Articles Variability in Airborne and Biological Measures of Exposure to Mercury in the Chloralkali Industry: Implications for Epidemiologic Studies Elaine Symanski,1 Gerd SAlisten, and Lars Barregdrd2 'University of Texas - Houston School of Public Health, Houston, Texas, USA; 2Department of Occupational Medicine, Sahigrenska University Hospital, Goteborg, Sweden Exposure assessment is a critical component of epidemiologic studies, and more sophisticated approaches require that variation in exposure be considered. We examined the intr- and interindi- vidual sources of variation in exposure to mercury vapor as measured in air, blood, and urine among four groups of workers during 1990-1997 at a Swedish chloralii plant. Consistent with the underlying kdnetics of mercury in the body, the variability of biological measures was dampened considerably relative to the variation in airborne levels. Owing to the efect of individual ria tion, estimat workers' exposures from a few measurements can attenuate measr of effiet To examine such effects on studies relating long-term exposure to a continuous health outcome, we evaluated the utility of each eosure measure by comparing the nece y sample sz required for accurate estimtion of a slope coefficient obtained from a ression analysis. No single measure outperfonned the others for all groups of workers. However, when workers were evaluated together, creatiine-corrected urnary mercury better disciminated workers' exposures than airborne or blood mercury leves. Thus, pilot studies should be conducted to examine variability in both air and bio- monitoring data because uantitatve information about the relative m itude of the intra- and interindividual sources of variation feeds directly into our efforts to desig an optial samplig strategy when evaluating health risks assocated with occupational or environmental contaminants. Key words. biological marker, exposure variability, mercury, variance components. Environ Heal Penpect 108:569-573 (2000). [Online 5 May 2000] http://ehpnetl.niehs. nib.goldocs/2000/08p569-573symanski/abstracthrml Of critical importance when assessing the utility of an exposure measure are questions related to exposure variability. Two major components of exposure variability are the variation that occurs from day-to-day (intraindividual variation) and the variation that occurs among workers (interindividual variation). For airborne measures of expo- sure, intraindividual variation occurs as a result of myriad factors related to the process and the work environment or it may be due, to a lesser extent, to measurement errors associated with sampling and analysis (1). Interindividual variation in airborne conta- minant levels among workers has been attributed to differences in job tasks or work practices (2). Intraindividual variation in the external exposure is transmitted, at least to some degree, to biological levels of the contami- nant or its metabolite in body fluids. In addition, sources of biological variability are likely to induce fluctuations in contaminant concentrations in urine or blood over time. Such variability in biomarker levels associat- ed with occupational exposure has been restricted thus far to a single investigation (3), but some endogenous constituents in blood and urine are marked by considerable intraindividual variation (4-7), and the same physiologic parameters are likely to exert similar effects on body burdens of contami- nants. Moreover, the variation between workers in exposure levels would contribute to interindividual variability in biomarker levels. Ethnicity, sex, age, anthropometric and lifestyle factors, and physiologic differ- ences in the rates of uptake, metabolism, and elimination are also likely to play a role. Intraindividual variability in exposure can induce error in exposure assessment and thereby can adversely affect epidemiologic studies by reducing the power to detect asso- ciations and by diminishing measures of effect (8-13). To assess the magnitude of the error in an exposure measure, the well-estab- lished techniques of analysis of variance can be applied (when repeated measurements on study subjects are available) to estimate the magnitude of the intra- and interindividual sources of variation. Information contained in the estimated variance components can then be used to assess the bias in measures of effect and to optimize study design in terms of the number of workers to be studied and the number of samples to collect. In