When selecting the least biased exposure surrogate, for example, the concentration of a biomarker in a urine sample, information on variability must be taken into consideration. We used mixed-effects models to estimate the variability and determinants of urinary cadmium (U-Cd) excretion using spot urine samples collected at six fixed times during 2 days about 1 week apart, from 24 healthy non-smokers. The urine samples were analysed for U-Cd, the concentrations were adjusted for dilution, and the excretion rates were calculated. Between-individual variability dominated the total variability for most measures of U-Cd excretion, especially for 24 h urine and first morning samples. The U-Cd excretion showed a circadian rhythm during the day, and time point of sampling was a significant factor in the mixed-effects models, thus a standardised sampling time, such as first morning urine samples, needs to be applied. Gender, urinary flow rate, age, and urinary protein excretions were also significant determinants for U-Cd excretion. The choice of biomarker for U-Cd excretion was found to be more important in individually-based studies of exposure–response relationships than in studies of comparing Cd levels of groups. When planning a study, this variability of U-Cd in spot samples must be acknowledged.
Exposure assessment is a critical component of occupational and environmental epidemiological studies. Assessment of long-term exposure is normally based on surrogate measures of exposure, such as a limited number of air or biomarker concentrations. In environmental epidemiology, biological monitoring is theoretically advantageous, first, because it reflects differences in uptake and kinetics between individuals and second, because it accounts for all possible routes of exposure, for example, inhalation, ingestion, and dermal contact, and thus reduces the need to monitor all environmental sources separately.1 The biological half-life of a biomarker has been shown to affect the exposure variability, with slower elimination leading to decreased variability, and so information on the variability of the potential biomarkers must be taken into consideration when selecting the least biased measure and when designing a study.2 Two major components of variability are the variation within individuals (i.e., the variation between different samples from the same individual), and the variation between individuals (in terms of average levels). Variability could also be introduced in the sampling procedure and analytical method; this will be included in the within-individual variance, although the analytical variability is generally small.3
In the case of cadmium (Cd), the main route of exposure for environmentally exposed individuals is through the diet. For smokers, tobacco use is also an important route of exposure.4, 5, 6 After ingestion or inhalation, 3–5% of the Cd in diet and 10–50% of the inhaled Cd is absorbed, bound to proteins in the blood, and eventually accumulated in the body, mainly in the kidneys, with a biological half-life of 20–40 years.4, 5, 6, 7 Gastrointestinal Cd absorption is also dependent on iron status and so young women with low iron stores might have increased Cd absorption.8 Absorbed Cd is excreted in urine and faeces, and urinary Cd (U-Cd) is widely used as a biomarker to assess the long-term exposure or body burden of Cd.5 Although Cd has a long biological half-life in the human body, and the long-term level in an individual is therefore thought to be stable, there is still variability within individuals when U-Cd is measured repeatedly. This variability is induced by factors, such as natural physiological variations, the choice of sampling strategy (e.g. time of the day), and the method used to adjust the U-Cd concentrations for diuresis. In a study population, there is also variability in U-Cd excretion between individuals, induced by factors such as level of exposure and differences in absorption.
Theoretically, 24 h urine (U24) samples represent the average U-Cd excretion more reliably than spot urine samples. However, U24 sampling is laborious, and incomplete or contaminated samples might pose a problem. A common alternative is to use overnight spot urine (UON) samples (first morning), as a high correlation has been shown between Cd in U24 and UON samples.9, 10 In large epidemiological studies or in reanalysis of samples from existing biobanks, UON samples could be difficult to obtain, and spot urine samples taken at any time of the day may have to be used.
Spot urine samples must be adjusted for dilution, especially if the results are to be compared with other studies or with guideline values. This adjustment could be done by using the creatinine concentration (U-Crea) or specific gravity (SG) of the urine sample.10, 11, 12
There is only limited information available about the variability of U-Cd excretion and its impact on the best choice of study design among healthy individuals with low-level environmental Cd exposure. In this study, we aimed to investigate the short-term variability of different measures of U-Cd excretion and its implications for study design, using repeated spot urine measurements from a healthy non-smoking population with low-level Cd exposure.
