prolonged dry seasons. To evaluate potential seasonal variation in renal biomarkers, we prospectively recorded the month of sample collection for each participant. We subsequently classified samples into long‑rain, short‑rain and dry‑season groups and compared median creatinine and urea concentrations across these groups using non‑parametric Kruskal–Wallis tests. No statistically significant differences were observed (P > .10 for all comparisons), suggesting that hydration and dietary changes associated with season had minimal impact on these biomarkers in our cohort.

2.4. Study population and eligibility criteria

The target population comprised children aged below 5 years accessing health services at KCRH or its outreach programs during the study period. Children were eligible if they were accompanied by a parent or legal guardian who provided written informed consent and if they were judged clinically healthy after evaluation by a medical officer. We excluded children with symptoms such as fever, diarrhea, respiratory infection, urinary abnormalities or known chronic medical conditions, including suspected or diagnosed with renal disease. To reduce inclusion of subclinical illness or dehydration, we performed a brief clinical examination, measured temperature and hydration status, obtained a urine dipstick and C‑reactive protein result, and reviewed perinatal history. Neonates with low birth weight (<2.5 kg), prematurity (<37 weeks), maternal hypertension or diabetes, or perinatal complications were excluded. Children on medications known to affect renal function were also excluded.

2.5. Sample size determination

The required sample size was calculated using Cochran’s formula for estimating a mean, as adapted by Charan and Biswas.^[14] For continuous variables, the number of observations (n) needed per stratum is given by:

$$\begin{array}{c}{\displaystyle }n=\frac{Z^2{\sigma }^2}{d^2}\end{array}$$ (1)

where Z is the standard normal deviate corresponding to the desired confidence level, σ is the estimated standard deviation (SD) of the analyte and d is the acceptable margin of error. We assumed a SD (σ) of 0.10 mg/dL for serum creatinine,^[15] a confidence level of 95 % (Z = 1.96) and a margin of error (d) of 0.05 mg/dL. Substituting the assumed values gives

$$\begin{array}{c}{\displaystyle }n=\frac{{1.96}^2{0.10}^2}{{0.05}^2}\approx 15.4\approx 16\mathrm{p}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{p}\mathrm{a}\mathrm{n}\mathrm{t}\mathrm{s}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{e}\text{-}\mathrm{b}\mathrm{y}\text{-}\mathrm{s}\mathrm{e}\mathrm{x}\mathrm{s}\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{u}\mathrm{m}.\end{array}$$

With 5 age categories and 2 sexes we aimed for 160 participants and added 10% to account for non‑responses, yielding a target of 176 children. We recognize that this sample size is below the ~120 per partition recommended by CLSI for establishing new reference intervals; therefore, our intervals are considered preliminary.

2.6. Sampling procedure

We used systematic sampling at point of care to mitigate selection bias: every third eligible child at each clinic or outreach site was approached. Written informed consent was obtained before any procedures. A structured questionnaire captured sociodemographic factors, recent illness and medication use, perinatal history, and nutrition. Anthropometry (light clothing, no shoes) was performed using calibrated equipment (length/height to 0.1 cm; weight to 0.1 kg). Nutritional status was characterized using WHO Anthro (version 1.0.3) to derive weight-for-age, length/height-for-age, and BMI-for-age z-scores (weight-for-age z-score, height-for-age z-score [HAZ], BAZ). Stunting/wasting/overweight were defined as z-scores <−2 or >+2 (WHO). Children with weight-for-age z-score, HAZ, or BAZ outside ± 2 were excluded to minimize malnutrition confounding. In the analytic sample, median BAZ was − 0.21 (interquartile range [IQR] − 0.71 to + 0.35) and median HAZ was − 0.37 (IQR − 0.83 to + 0.10); correlations between anthropometric z-scores and renal biomarkers were weak (|ρ|<0.15).

2.7. Blood collection and sample handling

Venous blood was collected 08:00 to 10:30 hours to limit diurnal variation. Children were seated or supine ≥ 10 minutes; caregivers ensured a ≥ 3 hours fasting interval for toddlers/preschoolers. Licensed phlebotomists drew 2 to 3 mL using single-use vacutainer systems. Tourniquet time was kept < 1 minutes; the site was disinfected with 70 % isopropyl alcohol and air-dried. Blood was collected into plain (red-top) tubes and allowed to clot 30 minutes (22–25°C) before centrifugation (3 000 × g, 5 minutes). Serum was inspected for hemolysis (analyzer hemolysis index threshold < 20), aliquoted into coded cryovials, stored at 2 to 8 °C, transported on ice, and analyzed within 6 hours. We used serum for electrolytes because the analyzer employed a direct ion-selective electrode (ISE) validated for serum; potassium was interpreted with caution given potential cellular release during clotting.

2.8. Laboratory analysis

All assays were performed in the KCRH Clinical Chemistry Laboratory. Serum creatinine and urea were quantified on a Biobase BK200 analyzer using manufacturer‑supplied reagents traceable to the isotope-dilution mass spectrometry (IDMS) reference method. Calibrators (Lot 2208A) were traceable to National Institute of Standards and Technology Standard Reference Material (SRM) 967, and 2 levels of commercial control sera were analyzed at the start and end of each batch. The creatinine method uses an enzymatic creatininase–creatinase–sarcosine oxidase sequence generating hydrogen peroxide measured photometrically. Urea was measured using a coupled urease–glutamate dehydrogenase method.^[16] Electrolytes (Na⁺, K⁺, Cl⁻) were analyzed using a direct ISE method^[17] on an Erma EL‑120 analyzer with daily calibration. The laboratory participates in external quality assessment (EQA) programs administered by the Kenya National Public Health Laboratory and the African Society for Laboratory Medicine; monthly reports during enrollment were within acceptable limits for all analytes.

