The epidemiologic evidence for the carcinogenicity of lead is inconsistent and requires improved exposure assessment to estimate risk. We evaluated historical occupational lead exposure for a population-based cohort of women (n=74,942) by calibrating a job-exposure matrix (JEM) with lead fume (n=20,084) and lead dust (n=5383) measurements collected over four decades in Shanghai, China. Using mixed-effect models, we calibrated intensity JEM ratings to the measurements using fixed-effects terms for year and JEM rating. We developed job/industry-specific estimates from the random-effects terms for job and industry. The model estimates were applied to subjects’ jobs when the JEM probability rating was high for either job or industry; remaining jobs were considered unexposed. The models predicted that exposure increased monotonically with JEM intensity rating and decreased 20–50-fold over time. The cumulative calibrated JEM estimates and job/industry-specific estimates were highly correlated (Pearson correlation=0.79–0.84). Overall, 5% of the person-years and 8% of the women were exposed to lead fume; 2% of the person-years and 4% of the women were exposed to lead dust. The most common lead-exposed jobs were manufacturing electronic equipment. These historical lead estimates should enhance our ability to detect associations between lead exposure and cancer risk in the future epidemiologic analyses.
Lead exposure is widespread in occupational settings, with at least 120 different occupations1 and an estimated 0.5–1.5 million US workers currently exposed to lead in their work.2 In occupational settings lead is predominantly inhaled as lead fume or lead dust. Lead fumes are generated when lead is vaporized during high temperature operations such as welding, cutting, burning, soldering, and melting. The lead vapors condense at room temperature into fumes composed of lead oxide particulates that are generally 1 μm or less in diameter3 and then coagulate to form dust particles. The fume particulates are small enough to reach the lung’s alveoli where absorption efficiencies are nearly 100%.4 Lead dust is also generated by a number of different mechanical operations including cutting, grinding, drilling, dry mixing, chipping, and sanding. The chemical composition and size of lead-dust particulates will vary based on the source material and the work being performed and thus the absorption efficiencies and the lung deposition patterns will vary.
Lead has been linked to numerous adverse health effects, such as abdominal colic, anemia, kidney damage, various neurological disorders, and cancer.5 Lead is classified as a probable carcinogen (Group 2A) by the International Agency for Research on Cancer based on sufficient animal data and limited human data.6 Thus far, epidemiological studies evaluating lead exposure and cancer risk have been inconclusive,6, 7, 8, 9 although recently there is more consistent evidence for increased risk of cancers of the stomach, brain, urinary bladder, and kidney with lead exposure.10, 11, 12, 13, 14, 15, 16, 17 Given the widespread degree of lead exposure and the difficulty in evaluating rare diseases such as brain and kidney cancers in industry-based studies, it is important to improve exposure assessment for general population and case–control studies to generate new evidence.
In this paper, we describe our exposure assessment efforts to estimate historical exposures to lead fume and lead dust for the Shanghai Women Health Study (SWHS), a prospective, population-based cohort study that consists of ∼75,000 Shanghai women.18 The SWHS cohort resided in a highly industrialized area of Shanghai, with nearly all women working outside home and thus provides an opportunity to examine occupational cancer risk factors among women. Historical exposure to lead fume and lead dust were estimated separately for this population because they can vary in their particle size and composition. To do so, we combined lead-fume- and lead-dust-specific job-exposure matrices (JEMs) with a database of lead-fume and lead-dust inspection measurements using a statistical modeling framework previously developed to estimate historical benzene exposure in this cohort.19 This framework used mixed-effects models to calibrate the JEM intensity ratings across time and by rating to the measurement data. The framework also used the models to calculate job and industry-specific lead exposure estimates across time based only on the measurements. The use of measurements to account for the change of lead exposure levels over time, in particular, was essential, because the work histories reported by the study subjects spanned 70 years, over which exposure circumstances changed substantially.20, 21 In this paper, we describe the models, the trends in lead-fume and lead-dust exposure in Shanghai, China across four decades and the application of the models to the SWHS cohort. The lead-fume and lead-dust exposure estimates developed here will be used to examine exposure-disease associations in the SWHS cohort in future analyses.
The Shanghai Women’s Health Study (SWHS) is a prospective, population-based cohort that was enrolled from 1997 to 2000 and has been previously described.18 The cohort includes 74,941 women aged 40–70 years who lived in urban communities in Shanghai, China. At study enrollment, the study subjects reported all jobs held for at least 1 year in a self-reported work history calendar, which included questions on the job title, type of business, factory name, description of the work tasks, and the employment dates for all jobs. Complete work histories were obtained for 74,658 women. The work history records were coded to 3-digit job and industry codes based on the Standard Chinese Classification of Industries and Occupations for the third national census (1982).
Shanghai Center for Disease Control and Prevention Lead Measurement Database
A database of inspection measurements collected between 1954 and 2000 during factory inspections by the Shanghai Center for Disease Control and Prevention (Shanghai CDC) was compiled. The database includes measurements for many agents, including lead dust and lead fume. Measurements were generally short-term, area samples taken with a variety of sampling methods. An analysis of the benzene measurements from this database was previously reported.19 Each measurement identified the factory name, job sampled, industry, sampling date, and air concentration and was classified to 3-digit job and industry codes based on the above-mentioned 1982 Standard Chinese Classification system. After we excluded environmental measurements and duplicates, 20,084 lead-fume and 5383 lead-dust measurements were available.