MATERIALS AND METHODS
Spot urine samples were collected from 30 healthy non-smoking study participants (15 men and 15 women; no diabetes, hypertension, or kidney disease) recruited from our department and among university students as previously described.13, 14 The spot urine samples were collected during two 24 h periods about 1 week apart, and the study participants were instructed to urinate at six fixed times (first morning void, 0930, 1200, 1430, 1730, and 2200 hours). The participants had no food or drink restrictions during the study. If urination was necessary between the fixed times, the next container was used to ensure a complete 24 h sampling. The collected samples were transferred to Minisorb tubes (NUNC, Roskilde, Denmark) and kept at 4 °C until analysis of proteins (Albumin (Alb) and α-1-microglobulin (A1M)) and U-Crea within 3 days of collection. Aliquots used for determination of U-Cd were frozen (−20 °C) and analysed 5 years later. One woman only provided urine samples on one of the two days, and was therefore excluded from the study. Samples from another woman, with repeatedly very low U24 excretion (around 500 ml), were regarded as incomplete or unrepresentative, and excluded from the study. The study was approved by the Ethics Committee of the University of Gothenburg and informed consent was obtained from all study participants.
U-Cd was analysed as described elsewhere.13 Briefly, the samples were analysed by inductively coupled plasma mass spectrometry (Thermo X7, Thermo Elemental, Winsford, UK) in samples diluted 10 times with an alkaline solution and corrected for molybdenum oxide-based interference, as molybdenum oxide formed during the analysis from the molybdenum (Mo) naturally present in urine might interfere with cadmium isotopes 111Cd and 114Cd.15, 16 All U-Cd samples were prepared in duplicate; the imprecision (calculated as the coefficient of variation for duplicate preparations) was 9.5%. The limit of detection (LOD) for U-Cd, calculated as three times the standard deviation of the blank, was 0.05 μg/l; U-Cd concentrations were below the LOD in 91 of the samples (27%). Three quality control samples were used (Trace Elements Urine, Seronorm AS, Billingstad, Norway, and Urine Reference Material, Le Centre de Toxicologie du Quebec, International Comparison Program, Canada); the results versus recommended values (± standard deviation) were 0.26±0.03 μg/l (N=12) versus 0.26–0.36 μg/l, 0.93±0.04 μg/l (N=12) versus 1.01±0.09 μg/l, and 4.9±0.12 μg/l (N=12) versus 5.1±0.26 μg/l, respectively.
Analyses of urinary albumin (U-Alb) were performed by an automated nephelometric immunochemical method using reagents and calibrator from Beckman Coulter (Fullerton, CA, USA). Internal reference samples were used in each analytical run, showing satisfactory results. The LOD for U-Alb was 2.4 mg/l; U-Alb was below this LOD in 110 of the samples (33%). Analyses of urinary alpha-1-microglobulin (U-A1M) were performed using the α1-microglobulin ELISA Kit K6710 (Immundiagnostik AG, Bensheim, Germany) as described elsewhere.14 Calibrators provided in each kit, with target values in the ranges 0.09–0.28 mg/l, were always within the acceptable range. The LOD for U-A1M was 0.1 mg/l; one sample (0.3%) was below the LOD.
Analyses of U-Crea were performed in fresh urine using the Jaffé method (Roche Diagnostics, Mannheim, Germany) with a LOD of 0.01 mmol/l. SG was measured in fresh urine with a Ceti Digit 012 refractometer (Medline, Oxfordshire, UK).
Exclusions and Data Transformations
Four study participants had more than six of their U-Cd samples (i.e., >50%) below the LOD and were excluded from the analysis to minimise the potential risk of attenuating the total variability. Thus, the final analysis included 24 participants (14 men and 10 women) and a total of 288 urine samples with a mean volume of 294 ml (range 20–990 ml), and a mean sampling time of 3.1 h (range 1.5–10.3 h). The UON samples (n=48) had a mean volume of 407 ml (range 90–990 ml) and a mean sampling time of 8.3 h (range 6.2–10.3 h). The calculated U24 samples (n=48) had a mean urine volume of 1764 ml (range 805–3935 ml) and a mean sampling time of 24.0 h (range 20.4–25.5 h). The 24 participants had a mean age of 40 years (range 23–59) and a mean BMI of 24.1 kg/m2 (range 19.1–28.7 kg/m2). Men had a significantly higher BMI compared with women (means 25.2 kg/m2 versus 22.6 kg/m2; P=0.02), but otherwise there were no differences in background variables between men and women.