Detailed instrument and reagent information were documented to ensure analytical traceability. Serum creatinine and urea reagents (Shenzhen Mindray Bio‑Medical Electronics Co., Ltd.) were IDMS‑calibrated and used calibrator lot 2208A with reagent lots 2210C and 2210D. Two commercial control sera (Bio‑Rad Liquichek®, level 1 and level 2) with assigned values traceable to National Institute of Standards and Technology Standard Reference Material 967 were run at 3 positions within each analytical batch (beginning, middle and end). Acceptable QC ranges were defined as the manufacturer’s assigned mean ± 2 SD; runs violating these limits prompted recalibration and repeat analysis. The laboratory participates in EQA schemes from the Kenya National Public Health Laboratory and the African Society for Laboratory Medicine, receiving monthly proficiency samples. During the enrollment period no Lot changes or instrument maintenance events affected the creatinine or urea assays. The ISE electrodes for electrolytes were replaced according to the manufacturer’s schedule, and calibration verification was performed weekly using assigned reference materials.

2.9. Quality assurance

Quality management followed CLSI EP28‑A3c^[12] and GP41 guidelines.^[18] Reagents were inspected for integrity and stored at 2 to 8°C; lot‑to‑lot comparisons were performed before use. Two levels of QC (target coefficient of variation ≤ 5%) were analyzed at the beginning, middle and end of each analytical run. Results exceeding ± 2 SD from assigned values triggered immediate investigation, recalibration and repeat measurement. During the study the laboratory’s coefficient of variation was ≈ 3.2 % for creatinine, 4.1 % for urea, 1.5 % for sodium, 2.0 % for potassium and 1.8 % for chloride. EQA results were within ± 1 SD of assigned values. To minimize pre‑analytical variability, sample separation was completed within 1 hour of collection and only non‑hemolyzed samples were analyzed.

2.10. Data management and analysis

Data was anonymized and stored in a password-protected database. Analyses were performed on R 4.4.0 with the generalized additive models for location, scale and shape (GAMLSS) and quantreg packages. Following standard recommendations,^[12] non‑parametric methods were used because analyte distributions were non‑Gaussian and sample sizes were modest. Harris–Boyd and Lahti partition tests determined whether age or sex partitions were necessary; where partitioning was not justified, data were pooled. Continuous age dependency was modeled using the GAMLSS with a Box–Cox t distribution using penalized B-spline smoothers. For analytes not meeting GAMLSS assumptions, quantile regression with natural cubic splines (3–5 knots at age percentiles) was used. Model choice (Box–Cox-t, Box-Cox Cole and Green (distribution) [Box‑Cox Cole and Green], skew-t) was based on the Akaike Information Criterion (AIC) and residual diagnostics (including worm plots). The percentile-plotting position index was determined according to equation 2 (Eq. 2), while percentile via linear interpolation between order statistics was determined using equation 2b (Eq. 3).

$$\begin{array}{c}{\displaystyle }{r}_p=(N+1)p\end{array}$$ (2)

$$\begin{array}{c}{\displaystyle }{x}_p=X_k+(r_p-k)[X_{k+1}-X_k],k=\left[r_p\right]\end{array}$$ (3)

Bootstrap resampling (1000 iterations) was performed according to equation 4 (Eq. 4) to obtain the 90 % CIs for percentile estimates at α_0.10.

$$\begin{array}{c}{\displaystyle }C{I}_{90\%}=[x_p*\left(\frac{\alpha }{2}\right),x_p*(1-\frac{\alpha }{2})],\alpha =0.10\end{array}$$ (4)

Prior to modeling, data distributions were explored using histograms and quantile–quantile plots. Potential outliers were identified using Tukey’s fences and clinically reviewed before exclusion. For a given variable, the lower and upper fences were computed as shown in equations 5 and 6.

$$\begin{array}{c}{\displaystyle }Lowerfence=Q_1-1.5(Q_3-Q_1)\end{array}$$ (5)

$$\begin{array}{c}{\displaystyle }Upperfence=Q_3+1.5(Q_3-Q_1)\end{array}$$ (6)

where Q1 and Q3 denote the first and third quartiles and (Q3 − Q1) is the IQR. Values falling outside these fences were flagged as potential outliers and evaluated for plausibility.

Age was treated as a continuous covariate, and the Z‑scores were calculated using the formula as shown in equation 5 (Eq. 7).

$$\begin{array}{c}{\displaystyle }Z\left(a\right)=\frac{X-\overline{x}\left(a\right)}{\sigma \left(a\right)}\end{array}$$ (7)

where X is the observed value, $\overline{x}$ is the age‑specific median (50th percentile), and σ(a) is the age-specific scale (from GAMLSS where available). When a model-based σ(a) was unavailable, we estimated it from the 2.5th/97.5th centiles according to equation 6 (Eq. 8):

$$\begin{array}{c}{\displaystyle }\sigma \left(a\right)\approx \frac{P_{97.5}\left(a\right)-P_{2.5}\left(a\right)}{2\times 1.96}\end{array}$$ (8)

To avoid inflating type‑I error, we did not perform multiple hypothesis tests; instead, results are presented as centiles with CIs and partition decisions.