the chloralkali industry, exposure to mercury vapor (Hg0) can occur during the production of chlorine through the electroly- sis of a brine solution in mercury cells (14). Exposure can be monitored by measuring mercury in the breathing zone of exposed workers using either active or passive personal sampling techniques (15) or by biomonitor- ing mercury in urine or whole blood. The pri- mary aim of the present study was to examine the intra- and interindividual sources of varia- tion in levels of mercury in the air, urine, and blood among workers at a Swedish chloralkali plant during the 1990s. Using information about exposure variability, a secondary objec- tive was to investigate whether airborne or biological measures of exposure might be more suitable for use in an epidemiologic study by comparing the minimum sample sizes necessary to minimize the attenuation of regression results when health-effects studies are carried out. Materials and Methods Compilation of the database. During the period 1990-1997, no major changes in pro- duction were implemented at the chloralkali plant involved in the current investigation. Review of the company's data on annual emissions of mercury from the cell hall revealed that the output remained relatively constant over the study period. To evaluate workers' exposures to mercury, both air and biological monitoring were conducted. One blood sample and two urine samples were typically collected on each worker per year. First-morning urine samples were collected at home in metal-free polyethylene bottles, and blood was collected by venipuncture in metal-free heparinized vacutainers at the health-care center of the plant. Personal exposures were evaluated during the full work shift by active sampling on Hydrar tubes (16). Nearly all workers participated in the biomonitoring program, but approxi- mately one-half of the workforce was moni- tored by personal sampling. We analyzed air samples using standard methods (16); determinations of mercury in biological samples were made using cold vapor atomic absorption spectrophotometry (17). To correct for urinary flow rate, mer- cury concentrations in urine were adjusted for creatinine, which was analyzed with a modified kinetic Jaffd method (18). There Address correspondence to E. Symanski, 1200 Herman Pressler Drive, RAS Suite W-642, Houston, TX 77030 USA. Telephone: (713) 500- 9238. Fax: (713) 500-9249. E-mail: esymanski@ sph.uth.tmc.edu This research was supported by grant KOI 0H00166 from the National Institute for Occupational Safety and Health, the Swedish Medical Research Council in Stockholm, and by the Wallenberg Foundation and the Medical Faculty at Gbteborg University. Received 21 October 1999; accepted 14 January 2000. Environmental Health Perspectives * VOLUME 108 1 NUMBER 61 June 2000 569 Articles * Symanski et al. was a change in the laboratory responsible for the analysis of the biomonitoring data in 1994. Quality assurance has been presented elsewhere for the analyses before 1994 (19). From 1996 onward, analyses of external ref- erence samples showed acceptable results (Centre de Toxicologie du Quebec, Sainte- Foy, Quebec, Canada; interlaboratory com- parison). The detection limit for airborne mercury was 0.5 jg/m3. For urinary and blood mercury, the limit of detection was 10 nmol/L through 1992 and 5 nmol/L thereafter. We used laboratory reports for 1990- 1997, provided by the company's health and personnel unit, to compile a database of both air and biological monitoring measure- ments. Before the data were entered, they were inspected to identify outliers, which were subsequently investigated to ascertain possible coding errors. In the absence of any errors, the original data were left intact. Blood, urinary, and air mercury values were recorded in units of nanomoles per liter, micrograms per gram creatinine, and micro- grams per cubic meter, respectively. Uncor- rected urinary values (nanomoles per liter) were compiled as well. For the biomonitor- ing database, measurements on workers exposed to mercury vapor for < 1 year were excluded because their exposure regimen was not sufficiently long to reasonably assume steady-state conditions. We omitted urine samples that were either too dilute (< 0.5 g creatinine/L) or too concentrated (> 3 g cre- atinine/L) (20). Very few urinary (< 1%) and blood (5%) mercury measurements were below the limit of detection; all such mea- surements were flagged and assigned a level of two-thirds the value of the