Study population. NHANES uses a complex multistage sampling design to obtain representative samples of the noninstitutionalized U.S. population (NCHS 2010a). We used data from NHANES III (1988–1994), collected in two phases (1988–1991 and 1991–1994), and from NHANES 1999–2008, collected in five phases (1999–2000, 2001–2002, 2003–2004, 2005–2006, and 2007–2008). In NHANES III, urine cadmium was measured in all participants ≥ 6 years of age. In NHANES 1999–2008, urine cadmium was measured in a random one-third subsample of participants ≥ 6 years of age. For this analysis we included 23,904 adults ≥ 20 years of age (7,967 in 1988–1991, 8,169 in 1991–1994, 1,299 in 1999–2000, 1,560 in 2001–2002, 1,532 in 2003–2004, 1,520 in 2005–2006, and 1,857 in 2007–2008). We excluded 628 pregnant women; 109 participants with missing urine creatinine measurements; 3,179 participants with missing smoking status, pack-years, or serum cotinine; and 229 participants with missing other variables of interest. A total of 19,759 participants (6,616 in 1988–1991, 7,075 in 1991–1994, 965 in 1999–2000, 1,166 in 2001–2002, 1,209 in 2003–2004, 1,160 in 2005–2006, and 1,568 in 2007–2008) were included in our analyses. Participants included in this analysis were similar to the corresponding NHANES-phase population with respect to sociodemographic variables (data not shown).
Urine cadmium. Cadmium in spot urine samples was measured at the Environmental Health Laboratory of the Centers for Disease Control and Prevention’s (CDC) National Center for Environmental Health (Atlanta, GA, USA) for all surveys. Extensive quality control procedures were followed, including confirmation that collection and storage materials were not contaminated with background cadmium or other metals (NCHS 2010a).
Urine cadmium was measured by graphite furnace atomic absorption spectrometry (model 3030; PerkinElmer, Norwalk, CT, USA) with Zeeman background correction in NHANES 1988–1994, by inductively coupled plasma mass spectrometer (ICP-MS; ELAN, PerkinElmer) in NHANES 1999–2002, and by ICP–dynamic reaction cell (DRC)-MS (ELAN DRC, PerkinElmer) in NHANES 2003–2008 (NCHS 2010a). In NHANES 1988–1994, specimens were analyzed in duplicate, and the average of the two measurements was reported (Paschal et al. 2000
). The interassay coefficients of variation ranged from 2.8% to 13.6%, and the limit of detection (LOD) was 0.03 µg/L (NCHS 2010a), resulting in 6% of observations below the LOD. In NHANES 1999–2008, the interassay coefficients of variation for urine cadmium ranged from 1.2% to 6.7%, and the LOD was 0.06 µg/L in 1999–2004 and 0.042 µg/L in 2005–2008, resulting in 4% of observations below the LOD. In all phases, the urine reference material from the National Institute of Standards and Technology (NIST 2010
) was analyzed periodically to ensure analytical accuracy, and observed concentrations were in good agreement with published values (96–101%) (Jarrett et al. 2008
; Paschal et al. 2000
). Two levels of in-house urine pools traceable to the reference material were used for daily quality control. One of two different levels of a blind quality control material was inserted in every analytical group of samples for an additional quality control check. In NHANES III, laboratory measures were within 10% of reference means for urinary cadmium (r2
= 0.97) (Paschal et al. 2000
). Precision data for NHANES 1999–2008 are discussed in detail in Jarrett et al. (2008)
. In brief, NHANES 1999–2002 data were mathematically adjusted to eliminate a bias in the original standard-mode quadrupole ICP-MS data caused by a molybdenum-based interference. The CDC lab used a new ICP-DRC-MS method starting in the NHANES 2003–2004 cycle to eliminate the molybdenum oxide interference. Therefore, starting with the 2003–2004 survey cycle, the need for mathematical correction of the urine cadmium data was eliminated (Jarrett et al. 2008
). For observations below the LOD and for values corrected for interference from molybdenum oxide equal to 0 in 1999–2002 data (n
= 4), urine cadmium value was imputed as the LOD divided by the square root of 2 (Hornung and Reed 1990
Creatinine-corrected urine cadmium data were reported in micrograms cadmium per gram creatinine. Urine creatinine was measured by the modified kinetic Jaffé method in 1988–2006 and by an enzymatic (creatinase) method in 2007–2008 (NCHS 2010a). We corrected urine creatinine determinations before 2007 as recommended by NHANES (NCHS 2010a).
Urine cadmium determinants.
Information on age, sex, race/ethnicity, education, smoking status, cigarette pack-years, occupation, body mass index (BMI), and serum cotinine was based on questionnaires, physical examination, and laboratory methods that have been described elsewhere (Benowitz et al. 1983
; NCHS 2010a; Yassin and Martonik 2004
). We classified study participants as current smokers if they answered yes to the question “Do you smoke cigarettes now?” or had serum cotinine levels > 10 ng/mL (Benowitz et al. 1983
). Former smokers were participants who answered yes to the question “Have you smoked at least 100 cigarettes during your entire life?” but were not current smokers. Pack-years were determined using answers to the following questions: “How old were you when you first started smoking cigarettes fairly regularly?”, “About how many cigarettes do you smoke per day now?” (or, because of changes in the smoking questionnaires in NHANES 2007–2008, “Average number of cigarettes/day during the past 30 days”), “For approximately how many years have you smoked this amount?”, “About how old were you when you last smoked cigarettes (fairly regularly)?”, and “About how many cigarettes per day did you usually smoke at that time?”