2.11. Ethics approval and consent to participate

The study received ethical approval from the Jomo Kenyatta University of Agriculture and Technology Institutional Ethics Review Board (Ref: JKU/2/4/896B), with additional clearance from the National Commission for Science, Technology and Innovation (License No: NACOSTI/P/25/4173660). Administrative permissions were obtained from KCRH’s Ethics and Research Committee. Besides, written informed consent was obtained from all parents or guardians. Confidentiality was ensured by assigning anonymized identifiers and restricting data access. All procedures conformed to ethical principles outlined in the Declaration of Helsinki, national research ethics guidelines, and CLSI GP41 guidelines for pediatric phlebotomy and ethical research involving human participants.^[18]

3. Results

3.1. Participant characteristics

We analyzed 176 children (90 males, 86 females) across 5 prespecified age strata (0–11, 12–23, 24–35, 36–47, and 48–60 months; per-stratum n = 34, 33, 39, 39, 31) as shown in Table 1. Recruitment spanned June 2024 to February 2025 and covered both rainy and dry seasons. Monthly medians did not differ (Kruskal–Wallis P > .10), indicating no detectable seasonality. WHO growth screening confirmed a healthy analytic cohort (BAZ/HAZ within − 2 to + 2; median BAZ − 0.21; median HAZ − 0.37).

Table 1.

Distribution of participants by age group and sex (n = 176).

Age (months)	Male	Female	Total
0–11	18	16	34
12–23	15	18	33
24–35	20	19	39
36–47	22	17	39
48–60	15	16	31
Total	90	86	176

3.2. Centile estimation and partitioning strategy

Analyses followed CLSI EP28-A3c. Given non-Gaussian shapes and modest per-cell n, we estimated the 2.5th, 50th (median), and 97.5th percentiles using nonparametric order statistics and derived 90% CIs for medians via a 1000-replicate percentile bootstrap. Harris–Boyd and Lahti tests (Supplementary File-1, Supplemental Digital Content 1) were used for sex-specific partitions only in early infancy (0–6 months) for BMI and creatinine; elsewhere, strata were pooled. Continuous-age models (GAMLSS with Box–Cox-t; quantile regression when assumptions were unmet) reproduced the centile trajectories and are visualized as median ± 90% CI in Supplementary Figure 1–6, Supplemental Digital Content 2.

3.3. BMI, kg/m²

BMI showed modest age-related variation with narrowing dispersion after 24 months, reflecting consolidation of growth trajectories. The median (P50) increased from 15.8 kg/m^-2- (±0.7, 90% CI) at 0 to 11 months to 16.5 kg/m^-2- (±1.0) at 48 to 60 months (+4.4% absolute change), while the 2.5^th to 97.5th narrowed beyond age 2 (Table 2). Sex effects were restricted to early infancy and did not warrant partitions beyond 6 months. Supplementary Figure 1, Supplemental Digital Content 2 and Supplementary File 2, Supplemental Digital Content 1 show the age-group medians with 90% CIs.

Table 2.

BMI centiles by age (CLSI-format 2.5th/50th/97.5th) with 90% CI for the median.

Age (months)	Percentile			Median 90% CI (±)
	2.5th	50th (median)	97.5th
0–11	11.4	15.8	20.2	0.7
12–23	10.5	15.9	21.4	0.8
24–35	11.0	14.6	18.1	0.5
36–47	10.7	14.7	18.7	0.6
48–60	10.1	16.5	23.0	1.0

CI = confidence interval.

3.4. Serum creatinine

Creatinine concentration rose steeply across infancy and early toddlerhood, mirroring maturation of GFR and waning of maternal influence, and then increased more slowly with lean mass accretion as shown in Table 3. The median increased from 17.4 µmol/L (2.5th–97.5th: 9.1–25.6) at 0 to 11 months to 31.2 µmol/L (22.4–40.0) at 48 to 60 months (~79% rise in the central tendency) (Table 3). The steepest slope occurred in the first year; between 24 and 60 months the trajectory flattened. Supplementary Figure 2, Supplemental Digital Content 4 and Supplementary File 2, Supplemental Digital Content 1 show medians with 90% CIs.

Table 3.

Creatinine centiles by age (µmol/L) with 90% CI for the median.

Age (months)	Percentile			Median 90% CI (±)
	2.5th	50th (median)	97.5th
0–11	9.1	17.4	25.6	1.2
12–23	12.8	20.4	28.0	1.1
24–35	14.8	21.2	27.6	1.0
36–47	16.3	24.6	33.0	1.2
48–60	22.4	31.2	40.0	1.3

CI = confidence interval.

3.5. Blood urea (mmol/L)

Urea increased gradually (median 2.59→3.63 mmol/L; +1.04 mmol/L), with overlapping reference limits across adjacent strata (Table 4). This pattern is compatible with diet diversification and urea-cycle maturation. Inter-centile widths ranged from 3.70 to 4.69 mmol/L; median 90% CI widths for the median were 0.56 to 0.70 mmol/L. Supplementary Figure 3, Supplemental Digital Content 5- and Supplementary File 2, Supplemental Digital Content 1 show numeric centiles, which are summarized in Table 4.