reported detec- tion limit (21). There were no undetectable air mercury values. With assistance from company personnel, we ascertained the job titles of the workers. We created four broad occupational cate- gories based on the following classifications: a) shift workers; b) cell cleaners, basement flushers, and mercury-pump repairmen (here- after referred to as "cell hall maintenance workers"); c) cell hall foremen, "egalisators" (voltage regulators of the mercury cells), and cell switchers (hereafter referred to as "cell hall production workers"); and d) instrument technicians, mechanical workshop workers and foremen, staff electricians, operating engi- neers, plastic workshop workers, and labora- tory workers (hereafter referred to as "non-cell hall workers"). Both the cell hall production workers and the cell hall maintenance workers typically spend the majority of their time in the cell hail, whereas the non-cell hall workers spend < 10% of their time in the cell hall. Shift workers run the process for 24 hr, rotate between day and night shifts, and perform numerous tasks in the control room, salt solu- tion hall, and cell hall. Due to small sample sizes and the irregularity of the work per- formed, data collected on workers who spend no time in the cell hall (e.g., washers or store- room workers) and external workers (e.g., painters or electricians) were not evaluated. We examined temporal effects by visually inspecting graphs of the annual mean levels for the air and biological monitoring data col- lected over the period 1990-1997. Although the air mercury levels appeared to fluctuate erratically above and below the mean value for the entire period, a shift in exposure levels in 1994 was apparent for the biological moni- toring data, especially urinary mercury. Thus, a systematic change in the urinary and blood mercury levels was evaluated when sources of exposure variability were examined. Also, his- tograms of the air, blood, and urinary mea- surements suggested that the data were approximately lognormally distributed; as such, the natural logarithms of the data were used in subsequent analyses. Intra- and interindividual sources of variation in exposure levels. To assess the intra- and interindividual sources of variation in airborne measures of exposure to mercury, we applied a one-way random-effects model (22. For the biological monitoring data, we applied a mixed-effects model to evaluate pos- sible differences that may have arisen from a change in laboratories in mid-1994. Briefly, the mixed model is specified as follows: Yi'k= =ln yk) =j+iY++Ps+k [+ 1 ] for i= 1 (1990-June 1994) or 2 (July 1994-1997) time periods, j= 1,2, ..., b workers, k = 1, ..., n. measurements per worker, and where u ijk = the exposure concentration for the ith worker on the kth day during the ith period, Y.k = the natural logarithm of the exposure concentration, iy=the overall mean (mean of Y.j, = the fixed effect due to the measurement having been collected during the ith period, p= the random effect measuring the devia- tion of the jth worker's true exposure from (g y+ al), and k= the random effect measuring the devia- tion of the jth worker's exposure from (py + ci + f on the kth day during the ith period. It is assumed that the ai values sum to zero and thus have a population variance defined as a la - 1 where a = 2 in our two-period situation. It is further assumed that ,B. and ?i.k are mutu- ally independent and normally distributed with zero means and variances c2 and or respectively. Thus, aS and cW represent the interindividual and intraindividual variance components. It follows that E(YI) = jy+ ai for all i, j, k, Var(Y1'A) = aB + cYWfor all i, j, k, and Cov(Y;, Yjk)= (B for k?# k' and for all i, and . For the air, blood, and urinary mercury data, analyses were run separately on each occupational group of workers and on all workers combined. Restricted maximum like- lihood estimates of the variance components were obtained using PROC MIXED available with SAS software (SAS Institute, Cary, NC). Effects of measurement error on accurate estimation of regression coefficients. To assess the influence of measurement error in air or biological levels of mercury, we constructed a hypothetical scenario in which estimates of average levels of the log-transformed mer- cury values in air, blood, or urine for each worker were used to examine the relation with a continuous health outcome measure (e.g., neuropsychologic deficits or changes in renal