Excretion rates per hour and per 24 h of cadmium (U-Cd/h, U-Cd/24 h), Alb (U-Alb/h, U-Alb/24 h), and A1M (U-A1M/h, U-A1M/24 h) were calculated from urinary concentrations, volumes, and sampling times. Urinary flow (UF) rates were calculated from urinary volumes and sampling times. The urinary concentrations of cadmium, Alb, and A1M were adjusted for diuresis using U-Crea (U-CdCrea, U-AlbCrea, U-A1MCrea) and SG (U-CdSG, U-AlbSG, U-A1MSG). SGstandard=1.015 was used in the SG adjustment calculations.11
The circadian rhythm of U-Cd excretion was studied by calculating geometric means of individual arithmetic means (of the 2 days) for each time point on untransformed data. Differences between men and women were tested using the Wilcoxon rank sum test (PROC NPAR1WAY). Statistical significance was set at P<0.05 for all tests except where otherwise stated, and two-sided confidence intervals (CIs) were used. Values below the LOD were replaced with the LOD divided by the square root of 2 in the statistical calculations (GSD <3.0 μg/l).17 All calculations were performed with version 9.1 of the SAS software package (SAS Institute, Cary, NC, USA).
The results for each Cd biomarker in spot urine samples, U24 samples, and UON samples (U-Cd, U-CdCrea, U-CdSG, U-Cd/h, U-Cd/24 h, UON-Cd, UON-CdCrea, UON-CdSG, and UON-Cd/h) were tested for normality using the Shapiro-Wilks test (PROC UNIVARIATE) and visual inspection of histograms to determine whether untransformed or log-transformed data should be used in the subsequent analyses. Natural log-transformation was used for all measures, as the data were highly skewed. As multiple samples were available for each study participant (i.e., the samples were correlated), mixed-effects models1 were used (PROC MIXED) separately for each of the Cd biomarkers:
for i=1, 2,.., 24 participants and j=1, 2,.., 12 measurements (j=1, 2 for U24 and UON samples). Xij is the exposure level for the ith person in the jth measurement, μY is the fixed mean (log-transformed) exposure level for the population, bi is the random effect of the ith person, and eij is the random error for the jth measurement of the ith person. The model also contains additional fixed effects for U covariates (determinants and interaction terms, discussed below) C1, C2, …, Cu; ∂u are regression coefficients representing the U covariates. The random effects bi and eij are assumed to be mutually independent and normally distributed with means of zero, between-individual variance σ2bY, and within-individual variance σ2wY. To determine whether a common fixed mean exposure levels and common variances could be used for men and women (i.e., if μY(men)=μY(women, σ2bY(men)=σ2bY(women), and σ2wY(men)=σ2wY(women)), we calculated mixed-effects models (Eq. (1)) containing intercept and gender as fixed effects. Three different variance structures were compared: common between- and within-individual variances for men and women, distinct between-individual but common within-individual variances, and distinct between- and within-individual variances, using a likelihood ratio test (significance level; P<0.05) where the difference in −2loglikelihood follows a χ2 distribution.1
Three different mixed-effects models (1A–1C) were used in the analysis. Model 1A (the null model) contained only random effects, a global mean, gender (categorical; men=0, women=1), and variance structure according to the result of the likelihood ratio test mentioned above. Restricted maximum likelihood (REML) estimates of μY, σ2wY, and σ2bY were determined from this model for all measures of U-Cd excretion (unadjusted concentrations, concentrations adjusted for U-Crea or SG, and U-Cd excretion rates) for all spot urine samples, for UON samples, and for U24 samples.
In model 1B, the determinants and variance components for U-Cd excretions in the total material of spot urine samples were estimated by adding more covariates and interaction terms to model 1A. The covariates were U-Alb, U-A1M, UF, BMI, time (categorical), and age; the interaction terms were gender times all these covariates. Finally, in model 1C, the determinants and variance components for U24 Cd excretion and UON Cd excretions were estimated. As the number of observations was lower in model 1C than in model 1B, the following covariates and interaction term were selected to be added to model 1A to form model 1C, on the basis of the results from model 1B: U-Alb, U-A1M, UF, age, and gender*UF.