Table 4.

Urea centiles by age (mmol/L) with 90% CI for the median.

Age (months)	Percentile			Median 90% CI (±)
	2.5th	50th (median)	97.5th
0–11	0.41	2.59	4.78	0.33
12–23	0.97	2.87	4.77	0.28
24–35	1.07	3.4	5.73	0.35
36–47	1.55	3.7	5.84	0.32
48–60	1.29	3.63	5.98	0.35

CI = confidence interval.

3.6. Serum sodium (Na⁺, mmol/L)

Sodium concentrations were tightly regulated and essentially age-invariant in this cohort as shown in Table 5. Medians clustered around 141 mmol/L with narrow 2.5^th–97.5^th limits (approx. 135–147 mmol/L) and small 90% CI widths, supporting age-agnostic intervals (Table 5). Supplementary Figure 4, Supplemental Digital Content 6- and Supplementary File 2, Supplemental Digital Content 1 show medians with 90% CIs.

Table 5.

Sodium centiles by age (mmol/L) with 90% CI for the median.

Age (months)	Percentile			Median 90% CI (±)
	2.5th	50th (median)	97.5th
0–11	135	141.0	147	0.9
12–23	136	141.0	146	0.7
24–35	135	141.0	147	0.9
36–47	137	141.0	146	0.7
48–60	135	141.0	146	0.8

CI = confidence interval.

3.7. Serum potassium (K⁺, mmol/L)

As the results in Table 6 show, potassium exhibited a gentle decline from infancy toward preschool ages while remaining within physiological bounds. The inter-centile range remained tight (~ 3.6–5.7 mmol/L) and the 90% CI for the median was uniformly small (Table 6). Supplementary Figure 5, Supplemental Digital Content 7- presents and Supplementary File 2, Supplemental Digital Content 1 provide medians with 90% CIs.

Table 6.

Potassium centiles by age (mmol/L) with 90% CI for the median.

Age (months)	Percentile			Median 90% CI (±)
	2.5th	50th (median)	97.5th
0–11	4.2	4.9	5.7	0.1
12–23	4.0	4.8	5.6	0.1
24–35	3.9	4.6	5.4	0.1
36–47	3.6	4.5	5.5	0.1
48–60	3.6	4.5	5.4	0.1

CI = confidence interval.

3.8. Serum chloride (Cl⁻, mmol/L)

Chloride remained stable across age strata with medians near 105 to 106 mmol/L and narrow 90% CIs throughout, supporting pooled intervals across ages (Table 7). Supplementary Figure 6, Supplemental Digital Content 3 and Supplementary File 2, Supplemental Digital Content 1 present the medians with 90% CIs.

Table 7.

Chloride centiles by age (mmol/L) with 90% CI for the median.

Age (months)	Percentile			Median 90% CI (±)
	2.5th	50th (median)	97.5th
0–11	101	105.7	110	0.7
12–23	100	105.6	111	0.8
24–35	102	106.2	111	0.7
36–47	101	105.4	110	0.6
48–60	100	105.3	111	0.9

CI = confidence interval.

3.9. Sex-based comparisons

To evaluate sex differences across analytes, we summarized means ± SD by sex, applied Welch two-sample tests, and report P values together with Hedges g (Table 8). Age-dependent centile trajectories for each analyte are provided in the Supplementary Figures (SF-S2–Supplementary Figure 6, Supplemental Digital Content 3), which correspond to the same age strata. Notably, consistent with partition tests, significant and clinically meaningful differences were confined to early life (BMI in 0–11 months; creatinine in 12–23 months), and decisions on reference interval partitioning followed the Harris–Boyd/Lahti criteria (Supplementary File 1, Supplemental Digital Content 1); multiple testing correction was not applied because hypothesis testing was not central to interval establishment.

Table 8.

Female vs. male comparisons (means ± SD, Welch P, Hedges g).