function). It was further assumed that there were no other explanatory vari- ables to consider as covariates in the linear model; as such, a simple linear regression model could be applied to examine the exposure-response relation. Under the standard assumptions that underlie simple (unweighted) linear regres- sion analysis (23), the expected value of the observed slope coefficient E(5) can be expressed as follows (24): E(I) = 1{+Ain1) [2] where P is the true slope coefficient, X is the ratio of the intra- and interindividual vari- ance components for the exposure variable (i.e., X = - 2 G 2) and n is the number of measurements obtained from each worker. Except in instances when the intraindividual variance component is zero, the observed slope coefficient (under expectation) is small- er than the true coefficient [i.e., E(1)/lP < 1]. For example, if only single measurements were available for each worker, the observed slope obtained from a regression analysis would be one-half as large as the true slope when the intra- and interindividual variances are equal to one another (L = 1). Following algebraic manipulation, Equation 2 can be easily rearranged to estimate samples sizes (n) that would be necessary to minimize the attenuation of an observed slope coefficient [E(4)/I] to specified levels. Effects of measurement error on accurate estimation of regression coefficients. Relying on estimates of the variance components VOLUME 108 1 NUMBER 6 1 June 2000 * Environmental Health Perspectives 570 Articles * Mercury levels in air, urine, and blood obtained from the models (62 and 62) sam- pie sizes were estimated to minimize the attenuation of an observed slope coefficient to 90%, 75%, and 60% of the true value. Assessments were made separately for mer- cury in air, blood, and urine for each group, as well as for all workers across job cate- gories. All statistical analyses were performed using SAS software (SAS Institute). Results Compilation of the database. Although infor- mation on lifestyle factors was not available, there were relatively few differences in the mean age of workers across job groups in our study (data not shown). During this period, 282 air measurements were collected on 42 workers. Far greater numbers of blood (n = 646) and urine samples (n = 955) were col- lected. Among all workers, the median num- ber of repeated measurements was 4 air sam- ples, 6 blood samples, and 13 urine samples. The arithmetic means ? 1 SD for air, blood, and urinary mercury levels during this 7-year period were 22 ? 35 pg/m3, 30 ? 23 nmol/L, and 10 ? 9 pg/g creatinine [79 ? 66 nmol/L], respectively. Correspondingly, the geometric means (geometric SD) were 12 pg/m3 (2.8) for airborne mercury, 24 nmol/L (2.8) for blood mercury, and 8 jig/g creatinine (2.1) [59 nmol/L (2.2)] for urinary mercury. Nearly 62% of the air measurements were performed for maintenance workers, whereas most blood and urine mercury samples were collected from shift workers, cell hall mainte- nance workers, and non-cell hall workers. When workers were dassified by occupational category, the highest and lowest exposures were typically observed for the cell hall mainte- nance workers and shift workers, respectively. Intra- and interindividual sources of variation in exposure levels. Point estimates of the variance components in the log-trans- formed air, blood, and urinary mercury data are shown in Table 1. For airborne mercury, the proportion of the total variability attribut- able to the intraindividual source of variation differed among groups. Cell hall maintenance workers and shift workers were characterized by extreme day-to-day variability; little varia- tion was detected between workers. There appeared to be as much or greater variation among individuals compared to variation across shifts in both the cell hall production workers and non-cell hall workers. There was greater variation between, rather than within, workers for the biomonitoring data when all workers were combined, whereas equivocal results were obtained when the analyses were conducted by occupational group. Based on the mixed-model analyses, exposure levels appeared to decrease in the latter period (mid- 1994 onward), especially for urinary mercury levels (data not shown). Table 2 shows the estimates of the num- ber of repeated measurements per worker required to minimize the attenuation of an observed slope coefficient to 90%, 75%, and 60% of its true value for air, blood, and urinary measures. When