Before inclusion in the mixed-effects models, correlations between possible covariates were investigated using the Spearman correlation coefficient. Variables with a correlation coefficient above 0.5 were not included simultaneously in the model, to avoid multicollinearity. The urinary protein excretions were highly skewed, and natural log-transformed protein excretions (ln(Alb) and ln(A1M)) were used for these determinants and their interaction terms in the models. The final models were then selected using backwards stepwise elimination (P<0.1 for inclusion). The elimination process followed the hierarchy of the data, starting with elimination of interaction terms followed by determinants. Covariates with P>0.1 were kept in the model if the corresponding interaction term had a P-value below 0.1. For categorical data (i.e., gender and time), women and UON samples were chosen as the reference categories.
When comparing the variances between model 1A (null model) and 1B, and those between model 1A and 1C, the same significance level for inclusion in the models was used (P<0.1), instead of P<0.05 as formerly used in model 1A.
The natural-scale mean exposure level was estimated as , where . Estimates of 95% between- and within-individual fold-ranges (bR0.95 and wR0.95) were determined for each measure of U-Cd excretion using model 1A, where and . The estimated ratio of the between-individual biomarker variance to total observed variance, intraclass correlation (ICC), was calculated as .1
Different types of study might have different priorities in selecting the least biased measure of U-Cd excretion. For instance, in an epidemiological study, it is important to minimise the attenuation in a hypothetical log(exposure) to log(response) relationship in an individual-based study design; that is, the ratio between the regression slope estimated in the study (βest) and the true regression slope (βtrue). The degree of attenuation for a given measure of U-Cd excretion was determined from estimated variance components in model 1A using the relationship:
where the bias is 1-b, λ=σ2wY/σ2bY, and n is the number of repeated measurements per individual.1
If the study instead aims to detect a difference in the level of Cd exposure between groups, the number of individuals needed for a certain power can be compared for different measures. The number of samples per group (m) required in order to detect a statistically significant (P<0.05) difference of 10, 25, 50, or 100% in the geometric mean values with a statistical power (P) of 80% was calculated for 1, 2, or 3 samples per individual (n) using the total variance and the ICC from Eq. (1) with just a global mean, and the following formula:18
where Zα and Zβ are the αth and βth percentiles of a standard Gaussian distribution (one-tailed), α is the desired type I error (α=0.05), β is the desired type II error (β=1-P), and d is the difference in means of log-transformed concentration between the two groups.18
The study population had a low environmental Cd exposure. The natural-scale U-Cd level in UON was 0.08 μg/g Crea for men and 0.17 μg/g Crea for women (95% CI: 0.06–0.12 μg/g Crea and 0.11–0.26 μg/g Crea, respectively; P=0.009). The 24 h U-Cd excretion was 0.18 μg in both men and women (95% CI: 0.15–0.21 μg; Table 1).
Variability of U-Cd Excretion
Table 1 presents the estimated total variance and variance components from model 1A for unadjusted U-Cd, U-Cd adjusted for U-Crea and SG, and U-Cd excretion rate, in all spot urine samples, UON samples, and 24 h U-Cd samples. Likelihood ratio tests showed that distinct between- and within-individual variance should be used for unadjusted U-Cd and U-Cd/h in spot urine samples, and a common variance structure for the rest of the measures (Table 1).
The total variance ranged between 0.20 and 0.62, and for most biomarkers the between-individual variance dominated the total variability. The fraction attributed to the between-individual variance (i.e., the ICC) increased when U-Cd was adjusted for dilution and when the UON samples and the 24 h U-Cd excretion were used (Table 1). Expressed as fold ranges, 95% of the U-Cd excretions for a given individual were roughly within a two- to fourfold range for the 24 h excretion and UON samples and within a 4- to 10-fold range for all spot urine samples. Meanwhile, 95% of the participants had a U-Cd excretion roughly within a 4- to 10-fold range; that is, highly exposed individuals had 4–10 times higher exposure compared with low-exposed individuals (Table 1).
