Study setting, design, and population
. The CKiD study is a prospective cohort study to identify risk factors for CKD progression (Copelovitch et al. 2011
; Furth et al. 2006
). As of 2011, 586 children 1–16 years of age with CKD of various etiologies and an estimated GFR of 30–90 mL/min per 1.73 m2
by the Schwartz formula (Schwartz et al. 1987
) have been enrolled from 48 clinical sites in the United States and Canada. The protocol for this study and the informed consent procedures were included in the main protocol for the CKiD study and approved by the institutional review boards at each participating center.
Enrollment of the CKiD cohort occurred over an approximately 2-year period. The present ancillary study collected whole blood aliquots for lead analysis in study participants starting several months after the cohort began year 2 study visits, and thus a portion of the cohort is missing year 2 lead values. Of 500 children completing year 2 visits, 382 had lead levels available (collected between January 2007 and December 2009). Of 211 children completing year 4 visits, 201 had lead levels available (collected between January 2008 and December 2009).
For the present cross-sectional analysis, we included all participants with blood lead levels from years 2 and/or 4 of the study (n = 456, contributing 583 lead measurements). We excluded participants who were missing data on Hispanic ethnicity (n = 7), body mass index (BMI) (n = 21), proteinuria (n = 24), income relative to the poverty level (n = 36), and hemoglobin (n = 10), leading to a final sample size of 391 participants contributing 485 lead measurements.
Analysis of blood lead. Lead and cadmium levels in whole blood were measured by high resolution inductively coupled plasma mass spectrometry at the University of California, Santa Cruz, Environmental Toxicology Laboratory (Smith DR). Samples were analyzed on an Element XR inductively coupled plasma mass spectrometer (Thermo Scientific, West Palm Beach, FL, USA) using standardized protocols including confirmation that storage materials were not contaminated with background lead. No samples were below the analytical limit of detection (< 0.1 µg/dL). Accuracy was assessed using National Institute of Standards and Technology (NIST; Gaithersburg, MD, USA) standard reference materials (SRMs). Analyses using SRMs reflecting blood lead levels of 1.6 µg/dL and 25.3 µg/dL had percent relative standard deviations (%RSDs) of 4.6 and 5.5, respectively. We assessed reproducibility by a) analyzing replicate samples at intervals throughout the same analytic run, b) analyzing samples in triplicate in the same run, and c) analyzing replicate samples in separate runs. Percent RSD for all reproducibility determinations was < 2.5%.
. GFR was measured at years 2 and 4 of the CKiD study based on plasma disappearance curves of iohexol (Ominipague; GE Healthcare, Princeton, NJ, USA). Iohexol (5 mL) was administered intravenously and blood samples were obtained at four time points at 10, 30, 120, and 300 min after infusion based on pilot data (Schwartz et al. 2006
). Of the 485 observations used herein, 30 (6.2%) did not have successful iohexol GFRs. In these cases, GFR was estimated by a CKiD-derived GFR estimating equation (Schwartz et al. 2009
eGFR = 40.7 × (height/serum creatinine)0.64 × (30/blood urea nitrogen)0.202,
with height in meters, and serum creatinine and blood urea nitrogen in milligrams per deciliter. GFR estimated by the “bedside CKiD” equation [eGFR = 41.3(height/serum creatinine)] (Schwartz et al. 2009
) was also examined in a sensitivity analysis. Serum creatinine and blood urea nitrogen were analyzed at the CKiD central laboratory on an Advia 2400 (Siemens Diagnostics, Tarrytown, NY, USA). It has been previously shown that the Siemens Bayer Advia creatinine measurement closely agrees with the high-performance liquid chromatography method traceable to reference isotope dilution mass spectroscopy developed by the NIST (Schwartz et al. 2006
. BMI was calculated as weight in kilograms divided by height in meters squared. BMI percentiles were calculated based on the CDC’s BMI-for-age sex-specific growth charts, and participants were categorized as obese if their BMI was at the 95th percentile or higher (CDC 2012a
). The diagnoses of CKD were reviewed by the CKiD Steering Committee and categorized as either glomerular or nonglomerular. Glomerular diagnoses include chronic glomerulonephritis, congenital nephrotic syndrome, diffuse mesangial sclerosis (Denys–Drash syndrome), diabetic nephropathy, familial nephritis, focal segmental glomerulosclerosis, hemolytic uremic syndrome, Henoch–Schönlein nephritis, idiopathic crescentic glomerulonephritis, IgA nephropathy, membranoproliferative glomerulonephritis types I and II, membranous nephropathy, sickle cell nephropathy, and systemic immunologic disease including systemic lupus erythematosus. Nonglomerular diagnoses included aplastic, hypoplastic, and dysplastic kidneys, cystinosis, medullary cystic disease/juvenile nephronophthisis, obstructive uropathy, oxalosis, autosomal dominant and recessive polycystic kidney disease, pyelonephritis/interstitial nephritis, reflux nephropathy, renal infarct, syndrome of agenesis of abdominal musculature, and Wilm’s tumor. A CKD diagnosis not included by one of the above was reviewed by the steering committee and, if necessary, discussed with the clinical site to be certain that it was properly categorized as glomerular or nonglomerular. Proteinuria was categorized by calculated first morning urine protein to creatinine ratio (UPC): none, UPC ≤ 0.2; significant, UPC > 0.2 to < 2.0; and nephrotic, UPC ≥ 2.0. Poverty was defined based on participant household size and income using 2009 U.S. Federal Poverty Guidelines (U.S. Department of Health and Human Services 2009). Anemia was defined as hemoglobin level less than the 5th percentile for age and sex. For secondary analyses, an “anemia status” variable was categorized as anemic participants, not treated with an erythropoiesis stimulating agent (ESA) (for example, erythropoietin); participants without anemia and not treated with an ESA; and participants treated with an ESA, with or without anemia.