Parameter	Age (Months)	Female ($\overline{\mathit{x}}$ ± SD)	Male ($\overline{\mathit{x}}$ ± SD)	P- value	Hedges g
BMI (kg/m2)	0–11	14.15 ± 2.07	17.46 ± 2.43	0.0001	1.43
	12–23	16.25 ± 3.50	15.49 ± 1.97	0.4274	-0.26
	24–35	14.76 ± 1.60	14.34 ± 2.16	0.5182	-0.21
	36–47	14.77 ± 1.56	14.67 ± 2.52	0.8925	-0.05
	48–60	17.34 ± 3.56	15.71 ± 3.03	0.1562	-0.48
Creatinine (µmol/L)	0–11	17.94 ± 3.82	16.81 ± 4.58	0.4327	-0.26
	12–23	18.66 ± 3.09	22.23 ± 4.38	0.0082	0.92
	24–35	21.51 ± 2.82	20.69 ± 3.84	0.4798	-0.24
	36–47	24.36 ± 4.85	24.93 ± 3.66	0.6992	0.13
	48–60	30.61 ± 5.50	31.72 ± 3.48	0.4818	0.24
Urea (mmol/L)	0–11	2.73 ± 1.28	2.46 ± 0.94	0.4805	-0.24
	12–23	2.93 ± 0.92	2.80 ± 0.99	0.6931	-0.13
	24–35	3.53 ± 1.29	3.33 ± 1.12	0.6305	-0.16
	36–47	3.48 ± 0.86	3.91 ± 1.33	0.2561	0.37
	48–60	3.45 ± 0.90	3.82 ± 1.49	0.3782	0.29
Sodium (mmol/L)	0–11	141.70 ± 2.76	140.20 ± 3.68	0.1884	-0.45
	12–23	141.04 ± 2.57	141.10 ± 2.65	0.9042	0.02
	24–35	140.70 ± 3.18	141.80 ± 2.93	0.2864	0.35
	36–47	141.60 ± 1.79	140.70 ± 2.80	0.2876	-0.37
	48–60	141.10 ± 2.63	140.20 ± 2.66	0.3136	-0.33
Potassium (mmol/L)	0–11	4.87 ± 0.34	5.03 ± 0.38	0.1774	0.43
	12–23	4.84 ± 0.45	4.67 ± 0.36	0.2160	-0.41
	24–35	4.57 ± 0.34	4.73 ± 0.49	0.2789	0.37
	36–47	4.49 ± 0.48	4.60 ± 0.52	0.5478	0.21
	48–60	4.43 ± 0.43	4.57 ± 0.45	0.3583	0.31
Chloride (mmol/L)	0–11	106.20 ± 2.52	105.20 ± 2.39	0.2210	-0.40
	12–23	105.00 ± 2.68	106.00 ± 2.63	0.2480	0.37
	24–35	106.50 ± 2.23	105.90 ± 2.26	0.4047	-0.26
	36–47	105.90 ± 1.88	104.90 ± 2.49	0.1765	-0.44
	48–60	105.40 ± 2.99	105.20 ± 2.92	0.7989	-0.07

SD = standard deviation.

In 0 to 11 months, males had higher BMI than females (17.46 ± 2.43 vs 14.15 ± 2.07 kg/m²; P = .0001; Hedges g = 1.43), meeting the threshold for statistical significance (Table 8; SFS2). In subsequent strata, sex differences were not statistically significant: 12 to 23 months (16.25 ± 3.50 vs 15.49 ± 1.97 kg/m²; P = .4274; g = −0.26), 24 to 35 months (14.76 ± 1.60 vs 14.34 ± 2.16 kg/m²; P = .5182; g = −0.21), 36 to 47 months (14.77 ± 1.56 vs 14.67 ± 2.52 kg/m²; P = .8925; g = −0.05), and 48 to 60 months (17.34 ± 3.56 vs 15.71 ± 3.03 kg/m²; P = .1562; g = −0.48) (Table 8; Supplementary Figure 1, Supplemental Digital Content 2).

In 12 to 23 months, males had higher creatinine than females (22.23 ± 4.38 vs 18.66 ± 3.09 µmol/L; P = .0082; Hedges g = 0.92), indicating a statistically significant difference (Table 8; Supplementary Figure 2, Supplemental Digital Content 4). No statistically significant sex differences were observed in 0 to 11 months (16.81 ± 4.58 vs 17.94 ± 3.82 µmol/L; P = .4327; g = −0.26), 24 to 35 months (20.69 ± 3.84 vs 21.51 ± 2.82 µmol/L; P = .4798; g = −0.24), 36 to 47 months (24.93 ± 3.66 vs 24.36 ± 4.85 µmol/L; P = .6992; g = 0.13), or 48 to 60 months (31.72 ± 3.48 vs 30.61 ± 5.50 µmol/L; P = .4818; g = 0.24) (Table 8; Supplementary Figure 2, Supplemental Digital Content 4).

Across evaluated age strata, sex differences in urea were not statistically significant: 0 to 11 months (2.46 ± 0.94 vs 2.73 ± 1.28 mmol/L; P = .4805; g = −0.24), 12 to 23 months (2.80 ± 0.99 vs 2.93 ± 0.92 mmol/L; P = .6931; g = −0.13), 24 to 35 months (3.33 ± 1.12 vs 3.53 ± 1.29 mmol/L; P = .6305; g = −0.16), and 36 to 47 months (3.91 ± 1.33 vs 3.48 ± 0.86 mmol/L; P = .2561; g = 0.37) (Table 8; Supplementary Figure 3, Supplemental Digital Content 5).

No statistically significant sex differences were observed for sodium in any stratum: 12 to 23 months (141.10 ± 2.65 vs 141.04 ± 2.57 mmol/L; P = .9042; g = 0.02), 24 to 35 months (141.80 ± 2.93 vs 140.70 ± 3.18 mmol/L; P = .2864; g = 0.35), 36 to 47 months (140.70 ± 2.80 vs 141.60 ± 1.79 mmol/L; P = .2876; g = −0.37), and 48 to 60 months (140.20 ± 2.66 vs 141.10 ± 2.63 mmol/L; P = .3136; g = −0.33) (Table 8; Supplementary Figure 4, Supplemental Digital Content 6).