Circadian Rhythm of U-Cd, Protein Excretion, and UF Rate
Figure 1 presents the geometric means of arithmetic means per individual for UF and urinary excretion of Cd, Alb, and AIM at each of the six fixed sampling times. All measures showed a circadian rhythm with higher excretion of U-Cd and U-proteins and higher UF during the day-time compared with overnight. UF and the excretion of U-Cd and U-A1M were higher before noon and decreased during the afternoon, whereas the excretion of U-Alb was more stable throughout the day (Figure 1). The time of urine sampling was still a significant factor (P<0.001) for U-Cd excretion when other determinants such as gender, age, UF, and protein excretion were included in the mixed-effects model 1B (Table 2). Compared with the UON samples, U-Cd excretion was generally significantly higher in the morning sample (0930 hours) and significantly lower in the afternoon samples (1430, 1730 and 2200 hours; Table 2).
Determinants for U-Cd Excretion and Effect on the Variance
Gender, time (for spot urine samples), age, UF, and protein excretion (mainly U-Alb) were significant determinants for most measures of U-Cd excretion (Table 2, Table 3). With covariates included in the models, the total variance decreased by an average of 41% for spot urine samples and 33% for U24 and UON samples compared with the null model. In general, a decrease was seen for both the between- and the within-individual variance (Figure 2). As indicated above, time was an important determinant for U-Cd excretion in spot urine samples. When time was removed from model 1B, the variance increased by an average of 11% (average difference −4% for the between-individual variance and +40% for the within-individual variance).
Implications for Study Design
Using the estimated variance components from model 1A and Eq. (2), we calculated the number of repeated samples per individual that would be required to ensure a maximum of 20% bias in a hypothetical log(exposure) to log(response) relationship, in an individual-based study. The results showed that at least two (but often more) repeated samples per individual were required when using biomarkers in spot urine samples, whereas one sample per individual was sufficient when using 24 h U-Cd excretion or most U-Cd measures in UON samples (Figure 3).
The number of individuals per group required to detect a difference between two groups with a power of 80% (α=0.05) was calculated using Eq. (3) (Table 4). The 24 h U-Cd excretion required the smallest group sizes, but apart from this there were no substantial differences between the biomarkers in spot urine or UON samples (Table 4). The number of individuals per group depended highly on the size of the difference, with about 10 individuals per group needed to detect a difference of 100% and about 500 individuals for a difference of 10%. The number of repeated samples per individual in the groups had only a small effect on the group sizes (Table 4). Separate calculations for men and women are provided as Supplementary Information. In general, the number of individuals needed per group was lower for men than for women.
Variation in repeated U-Cd samples has been noted previously by others.3, 12, 19, 20, 21, 22 In this study, we estimated the short-term variability (within a day, and between two days mostly 4–6 days apart) of U-Cd excretion among healthy non-smoking men and women. This variability can lead to a biased prediction if results are used for epidemiologic studies of exposure–response relationships or studies comparing exposure levels between different groups.2
In our study, the major part of the variability of the investigated measures of U-Cd excretion could be attributed to the between-individual variance. A review on biomonitoring of occupational exposures reported the same result for exposures to substances with long half-life.23 Compared with all spot urine samples, 24 h U-Cd excretion and measures in UON samples had a lower σ2wY/σ2bY, indicating a larger reproducibility.
Others have reported a reduction of the variance when U-Cd concentrations in urine were adjusted for dilution.3, 19, 21 In our study, this was the case for adjustment by creatinine in spot urine samples but not for the UON samples adjusted for SG. However, adjustment for dilution increased the ICC for all measures.
In this study, 95% of the levels for a given individual varied from about 2- to 4-fold for U24 samples and from 4- to 10-fold for UON samples. A similar within-individual variation in U-Cd concentrations adjusted for dilution was seen in second morning samples taken for 10 subsequent weeks among 17 environmentally exposed women in Japan.21
In all spot urine samples, we were able to study the circadian rhythm for U-Cd excretion. Although the long-term U-Cd excretion for each individual is thought to be stable, due to the long half-life of Cd in the human body,4, 5, 6, 7 we found a circadian rhythm in the U-Cd excretion over 24 h. The highest levels were seen in the 0930 hours samples and the levels then decreased (Figure 1). The same result was also found for model 1B when more determinants were included in the models (Table 2). In order to find the reason for this circadian rhythm, we examined the variation in U-protein excretions and UF; the same pattern as for U-Cd was seen for U-protein excretion, especially U-A1M, and for UF. However, as time point of sampling was a significant factor for U-Cd excretion even after U-protein and UF were included in model 1B, these factors could not fully explain the variation in U-Cd excretion during the 24-h time period. Another factor that could affect U-Cd excretion (and the urinary excretion of proteins and UF) is circadian variations in the kidney function. In a study of 11 healthy individuals, a circadian rhythm in glomerular filtration rate (GFR) and renal plasma flow was seen for all participants during bedrest and standardised protein intake.24 GFR was found to be higher during the day-time compared with overnight, and UF was found to have the same circadian rhythm as GFR.24 A study of occupationally exposed subjects revealed a circadian rhythm for U-Cd excretion, with a decreased excretion overnight that was attributed to the rhythm of GFR.22 When the effect of GFR on U-Cd excretion was investigated by water loading and water restriction among 19 occupationally exposed subjects, a significant effect of GFR was seen on the U-Cd excretion.25 As a result of this circadian rhythm in U-Cd excretion during the day, a standardised sampling time, such as first morning urine sample, needs to be applied in order to be able to compare levels within and between studies.