Statistical analysis. Median and interquartile ranges (25th–75th percentiles) for blood lead levels and GFR were calculated for the entire study population. p-Values were determined using the median command in Stata which performs a nonparametric K-sample test on the equality of the medians and provides a Pearson chi-square test statistic. Linear regression was used to estimate associations between blood lead levels and GFR. Non-independence between measures from the same person (n = 94 with two measurements) was accounted for using robust standard errors. As a sensitivity analysis, models were rerun using linear mixed effect models in SAS and showed similar results (data not shown). Lead exposure, the explanatory variable in the linear regression model, was modeled as an untransformed continuous variable or as a natural log (ln)–transformed continuous variable. Because inferences based on ln-transformed lead were comparable (data not shown), results are reported for lead modeled as an untransformed variable for ease of interpretation. GFR was ln-transformed because it was not normally distributed. Continuous covariates (age, BMI z-score, and urine protein:creatinine ratio in the main analysis, and ln-transformed blood cadmium and ln-transformed hemoglobin in secondary analyses) were centered at the median.
Linear regression models were fitted with increasing degrees of adjustment. First we adjusted for age (continuous), sex, race (black, white, or other), Hispanic ethnicity, BMI z
-score (continuous), and poverty (yes/no). Second, the model was further adjusted for CKD diagnosis (glomerular or nonglomerular) and urine protein to creatinine ratio (continuous). Finally, the model was further adjusted for ln-transformed blood cadmium level (continuous). The estimated percent change in GFR associated with a 1-µg/dL increase in blood lead was approximated by 100 × β, where β is the coefficient for blood lead from the linear regression model of ln-GFR. For ease of interpretation, the main result is also reported for GFR as an untransformed dependent variable (with units of milliliters per minute per 1.73 m2
). To accomplish this, the beta and intercept from the original ln-tranformed GFR model are exponentiated, and thus the estimate corresponds to the change in GFR in milliliters per minute per 1.73 m2
for an individual who is female, white, not Hispanic, not impoverished, not diagnosed with glomerular CKD, and of median age, BMI z
-score, urine protein to creatinine ratio, and ln-transformed blood cadmium level (the reference category of each variable). Hypertension (yes/no) and blood pressure variables (systolic/diastolic blood pressure z
-scores/percentiles) were also evaluated as covariates but were not included in the fully adjusted final model because they did not influence the magnitude of the association between lead and GFR (data not shown) (National High Blood Pressure Education Program Working Group on High Blood Pressure in Children and Adolescents 2004
). Analyses were also restricted to participants without missing iohexol GFR (455 measurements) with similar results (data not shown).
To evaluate possible nonlinear associations between blood lead level and ln-GFR, a linear–linear spline regression analysis in fully adjusted models was examined with the cut point (1 µg/dL) selected post hoc to maximize the differences in the slopes of the linear segments above and below the cut point.
In secondary analyses, models were stratified by the participant characteristics presented in , except for anemia. For the stratified analysis, proteinuria was defined as a urine protein to creatinine ratio > 0.2. p-Values for interaction are the Wald p-values for cross-product (interaction) terms between lead and each participant characteristic. In addition, we estimated associations stratified by anemia status, with and without hemoglobin adjustment.
Blood lead levels and GFR by participant characteristic.a
All statistical analyses were two-sided. The threshold for statistical significance for all analyses was set to 0.05. Data analyses were performed using Stata versions 11.0 and 12.0 (StataCorp, College Station, TX, USA) and SAS version 9.1 (SAS Institute Inc., Cary, NC, USA) statistical software.