Potassium showed no statistically significant sex differences: 0 to 11 months (5.03 ± 0.38 vs 4.87 ± 0.34 mmol/L; P = .1774; g = 0.43), 12 to 23 months (4.67 ± 0.36 vs 4.84 ± 0.45 mmol/L; P = .2160; g = −0.41), 24 to 35 months (4.73 ± 0.49 vs 4.57 ± 0.34 mmol/L; P = .2789; g = 0.37), 36 to 47 months (4.60 ± 0.52 vs 4.49 ± 0.48 mmol/L; P = .5478; g = 0.21), and 48 to 60 months (4.57 ± 0.45 vs 4.43 ± 0.43 mmol/L; P = .3583; g = 0.31) (Table 8; Supplementary Figure 5, Supplemental Digital Content 7).

Chloride also showed no statistically significant sex differences across strata: 0 to 11 months (105.20 ± 2.39 vs 106.20 ± 2.52 mmol/L; P = .2210; g = −0.40), 12 to 23 months (106.00 ± 2.63 vs 105.00 ± 2.68 mmol/L; P = .2480; g = 0.37), 24 to 35 months (105.90 ± 2.26 vs 106.50 ± 2.23 mmol/L; P = .4047; g = −0.26), 36 to 47 months (104.90 ± 2.49 vs 105.90 ± 1.88 mmol/L; P = .1765; g = −0.44), and 48 to 60 months (105.20 ± 2.92 vs 105.40 ± 2.99 mmol/L; P = .7989; g = −0.07) (Table 8; Supplementary Figure 6, Supplemental Digital Content 3).

3.10. Association of anthropometry with biomarkers

Across age strata, correlations between BMI-for-age (BAZ) or height-for-age (HAZ) and renal biomarkers were weak and nonsignificant (|ρ| < 0.15; P > .10; Table 9; Supplementary File 1, Supplemental Digital Content 1). Moreover, a matrix of n per age‑by‑sex cell alongside full percentile tables at monthly increments is provided in Supplementary File 2, Supplemental Digital Content 1 and an interactive Excel calculator for z‑scores to facilitate clinical implementation and allow clinicians to input age in months and obtain estimated centiles and z‑scores is provided in Supplementary File 3, Supplemental Digital Content 1.

Table 9.

Summary of Spearman correlations (ρ) between anthropometric z-scores and biomarkers

Biomarker	ρ (BAZ)	ρ (HAZ)	P-value
Creatinine	0.12	0.08	> 0.10
Urea	0.1	0.11	> 0.10
Sodium	0.05	0.04	> 0.10
Potassium	-0.04	-0.02	> 0.10
Chloride	-0.03	-0.05	> 0.10

BAZ = BMI-for-age z-score, HAZ = height-for-age z-score.

4. Discussion

Pediatric laboratory decision-making in sub-Saharan Africa still leans heavily on reference intervals generated in high-income populations, despite clear biological and contextual differences in infant physiology, growth, diet, altitude, and disease exposure.^[19] This mismatch risks misclassification at the very ages when renal function is changing fastest and when broad “0–11 months” bins are least defensible. We undertook this study to generate locally derived, age-resolved centiles for BMI and core renal chemistry in Kenyan children under 5, using CLSI EP28-A3c–concordant methods, continuous-age modeling, and fully traceable analytics.^[12] By pairing rigorous pre-analytical control with method standardization, we aimed to produce intervals, and z-score equations, that clinicians and laboratories can implement with confidence, and that health systems can verify and scale. The result is a maturation-aware foundation tailored to our setting, designed to improve diagnostic accuracy, reduce avoidable referrals, and support safer pediatric care.

BMI in early childhood integrates adiposity and lean mass and is a practical screening tool for undernutrition and emerging obesity risk, both of which carry downstream effects on infection susceptibility, neurodevelopment, and later cardiometabolic health.^[8] In < 5-year-olds, BMI-for-age centiles and z-scores provide a maturation-aware frame for interpreting growth patterns across rapid developmental stages.^[8] The centiles show a steady increase in girls across the preschool years, consistent with linear growth and gradual fat accretion. In boys, BMI is higher in infancy, dips during toddlerhood, and then rises toward the fifth year, plausibly reflecting changes in activity, energy expenditure, and transient differences in body composition during early life.^[20,21] These trajectories align with WHO BMI-for-age reference curves and support the use of sex- and age-specific BMI references in clinical care and public health surveillance.^[22] Importantly, the absence of meaningful correlations between BMI and renal markers across age groups indicates that, in healthy children under 5, anthropometry alone does not predict renal biochemistry once age is accounted for; weak, nonsignificant associations observed in older strata likely reflect early muscle development rather than kidney impairment.^[23,24] Clinically, the provision of age-specific 2.5th/50th/97.5th percentiles allows maturation-aware interpretation of BMI together with renal indices, reducing misclassification that can occur when adult thresholds or nonlocal pediatric intervals are applied.