Gender and age were significant determinants of U-Cd excretion for most U-Cd biomarkers in our study; this is as expected, as women generally have higher Cd absorption because of iron deficiency,8 and Cd accumulates in the human body because of the long half-life of Cd in humans.4, 5, 6, 7 There was a significant effect of UF on the U-Cd excretion, especially for all spot urine samples. This indicates that the adjustment for dilution was not perfect, as discussed by others.10, 11, 26, 27
For most of the U-Cd biomarkers, U-protein excretion was found to significantly predict the U-Cd excretion among the study participants, a result which has also been as reported recently.13, 28, 29 The association could be interpreted as normal variability in renal physiology affecting both U-Cd and U-protein. The impact of U-protein and UF on 24 h U-Cd excretion was limited and not statistically significant; the reason for this could be that temporary variation in renal physiology has less influence when the urine sampling time is long. Moreover, the determinants of 24 h U-Cd and UON-Cd excretions were analysed using a smaller number of observations than for spot urine samples.
In individually-based epidemiological studies, when a biomarker is used to construct an empirical log(exposure) to log(response) relationship, the ratio of within-individual to between-individual variance dictates the likely degree of attenuation according to Eq. (2). Hence, this ratio can be used to select the least biased biomarker. A biomarker with a low such ratio has a high contrast in exposure between the participants, and generates less bias with the same number of repeated samples per individual compared with a biomarker with a higher ratio.1, 2 Using only one sample per study participant, Cd in U24 or UON samples adjusted for dilution would result in a high accuracy (a bias <20%) in the observed slope coefficient (Figure 3), whereas biomarkers in spot urine samples typically need more than two repeated samples per participant to reach this level of accuracy. Therefore, U-Cd concentrations in UON samples adjusted for dilution would be the favoured biomarkers in this kind of study. Our results indicate that adjustment for creatinine is better than adjustment for SG in this respect.
For studies of differences in exposure levels between groups, Eq. (3) suggests that the combination of a small total variance and a small σ2bY is best. U-Cd in 24 h and U-Cd/h permitted somewhat lower group sizes, but apart from this, no large differences were seen between the other measures (Table 4). The result did not differ even between all spot urine and UON samples, reflecting the fact that the total variance was similar for these types of samples. The variances were higher in women than in men, and so larger group sizes (50% for spot urine samples and 150% for UON samples) are required for women to obtain the same accuracy as for men (Supplementary Information).
The number of repeated samples per individual did not have a large effect on the group sizes, indicating that a high study power would be better achieved by increasing sample size than by increasing the number of measurements per person, as has also been shown for U-proteins.30 As U24 samples are laborious to collect, and incomplete or contaminated samples might pose problems, when aiming to achieve a high power in this kind of study appropriate group size for an anticipated difference between the investigated groups is more important than the choice of U-Cd biomarker.
A strength of this study was that the repeated samples were collected over a short time-span, in order to avoid any impact of age or life-style changes. The study included both men and women, which was also accounted for in the models. The study design enabled us to investigate circadian rhythms and to estimate the variability of different U-Cd biomarkers. The study population had a low level of environmental Cd exposure, and only non-smokers were recruited; this limits the generalizability to healthy non-smoking men and women with a low level of Cd exposure. A limitation is that in the final data set, 18% of the samples used had U-Cd below the LOD, but these values were estimated according to the distribution of data17 and participants with more than 50% of samples below the LOD were excluded to minimize the potential risk of attenuating the total variability. The study participants were not randomly selected from the general population, but we believe that they are representative of the general population with regard to their between-individual and within-individual variability of U-Cd excretion.