Occupations and industries associated with cadmium exposure were based on the study by Yassin and Martonik (2004)
and included self-reported occupations related to transportation, metal, mining, and repairing service industries. Duration of the longest cadmium-associated occupation was determined using answers to the questions: “What kind of work were you doing (in the past 1 or 2 weeks, depending on the survey period)?”, “About how long have you worked for (employer) as a(n) (occupation) (in the past 1 or 2 weeks, depending on the survey period)?”, “Thinking of all the paid jobs or businesses you ever had, what kind of work were you doing the longest?”, and “About how long did you work at that job or business (the longest)?” For participants with the longest held job not being cadmium related, the current job was also considered as a potential source of cadmium. Information on occupation was not available in NHANES 2005–2008.
Serum cotinine was measured by an isotope-dilution high-performance liquid chromatography/atmospheric pressure chemical ionization tandem MS method (NCHS 2010a). The LOD for serum cotinine was 0.05 ng/mL in NHANES III and 0.015 ng/mL in NHANES 1999–2008. For the 4.9% and 9.0% of participants below the LOD in NHANES III and NHANES 1999–2008, respectively, cotinine concentrations were replaced by the LOD divided by the square root of 2.
Statistical analyses were performed using the survey package (Lumley 2004
) in R software (version 2.12.1; R Development Core Team 2011) to account for the complex sampling design and weights in NHANES 1988–2008 and to obtain appropriate standard errors for all estimates. Urine cadmium levels were right skewed and were log transformed for the analyses. Age-adjusted cadmium concentrations were obtained using the residuals from linear regression models of log-urine cadmium concentrations corrected for creatinine modeled by age as restricted cubic splines with 5 knots, and adding back the cadmium-weighted means of the corresponding NHANES samples. Age-adjusted creatinine-corrected urine cadmium concentrations were reported as geometric means.
Starting in 1999, urine cadmium was measured in a one-third random subsample of NHANES participants only. To increase the sample size in regression models over time, we grouped multiple phases together (1988–1994, 1999–2002, and 2003–2008). Temporal trends in urine cadmium concentrations were evaluated by estimating geometric mean ratios and 95% confidence intervals (CIs) of urine cadmium concentrations in NHANES 1999–2002 and NHANES 2003–2008 compared with NHANES 1988–1994. The geometric mean ratios were obtained by exponentiating linear combinations of β-coefficients in regression models adjusted for age, sex, and race/ethnicity with log-transformed cadmium as the dependent variable and survey phase group and cadmium determinants as interacting independent variables. Subsequently, percent reductions in geometric means over time [estimated as (1 – geometric mean ratio) × 100] and 95% CIs were calculated overall and by subgroups of cadmium determinants. We considered the following subgroups of cadmium determinants: age (< 35, 35–49, 50–65, and ≥ 65 years), sex (men and women), race/ethnicity (white, African American, Mexican American, and other), education (< high school and high school, or higher education), BMI (< 25, 25–30, and ≥ 30 kg/m2), cigarette smoking status (never, former, current), cigarette pack-years (0, 0–10, 10–20, and > 20 pack-years), serum cotinine (< 0.05, 0.05–10, 10–200, ≥ 200 ng/mL), and duration of the longest occupation associated to cadmium (0, 0–10, 10–20, > 20 years in the subset of participants with information available). We estimated p-trend values by applying the Wald test to the appropriate regression coefficients.
Toxicokinetic parameters related to cadmium absorption, accumulation in the renal cortex, and excretion, such as low iron stores or reduced kidney function, could induce variation in urine cadmium concentrations independently of variations in exposure (Amzal et al. 2009
; Ruiz et al. 2010
). Therefore, we performed a sensitivity analysis using models adjusted for serum iron, and glomerular filtration rate, in addition to age, sex, and race/ethnicity. Results were similar to those reported here (data not shown).
The relative contribution of cadmium determinants to the trend in urine cadmium concentrations was calculated as the relative change in the β-coefficient for survey phase category (grouped as 1999–2008 vs. 1988–1994) in regression models for log-transformed cadmium after sequentially introducing cadmium determinants in linear regression models. In this analysis, NHANES 1999–2002 and 2003–2008 were combined and compared with NHANES 1988–1994 for simplicity, because results in NHANES 1999–2002 and NHANES 2003–2008 separately yielded similar findings. Because occupation variables were available only until 2004, the contribution of occupation to the change in urine cadmium concentrations was evaluated only for the available period.
We also conducted a sensitivity analysis comparing estimates from models of urine cadmium concentrations (micrograms per liter) adjusted for log-transformed urine creatinine (data not shown) with those from models of creatinine-corrected urine cadmium concentrations (micrograms per gram). Results were comparable to those reported here.