workers were evalu- ated together irrespective of occupational category, the sampling requirements were reduced for mercury measured in blood or urine as compared to those for air. Across occupational groups, the sampling demands varied, and in some instances sizeable differ- ences were noted. Discussion Effects related to intraindividual variation have long been recognized in the statistical and epidemiologic literature (10,11). How- ever, the quantification of the inter- and intraindividual sources of exposure variability in the occupational arena has focused primarily on airborne contaminant levels (22,25-27). Similar investigations of varia- tion in biological measures of exposure to workplace contaminants have, to our knowl- edge, been restricted to a single study of workers exposed to styrene at a boat manu- facturing plant (3). In our study we found that a substantial percentage of the variability in airborne mercury levels was due to day-to- day variation, which was nearly 50% or high- er in all groups of workers. This finding is in agreement with an investigation of variability in airborne contaminants across a broad cross-section of workplaces worldwide (22). Our results also confirm that fluctuations of daily airborne mercury levels are smoothed in both the body burdens of mercury in blood and, to a greater extent, to that in urine. Given that the damping of variability in air exposures is highly dependent on the contaminant's half-life in the body (28), these results are consistent with the underly- ing kinetics of mercury in blood and in urine, with slow elimination phases of several weeks and 2-3 months (29-31), respectively. Based on kinetic considerations alone, urinary mercury may be deemed a superior measure relative to blood mercury because exposures are integrated over longer periods (32). Yet our results for the entire group of chloralkali plant workers indicate that similar numbers of measurements would be required if blood or uncorrected urinary mercury were used to estimate individual workers' mean levels in a regression analysis (Table 2). Because variations in urinary flow rate (e.g., due to variable water intake) increase the variability in urinary mercury concentrations in spot samples (33), creatinine-corrected uri- nary mercury produced less variable results and thus yielded the expected benefits when compared to mercury in blood. Nevertheless, in situations when the primary aim of biolog- ical monitoring is to detect temporary peak exposures rather than to assess the long-term body burden of mercury, mercury in blood would be a superior measure, owing to the damping of such peaks in urinary levels. Table 1. Inter- and intraindividual sources of variation (5B and d,) for log-transformed air, blood, and urinary mercury data collected on Swedish chloralkali plant workers during 1990-1997. n ba Median n. a5B 2 ib Airborne Hg (pg/m3) Shift workers 56 17 4 0.09 0.70 7.9 Cell hall production workers 19 4 4 0.56 0.46 0.83 Cell hall maintenance workers 174 15 9 0.06 0.82 13 Non-cell hall workers 33 8 3 0.50 0.50 1.00 All workers 282 42 4 0.39 0.74 1.9 Blood Hg (nmol/L) Shift workers 185 38 5 0.06 0.23 4.16 Cell hall production workers 76 6 13 0.13 0.10 0.81 Cell hall maintenance workers 176 18 7 0.14 0.17 1.27 Non-cell hall workers 209 30 7 0.36 0.23 0.65 All workers 646 87 6 0.23 0.20 0.87 Urinary Hg (nmol/L)c Shift workers 472 41 14 0.08 0.18 2.41 Cell hall production workers 49 6 8 0.04 0.11 2.95 Cell hall maintenance workers 130 17 4 0.05 0.23 4.33 Non-cell hall workers 296 30 9 0.31 0.31 1.01 All workers 947 88 13 0.32 0.23 0.73 Urinary Hg (pg/g creatinine) Shift workers 474 41 14 0.05 0.11 2.2 Cell hall production workers 49 6 8 0 0.08 - Cell hall maintenance workers 130 17 4 0.08 0.12 1.6 Non-cell hall workers 302 30 8 0.32 0.19 0.59 All workers 955 88 13 0.32 0.14 0.45 Abbreviations: n, total number of measurements; b, number of workers; n., number of repeated measurements per worker. aA few workers held more than one job title over the study period. bX -vAo c2 a few instances, the lab reports indi- cated urinary mercury levels only in units of micrograms per gram creatinine. Thus, there were slightly fewer uncorrect- ed measurements (nanomoles per liter) compared to the creatinine-corrected values. Environmental Health Perspectives * VOLUME 108 1 NUMBER 6 1 June 2000 571 Articles * Symanski et al. The proportion of the intraindividual variability to the total variance generally decreased in levels of