Serum creatinine, an endogenous product of muscle metabolism, serves as the most widely used surrogate for GFR in routine care, but its concentration is age- and muscle-mass dependent and method sensitive.^[25] Accurate, IDMS-traceable quantification is essential in infancy, when maternal transfer and rapid renal maturation can confound interpretation.^[10] The creatinine centiles increased sharply through infancy and into the second year of life before rising more gradually, mirroring well-described increases in glomerular filtration and accrual of lean mass after birth. This pattern is biologically coherent and consistent with large pediatric reference initiatives, including CALIPER/IFCC.^[2,9,26,27] A sex difference appears in the second year, with higher male centiles, plausibly reflecting earlier lean-mass accrual and androgen-mediated effects on muscle metabolism.^[11,28,29] These findings imply that for age- and, in early life, sex-aware interpretation to avoid under- or overestimating renal impairment. Absolute values in Kenyan toddlers tend to be modestly lower than those reported in some high-income cohorts, a difference that tracks with lean-mass, dietary protein exposure, and residual calibration nuances even among IDMS-traceable enzymatic assays.^[9,25] Because creatinine is method dependent, laboratories using Jaffe methods should verify transferability and consider bias adjustments before adoption.^[1,10]

Blood urea reflects the interplay of protein intake, hepatic urea cycle function, hydration status, and renal excretion, making it a useful, though less specific, adjunct to creatinine in assessing nitrogen balance and prerenal states.^[25] Age, diet, and catabolic stress can shift its distribution, warranting population-appropriate centiles. The urea centiles obtained in this study showed a gradual age-related rise, consistent with maturation of hepatic urea synthesis and increasing dietary protein intake across the first years of life. The distribution remains tight in both sexes and observed sex differences are small. Compared with some high-income pediatric cohorts, central tendencies in this Kenyan sample are modestly lower, plausibly reflecting protein intake patterns and growth context, though the shape of the trajectory is congruent with international experience.^[9–11] Clinically, elevated urea outside age-appropriate limits should prompt assessment for dehydration, catabolic states, gastrointestinal blood loss, or impaired renal clearance, while unusually low values may signal severe malnutrition or hepatic dysfunction in young children.^[30,31] As with creatinine, method standardization and quality assurance are prerequisites for safe transfer of urea limits into routine use.

As the principal extracellular cation, sodium is the dominant determinant of plasma osmolality and extracellular volume, whereby even small deviations can signal dehydration, dysnatremias from endocrine/renal causes, or iatrogenic shifts.^[32] Reliable interpretation in children depends on direct ISE measurement and tight pre-analytical control. Sodium distributions were narrow and essentially age- and sex-invariant across the 0 to 60-month window, reflecting robust homeostatic control via renal tubular handling and neurohormonal feedback loops.^[33] This stability aligns with reports using direct ISE in pediatric populations.^[17,32] The absence of material age or sex effects supports uniform pediatric limits for sodium in early childhood when measurement principles and pre-analytics are comparable.^[32] Clinically, this provides a reliable benchmark for evaluating dehydration, salt-wasting disorders, and iatrogenic shifts related to fluids, without the need for complex age-specific thresholds within this age range.

Chloride is the principal extracellular anion central to acid–base physiology via the strong-ion difference and tracks with renal and gastrointestinal handling of chloride and bicarbonate.^[33] Hypo- and hyperchloremia provide complementary information to sodium and bicarbonate in evaluating fluid balance and metabolic derangements.^[33] In this study, chloride level mirrored sodium in its stability across early childhood, with tight central tendencies and minimal dispersion in both sexes. This pattern is expected given coupled renal handling of sodium and chloride and the central role of chloride in acid–base balance. The practical implication is that a single pediatric interval can be applied across sexes and early childhood ages when direct ISE is used and pre-analytics are well controlled.^[17,32] Deviations from the centiles should prompt evaluation for metabolic acidosis/alkalosis, gastrointestinal losses, and fluid balance disturbances.

Potassium governs resting membrane potential and myocardial conduction, so pediatric thresholds must balance the higher physiological range seen in early life with the real risk of arrhythmia at extremes.^[34] Potassium showed a gentle decline from infancy into the preschool years in this study. This trajectory is physiologically plausible and reflects developmental changes in renal potassium handling, intracellular–extracellular partitioning, and dietary transition. Slightly higher values in infancy are unsurprising given immature renal function and greater intracellular potassium content during rapid growth.^[35,36] The small, non-significant male elevations observed in early life are consistent with transient androgen influences on tubular transport described in neonatal studies.^[37] Because potassium is sensitive to pre-analytical artifacts, strict control of tourniquet time, fist clenching, and time-to-separation is essential to preserve the clinical meaning of observed centiles.^[38] In practice, age-aware interpretation helps avoid overdiagnosis of hyperkalemia in infants and under-recognition of clinically important shifts in older toddlers.

Across analytes, sex differences were limited to early life where males recorded higher BMI during infancy and higher creatinine during the second year than females. Beyond roughly 2 years of age, sex contrasts were small and statistically unconvincing across all measures. Reporting effect sizes (Hedges g) alongside P values provides a calibrated sense of magnitude and guards against overinterpretation of isolated significant tests.^[39] Crucially, partition decisions were based on Harris–Boyd/Lahti criteria rather than on multiplicity-prone hypothesis testing, resulting in pooled sex intervals outside early infancy,^[40,41] an approach that we consider both biologically parsimonious and CLSI-concordant.