In conclusion, between-individual variability dominated the total variability for most U-Cd measures, especially in U24 and overnight samples. The U-Cd excretion varied during the day. Gender, UF, age, and U-protein were also significant determinants for U-Cd excretion, explaining part of the variability in this measure. When planning a study, this variability must be acknowledged.
Rappaport SM, Kupper LL . Quantitative Exposure Assessment. Stephen Rappaport: El Cerrito, California, USA. 2008.
Lin YS, Kupper LL, Rappaport SM . Air samples versus biomarkers for epidemiology. Occup Environ Med 2005; 62: 750–760.
Mason HJ, Williams NR, Morgan MG, Stevenson AJ, Armitage S . Influence of biological and analytical variation on urine measurements for monitoring exposure to cadmium. Occup Environ Med 1998; 55: 132–137.
EFSA. Cadmium in food—scientific opinion of panel on contaminants in the food chain. EFSA J 2009; 980: 1–139.
Jarup L, Berglund M, Elinder CG, Nordberg G, Vahter M . Health effects of cadmium exposure—a review of the literature and a risk estimate. Scand J Work Environ Health 1998; 24 (Suppl 1): 1–51.
Nordberg GF, Nogawa K, Nordberg M, Friburg LT Cadmium. In: Nordberg GF, Fowler B, Nordberg M, Friberg LT (eds). Handbook on the Toxicology of Metals, Chapter 23 3rd edn Academic Press: London. 2007 pp 445–486.
Akerstrom M, Barregard L, Lundh T, Sallsten G . The relationship between cadmium in kidney and cadmium in urine and blood in an environmentally exposed population. Toxicol Appl Pharmacol 2013; 268: 286–293.
Berglund M, Akesson A, Nermell B, Vahter M . Intestinal absorption of dietary cadmium in women depends on body iron stores and fiber intake. Environ Health Perspect 1994; 102: 1058–1066.
Akerstrom M, Lundh T, Barregard L, Sallsten G . Sampling of urinary cadmium: differences between 24-h urine and overnight spot urine sampling, and impact of adjustment for dilution. Int Arch Occup Environ Health 2012; 85: 189–196.
Berlin A, Alessio L, Sesana G, Dell'Orto A, Ghezzi I . Problems concerning the usefulness of adjustment of urinary cadmium for creatinine and specific gravity. Int Arch Occup Environ Health 1985; 55: 107–111.
Suwazono Y, Akesson A, Alfven T, Jarup L, Vahter M . Creatinine versus specific gravity-adjusted urinary cadmium concentrations. Biomarkers 2005; 10: 117–126.
Trevisan A, Nicoletto G, Maso S, Grandesso G, Odynets A, Secondin L . Biological monitoring of cadmium exposure: reliability of spot urine samples. Int Arch Occup Environ Health 1994; 65: 373–375.
Akerstrom M, Sallsten G, Lundh T, Barregard L . Associations between urinary excretion of cadmium and proteins in a nonsmoking population: renal toxicity or normal physiology? Environ Health Perspect 2013; 121: 187–191.
Andersson L, Haraldsson B, Johansson C, Barregard L . Methodological issues on the use of urinary alpha-1-microglobuline in epidemiological studies. Nephrol Dial Transplant 2008; 23: 1252–1256.
Akerstrom M, Lundh T, Barregard L, Sallsten G . Effect of molybdenum oxide interference on urinary cadmium analyses [Letter]. Int Arch Occup Environ Health 2013; 86: 615–617.
Jarrett JM, Xiao G, Caldwell KL, Henahan D, Shakirova G, Jones RL . Eliminating molybdenum oxide interference in urine cadmium biomonitoring using ICP-DRC-MS. J Analytical Atomic Spectrometry 2008; 23: 962–967.
Hornung RW, Reed LD . Estimation of average concentration in the presence of nondetectable values. Appl Occup Enviro Hyg 1990; 5: 46–51.
Li Z, Romanoff LC, Lewin MD, Porter EN, Trinidad DA, Needham LL et al. Variability of urinary concentrations of polycyclic aromatic hydrocarbon metabolite in general population and comparison of spot, first-morning, and 24-h void sampling. J Expo Sci Environ Epidemiol 2010; 20: 526–535.