mercury in blood or urine when compared to air mercury levels. A notable exception was the group of shift workers for which the percentage of variation attributable to intraindividual variability was higher in biological levels (especially in blood mercury) as compared to airborne levels. In this group, the geometric mean level of blood mercury (reflecting both inorganic and organic mercury exposure) was 18 nmol/L, which is only slightly higher than that found in the general Swedish population (34,35). It is likely that the greater intraindividual varia- tion relative to the total variability in blood mercury levels in shift workers is due to fluc- tuations in exposures from nonoccupational sources (primarily from contaminated fish and amalgam fillings) (35), which play a big- ger role in influencing body burdens of cont- aminants when workplace exposures are low. Relying on quantitative estimates of the intra- and interindividual sources of variation in exposure to mercury as measured in the air, blood, and urine among workers at a Swedish chloralkali plant, we also evaluated effects on regression results should such data be used to examine long-term health effects associated with mercury exposure. As shown in Figure 1, our results suggest that the underestimation of the regression coefficient can be substantial when limited numbers of measurements are collected (although the benefits of collecting additional measure- ments diminish with increasing sample size). Although requisite sample sizes are not the only factor to consider when evaluating exposure measures, estimating the distribu- tion of measurement errors and quantifying differences among measures provide invalu- able information that can be used to plan future investigations. Although random and mixed-effects models offer clear advantages when evaluating the nature of workplace exposures, important questions must be addressed related to the variance-covariance structure of the data. In the statistical models that were applied, we assumed that the covariance between mea- surements collected on the same worker was the same regardless of the interval separating them. Shift-long airborne samples were often collected repeatedly on the same worker over the course of a few days, but studies have indicated relatively little serial correlation in air monitoring data (36-38). Biological mon- itoring data may lack independence; however, the extent to which such data are autocorre- lated will be a function of both the half-life of the contaminant in the body and the timing of sampling. In our study, when the interval between measurements for each individual was computed, we found that only 10% of the urinary data were < 4 months apart. Likewise, only 1% of the blood measure- ments was collected at intervals of < 1 month. Thus, errors associated with an improper specification of the variance-covariance struc- ture are unlikely to have affected our results. Another issue related to the proper speci- fication of the model when data across occu- pational groups are combined stems from the assumption that the intraindividual variance was the same for all groups of workers. Based on our stratified analyses, it appears that the magnitude of the intraindividual variance varies by occupational group. Thus, the assumption of variance homo- geneity may be violated when data across groups are combined. Although some effects related to the misspecification of variance components models have been evaluated (39), the robustness of such models when underlying assumptions are violated war- rants further investigation. Finally, our investigation focused on an exposure assessment strategy that relies on an individual-based approach in which each individual worker's exposure is evaluated. Should a group-based approach be adopted instead, study questions may be focused either on evaluating the individual group mean exposure levels (in which case the mixed- effects model would evaluate job group as a fixed effect) or on assessing the degree of vari- ability across groups (in which case the mixed-effects model would evaluate job group as a random effect). Questions related to the relative merits of both approaches are being evaluated in another investigation. Whether biological monitoring offers advantages compared to air monitoring depends on kinetic factors as well as on the relative magnitude of the inter- and intrain- dividual sources of variation in each expo- sure measure. It is interesting to note