While the shapes of the age trajectories match those reported in North American and European cohorts, absolute creatinine and urea values in Kenyan toddlers are modestly lower, possibly due to differences in lean-mass accrual and protein intake as well as residual method effects ^[20,23]. Electrolyte stability across early childhood agrees with pediatric reports obtained by direct ISE.^[25,42,43] These patterns underscore shared physiology across populations when analytic methods and pre-analytics are aligned, while also highlighting the value of locally derived centiles to capture context-specific distributional shifts that could otherwise lead to misclassification. Clinical screening with WHO growth assessments and exclusion criteria reduced confounding from malnutrition and subclinical illness. However, we must acknowledge that modest sample sizes within some age-by-sex cells relative to the ~120 per partition often recommended for de novo establishment, may widen CIs at the tails and in the youngest ages. A more fundamental limitation is the cross-sectional design itself: because each child was sampled only once, the study cannot capture within-subject biological variation, including day-to-day fluctuation in creatinine, urea, or electrolyte concentrations, and cannot fully disentangle true age-related maturational change from cohort effects, that is, whether differences between younger and older children reflect renal physiological maturation or simply reflect that they were born into, and sampled from, different birth cohorts. Because recruitment occurred within a single calendar window (June 2024-February 2025), the observed age trajectories may also be confounded by secular trends specific to that period, for example transient shifts in food security, infection burden, or care-seeking behavior, that cannot be separated from genuine maturational change using cross-sectional data alone. These features constrain the robustness of the derived reference intervals: the reported centiles describe the distribution of the sampled population at a single point in time and should not be overinterpreted as a description of how an individual child’s own biomarker values will change as they age. Nonparametric estimation, bootstrapping, and smoothing^[44,45] mitigate sampling noise but cannot substitute for repeated-measures data. Longitudinal follow-up of the same children over time is therefore needed to establish true within-subject physiological variation, confirm that the observed age-related trends reflect genuine maturation rather than cohort or secular effects, and refine and validate these preliminary reference intervals.

The centiles and medians, together with age-dependent z-scores, are readily implementable in laboratory information systems to provide maturation-aware interpretation. Elevated creatinine in older infants and toddlers should trigger assessment for dehydration, obstructive uropathy, or parenchymal injury; unusually low creatinine in young infants warrants evaluation for malnutrition or hepatic disease. Stable sodium and chloride intervals simplify decision-making across sexes, while age-aware potassium interpretation reduces false alarms in infancy. Because creatinine and electrolytes are method sensitive, laboratories using Jaffe creatinine or indirect ISE should perform local verification and consider bias adjustments before adoption.^[20,23,27] In systems where de novo studies are not yet feasible, CLSI-style verification of established pediatric intervals (e.g., CALIPER/IFCC) against smaller, well-screened local samples is a pragmatic interim step.^[27]

Going forward, enlarged neonatal and early-infancy cohorts are needed to support biologically informed micro-partitions, alongside longitudinal follow-up to separate maturation from cross-sectional differences and to quantify within-child variability. Expanding the panel to include cystatin C and emerging acute kidney injury biomarkers would broaden clinical applicability and reduce creatinine’s dependence on muscle mass.^[15] Multi-site collaborations using harmonized pre-analytics, shared calibrators, and common data standards would accelerate regional harmonization, tighten CIs where they are currently widest, and strengthen the evidence base for pediatric laboratory medicine in sub-Saharan Africa.

5. Conclusions and recommendations

We established locally derived, age-dependent centiles (2.5th/50th/97.5th with CIs) for BMI, creatinine, urea, sodium, potassium, and chloride in Kenyan children under 5 using continuous-age modeling. Findings were physiologically coherent: creatinine rose steeply in infancy then slowed; urea increased modestly; sodium and chloride were stable across age and sex; potassium declined slightly from infancy to preschool. Sex effects were confined to early life, higher BMI in males at 0 to 11 months and higher creatinine at 12 to 23 months, supporting pooled sex intervals thereafter. Results are method-specific (IDMS-traceable enzymatic creatinine; direct ISE) with acceptable QC/EQA. These centiles should be adopted after local CLSI verification, with method-comparison and bias adjustment where Jaffe creatinine or indirect ISE are used. Laboratories should embed age-dependent z-scores in laboratory information systems and electronic health records and maintain strict pre-analytical control. Clinically, interpret creatinine by age (and early-life sex), apply unified limits for sodium and chloride, and use age-aware thresholds for potassium. Future work should enlarge neonatal cohorts for biologically informed micro-partitions, pursue longitudinal multi-site studies to establish true within-subject physiological variation and disentangle maturational from cohort and secular effects, and incorporate cystatin C and emerging acute kidney injury biomarkers.

Acknowledgments

We appreciate the respondents who took part in this study, as without them, this study would not have been possible. We also acknowledge the clinicians and laboratory technologists at the Kitui County Referral Hospital for their cooperation and technical assistance during this study; no individuals are identified by name in this Acknowledgments.

Author contributions

Conceptualization: Brian Nganda, Peter Karanja, Kennedy Muna, Gervason Moriasi.

Data curation: Brian Nganda, Gervason Moriasi.

Formal analysis: Brian Nganda, Gervason Moriasi.

Funding acquisition: Brian Nganda.

Investigation: Brian Nganda, Gervason Moriasi.

Methodology: Brian Nganda, Kennedy Muna, Gervason Moriasi.

Project administration: Brian Nganda.

Resources: Brian Nganda.

Validation: Brian Nganda, Peter Karanja, Kennedy Muna, Gervason Moriasi.

Visualization: Brian Nganda, Gervason Moriasi.

Supervision: Peter Karanja, Kennedy Muna.

Software: Gervason Moriasi.

Writing – original draft: Brian Nganda, Gervason Moriasi.

Writing – review & editing: Brian Nganda, Peter Karanja, Kennedy Muna, Gervason Moriasi.