Arisawa K, Nakano A, Honda S, Saito H . Reproducibility of urinary beta 2-microglobulin and cadmium excretion among residents in a cadmium-polluted area during a 3-year period. Toxicol Lett 1997; 91: 147–152.
Ikeda M, Ezaki T, Tsukahara T, Moriguchi J, Furuki K, Fukui Y et al. Reproducibility of urinary cadmium, alpha1-microglobulin, and beta2-microglobulin levels in health screening of the general population. Arch Environ Contam Toxicol 2005; 48: 135–140.
Yamagami T, Suna T, Fukui Y, Ohashi F, Takada S, Sakurai H et al. Biological variations in cadmium, alpha 1-microglobulin, beta 2-microglobulin and N-acetyl-beta-D-glucosaminidase in adult women in a non-polluted area. Int Arch Occup Environ Health 2008; 81: 263–271.
Yokoyama K, Araki S, Sato H, Aono H . Circadian rhythms of seven heavy metals in plasma, erythrocytes and urine in men: observation in metal workers. Ind Health 2000; 38: 205–212.
Symanski E, Greeson NM . Assessment of variability in biomonitoring data using a large database of biological measures of exposure. AIHA J (Fairfax, Va) 2002; 63: 390–401.
Koopman MG, Koomen GC, Krediet RT, de Moor EA, Hoek FJ, Arisz L . Circadian rhythm of glomerular filtration rate in normal individuals. Clin Sci (Lond) 1989; 77: 105–111.
Araki S, Aono H, Yokoyama K, Murata K . Filterable plasma concentration, glomerular filtration, tubular balance, and renal clearance of heavy metals and organic substances in metal workers. Arch Environ Health 1986; 41: 216–221.
Elkins HB, Pagnotto LD, Smith HL . Concentration adjustments in urinalysis. Am Ind Hyg Assoc J 1974; 35: 559–565.
Greenberg GN, Levine RJ . Urinary creatinine excretion is not stable: a new method for assessing urinary toxic substance concentrations. J Occup Med 1989; 31: 832–838.
Chaumont A, Nickmilder M, Dumont X, Lundh T, Skerfving S, Bernard A . Associations between proteins and heavy metals in urine at low environmental exposures: evidence of reverse causality. Toxicol Lett 2012; 210: 345–352.
Haddam N, Samira S, Dumont X, Taleb A, Lison D, Haufroid V et al. Confounders in the assessment of the renal effects associated with low-level urinary cadmium: an analysis in industrial workers. Environ Health 2011; 10: 37.
Stengel B, Watier L, Chouquet C, Cenee S, Philippon C, Hemon D . Influence of renal biomarker variability on the design and interpretation of occupational or environmental studies. Toxicol Lett 1999; 106: 69–77.
We thank Lena Samuelsson and Caroline Johansson for help with the data collection. For funding, our thanks go to the Graduate School in Environment and Health, a cooperation between the University of Gothenburg, the Chalmers University of Technology, and the Västra Götaland Region, coordinated by the Centre for Environment and Sustainability (GMV).
The authors declare no conflict of interest.
Supplementary Information accompanies the paper on the Journal of Exposure Science and Environmental Epidemiology website
About this article
Cite this article
Akerstrom, M., Barregard, L., Lundh, T. et al. Variability of urinary cadmium excretion in spot urine samples, first morning voids, and 24 h urine in a healthy non-smoking population: Implications for study design. J Expo Sci Environ Epidemiol 24, 171–179 (2014). https://doi.org/10.1038/jes.2013.58
- spot urine
- urinary excretion
- 24 h urine
- study design
Exposure to heavy metals and the risk of osteopenia or osteoporosis: a systematic review and meta-analysis
Osteoporosis International (2020)
Temporal stability of urinary cadmium in samples collected several years apart in a population of older persons
International Journal of Hygiene and Environmental Health (2019)
The association between cumulative cadmium intake and osteoporosis and risk of fracture in a Chinese population
Journal of Exposure Science & Environmental Epidemiology (2019)
Cadmium exposure and early renal effects in the children and adults living in a tungsten-molybdenum mining areas of South China
Environmental Science and Pollution Research (2018)
Journal of Exposure Science & Environmental Epidemiology (2018)