that Rappaport et al. (3) found that personal sampling measurements of airborne styrene Table 2. The number of measurements per worker that would be required to obtain an observed coefficient that is 90, 75, and 60% of the true value.a 90% Airborne Hg (pg/m3) Shift workers Cell hall production workers Cell hall maintenance workers Non-cell hall workers All workers Blood Hg (nmol/L) Shift workers Cell hall production workers Cell hall maintenance workers Non-cell hall workers All workers Urinary Hg (nmol/[) Shift workers Cell hall production workers Cell hall maintenance workers Non-cell hall workers All workers Urinary Hg (pg/g creatinine) Shift workers Cell hall production workersb Cell hall maintenance workers Non-cell hall workers All workers 75% 71 7 119 9 17 24 2 40 3 6 37 7 11 6 8 12 2 4 2 3 22 27 39 9 7 7 9 13 3 2 20 15 5 4 7 5 2 1 I slope - . 60% - o U S 12 m. 20 1 -4 20 M 3 .30 1 1 6 a ._ 1 00 2 E 4 4 7 2 3 2 l 1 No. of ropeatd emsuremeatWworker Figure 1. Accuracy in the observed slope coeffi- cient as a function of the number of measurements collected on each worker for the log-transformed air (Air-Hg), blood (B-Hg), and urinary mercury (U- Hg) data collected on Swedish chloralkali workers during 1990-1997 (see Equation 2 in text). The esti- mated ratio of the intra- to interindividual variance component (X) was 1.9 for Air-Hg, 0.87 for B-Hg, 0.73 for uncorrected U-Hg, and 0.45 for creatinine- corrected U-Hg. VOLUME 108 1 NUMBER 6 1 June 2000 * Environmental Health Perspectives "Slope coefficient estimated from a simple linear regression relating mercury exposure to a continuous health outcome. bNo calculation was possible given that d2 = 0. 572 Articles * Mercury levels in air, urine, and blood yielded the least biased measure when com- pared to measurements of styrene in exhaled air among boat-manufacturing workers, whereas one of the biological measures of exposure performed the most efficiently in the current investigation. In any case, our investigation demonstrates that quantitative information about intra- and interindividual sources of variation in exposure can be used to design efficient sampling strategies when evaluating health risks associated with work- place or environmental contaminants. REFERENCES AND NOTES 1. Nicas M, Simmons BP, Spear RC. Environmental versus analytical variability in exposure measurements. Am Ind Hyg Assoc J 52:553-557 (1991). 2. Rappaport SM. Assessment of long-term exposures to toxic substances in air. Ann Occup Hyg 35:61-121 (1991). 3. Rappaport SM, Symanski E, Yager JW, Kupper LL. The relationship between environmental monitoring and bio- logical markers in exposure assessment. Environ Health Perspect 103(suppl 3):49-54 (1995). 4. Harris EK, Kanofsky P, Shakarji G, Cotlove E. Biological and analytical components of variation in long-term studies of serum constituents in normal subjects. II. Estimating biological components of variation. Clin Chem 16:1022-1072 (1970). 5. Pickup JF, Harris EK, Kearns M, Brown SS. Intra-individ- ual variation of some serum constituents and its rele- vance to population-based reference ranges. Clin Chem 23:842-850 (1977). 6. Knuiman JT, Hautvast JGA, Van Der Heijden L, Geboers J, Joossens JV, Tornqvist H, lsaksson B, Pietinen P, Tuomilehto J, Flynn A, et al. A multi-centre study on with- in-person variability in the urinary excretion of sodium, potassium, calcium, magnesium and creatinine in 8 European centres. Hum Nutr: Clin Nutr 40:343-348 11988). 7. Borel MJ, Smith SM, Derr J, Beard JL. Day-to-day varia- tion in iron-status indices in healthy men and women. Am J Clin Nutr 54:729-735 (1991). 8. Brunekreef B, Noy D, Clausing P. Variability of exposure measurements in environmental epidemiology. Am J Epidemiol 125:892-898 (1987). 9. Armstrong BG. Effect of measurement error on epidemi- ological studies of environmental and occupational exposures. Occup Environ Med 55:651-656 (1998). 10. Cochran WG. Errors of measurements in statistics. Technometrics 10:637-666 (1968). 11. Liu K, Stamler J, Dyer A, McKeever J, McKeever P. Statistical methods to assess and minimize the role of intra-individual variability in obscuring the relationship between dietary lipids and serum cholesterol. J Chronic Dis 31:399-418 (1978). 12. Armstrong BG. The effects of measurement errors on relative risk regressions. Am J Epidemiol 132:1176-1184 (1990). 13. Rosner B, Spiegelman D, Willett WC. Correction of logis- tic regression relative risk estimates and confidence intervals for random within-person measurement error. Am J Epidemiol 136:1400-1414 (1992). 14. Sallsten G, BarregArd L, Jarvholm B. Mercury in the Swedish chloralkali industry-an evaluation of the expo- sure and preventive measures over 40 years. Ann Occup Hyg 34:205-214 (1990). 15. Sallsten G, Barregard L, Langworth S, Vesterberg 0. Exposure to mercury in industry and dentistry: a field comparison between active and diffusive samplers. Appi Occup Environ Hyg 7:434-440 (1992). 16. NIOSH. Mercury. NIOSH Method 6009. In: NIOSH Manual of Analytical Methods, 3rd ed. Cincinnati, OH:National Institute for Occupational Safety and Health, 1989;6009-1 to 6008-4. 17. Einarsson 0, Lindstedt B, Bergstrom T. A computerized automatic apparatus for determination of mercury in bio- logical samples. J Autom Chem 6:74-79 (1984). 18. Lustgarten JA, Wenk RE. Simple, rapid kinetic method for serum creatinine measurement. Clin Chem 18:1419-1422 (1972). 19. Vesterberg 0. Automatic method for quantitation of mer- cury in blood, plasma, and urine. J Biochem Biophys Methods 23:227-235 (1991). 20. Alessio L, Berlin A, Dell'Orto A, Toffoletto F, Ghezzi I. Reliability of urinary creatinine as a parameter used to adjust values of urinary biological indicators. Int Arch Occup Environ Health 55:99-106 (1985). 21. Hornung RW, Reed LD. Estimation of average concentra- tion in the presence of nondetectable values. AppI Occup Environ Hyg 5:46-51 (1990). 22. Kromhout H, Symanski E, Rappaport SM. A comprehen- sive evaluation of within- and between-worker compo- nents of occupational exposure to chemical agents. Ann Occup Hyg 37:253-270 (1993). 23. Kleinbaum DG, Kupper LL, Muller KE, Nizam A. Applied Regression Analysis and Other Multivariable Methods, 3rd ed. Pacific Grove, CA:Duxbury Press, 1998. 24. Tielemans E, Kupper LL, Kromhout H, Heederik D, Houba R. Individual-based and group-based occupational expo- sure assessment: some equations to evaluate different strategies. Ann Occup Hyg 42:115-119 (1998). 25. Rappaport SM, Kromhout H, Symanski E. Variation of exposure between workers in homogeneous exposure groups. Am Ind Hyg Assoc J 54:654-662 (1993). 26. Kumagai S, Kusaka Y, Goto S. Cobalt exposure level and variability in the hard metal industry of Japan. Am Ind Hyg Assoc J 57:365-369 (19961. 27. Lagorio S, lavarone 1, lacovella N, Proietto AR, Fuselli S, Baldassarri LT, Carere A. Variability of benzene exposure among filling station attendants. Occup Hyg 4:15-30 (19971. 28. Rappaport SM. Smoothing of exposure variability at the receptor: implications for health standards. Ann Occup Hyg 20:201-214 11985). 29. Sallsten G, BarregArd L, Schitz A. Decrease in mercury concentration in blood after long term exposure: a kinet- ic study of chloralkali workers. Br J Ind Med 50:814-821 (1993). 30. Sallsten G, Barregard L, Schuitz A. Clearance half life of mercury in urine after the cessation of long term occu- pational exposure: influence of a chelating agent (DMPS) on excretion of mercury in urine. Occup Environ Med 51:337-342 (1994). 31. Roels HA, Boeckx M, Ceulemans E, Lauwerys RR. Urinary excretion of mercury after occupational expo- sure to mercury vapor and influence of the chelating agent meso-2,3-dimercaptosuccinic acid (DMSA). Br J Ind Med 48:247-253 (1991). 32. Droz PO, Berode M, Wu MM. Evaluation of concomitant biological and air monitoring results. AppI Occup Environ Hyg 6:465-474 (1991). 33. Wallis G, Barber T. Variability in urinary mercury excre- tion. J Occup Med 24:590-595 (1982). 34. Barregard L. Biological monitoring of exposure to mer- cury vapor. Scand J Work Environ Health 19(suppl 1):45-49 (1993). 35. Langworth S, Elinder C-G, Gothe C-J, Vesterberg 0. Biological monitoring of environmental and occupational exposure to mercury. Int Arch Occup Environ Health 63:161-167 (1991). 36. Francis M, Selvin S, Spear R, Rappaport SM. The effect of autocorrelation on the estimation of worker's daily exposures. Am Ind Hyg Assoc J 50:37-43 11989). 37. Kumagai S, Matsunaga I, Kusaka Y. Autocorrelation of short-term and daily average exposure levels in work- places. Am Ind Hyg Assoc J 54:341-350 11993). 38. Symanski E, Rappaport SM. An investigation of the dependence of exposure variability on the interval between measurements. Ann Occup Hyg 38:361-372 (1994). 39. Symanski E, Kupper LL, Kromhout H, Rappaport SM. An investigation of systematic changes in occupational exposure. Am Ind Hyg Assoc J 57:724-735 (1996). Environmental Health Perspectives * VOLUME 1081 NUMBER 6 1 June 2000 573