Population-based JEMs for lead fume and for lead dust were developed separately for the Shanghai Women’s Health Study. Industrial hygienists (JC, SX) rated the intensity and probability of lead-fume and lead-dust exposure on an ordinal scale (0 to 3) for all 3-digit job and industry codes reported in the subjects’ work histories. The ratings were assigned separately for each job and industry; thus, each work history record was assigned both a job rating and an industry rating for probability and for intensity. Probability was defined as the estimated proportion of exposed workers in the job or the industry: 0 for no exposure; 1 for <5% workers exposed; 2 for 5–49% workers exposed; and 3 for >50% of workers exposed. Intensity rating was estimated based on the maximum allowable concentration (MAC), which has been 0.03 mg/m3 for lead fume and 0.05 mg/m3 for lead dust, respectively, since 1979.22 The intensity categories were defined as follows: 0 for negligible exposure; 1 for <10% of MAC; 2 for 10–100% of MAC; and 3 for >100% of MAC.
Measurement Data Treatment
In the Shanghai measurement database there were 60 three-digit job and 90 three-digit industry codes with lead-fume or lead-dust measurements resulting in 881 unique job/industry combinations. As most combinations had few measurements, two industrial hygienists (DK, SL) reviewed these codes and combined the jobs, industries, and industries within jobs based on likely similarities in exposure scenarios. Briefly, 3-digit occupation codes were collapsed into 34 job groups and 3-digit industry codes were collapsed into 44 industry groups based on professional judgment. If a resulting occupation/industry combination had fewer than 10 measurements, the team further combined similar industry groups within that job group. The procedure resulted in 34 final job groups and 242 final job/industry groups (industry collapsed within job). This procedure was done blind to the JEM ratings and magnitude of the exposure concentrations.
The lead-fume and lead-dust measurements were positively skewed and were log-transformed (base e) for all analyses. We treated any measurement with a reported value below 0.01 mg/m3 (the most frequently reported low-level exposure concentration) as a sample below the limit of detection (LOD) for both lead fume (33% <LOD) and lead dust (13% <LOD) throughout the study period. For the LOD samples, we imputed values assuming that the data was log-normally distributed across database, with an exposure distribution defined by all the measurements above the LOD.23 A value for each LOD sample was randomly drawn from the exposure distribution, with the draw repeated five times, to obtain five data sets of imputed estimates for measurements below the LOD and real measurements for measurements above the LOD.
The modeling framework, used here to combine the JEM ratings and the inspection measurements to predict lead-fume and lead-dust concentrations for this cohort, was previously developed to estimate historical benzene exposure for the same cohort.19 We used a mixed-effects model, with the JEM intensity ratings and calendar year incorporated as fixed effects, and the job and the industry nested within job incorporated as random effects. The fixed effects allowed us to calibrate each JEM rating to a concentration scale across time using the measurements. The random effects for job and industry allowed us to calculate job-specific and job/industry-specific estimates by using both the fixed-effects parameters and the best linear unbiased predictors (BLUPs) obtained from an empirical Bayes estimation procedure.24 The BLUPs are a shrinkage estimator that “shrinks” the estimate toward the corresponding JEM rating estimate when the measurements are sparse and/or highly variable and “pulls” the estimate toward the job or job/industry mean when more measurements are available and/or the exposure variability is low.25 The structure of the model is shown in Equation 1.
Log-transformed lead fume and lead dust, Ln(Yjifm), were examined in separate mixed-effects models. The fixed effects included a spline time trend for year of measurement (Y, see Supplementary Table S1) and JEM intensity ratings for job (Ijob) and industry (Iind). For the time trend, we varied the number of knots (from 0 to 5) and degrees of freedom (from 1 to 4) in increments of 1 to determine the b-spline time trend, which minimized the model’s -2 log-likelihood fit statistics. The random effects were job group (Jj), industry group nested within job group (Ind(J)ji), and inspection occurrence (Ojif). Inspection occurrence accounted for potential correlation in measurements collected within the same factory on the same inspection date (number of measurements per occurrence: mean=3.6; maximum=59). Each random effect was assumed to be normally distributed, with a mean of 0. Variance components were calculated using the restricted maximum likelihood (REML) method. The model parameters and variance components were calculated separately for each of the five imputed data sets and combined using “proc mianalyze” to derive the overall parameter estimates, SEs, and confidence limits for each model parameter (SAS code provided in Friesen et al.19). All analyses were conducted using SAS (version 9.2; SAS Institute, Cary, NC, USA).
Application to Cohort
We used the JEM probability rating to determine when to assign an exposure estimate to the study subjects. We used stringent criteria to define lead exposure, because specificity is more important than sensitivity when the exposure prevalence is low.26 For both lead fume and lead dust, a work history record was assumed exposed and the predicted concentrations from the model assigned if the job probability rating was 3 or if the industry probability rating was 3 and the job probability rating was ≥1. Work history records that did not meet the above criteria were assumed unexposed.
For the exposed jobs, we calculated calibrated JEM estimates and job/industry-specific estimates for each study year. The calibrated JEM lead-fume and lead-dust estimates were calculated using the fixed-effect parameters from the relevant model. Job/industry-specific estimates were calculated using the fixed-effect parameters and the BLUP estimates from the random effects for job and industry. If the subject’s reported job/industry combination was not included in the exposure database or there were fewer than five measurements for that job/industry combination, we did not use the BLUP estimate for that industry. Instead, we calculated the lead estimates using only the fixed-effect parameters and the BLUP estimate for that job group. For study years 1965 through 2000, the b-spline time trend was used to predict annual exposure levels for both the calibrated JEM estimates and the job/industry-specific. For study years pre-1965, we assigned the predicted concentrations from 1965, because the data was too sparse to provide a stable time trend in this period (<5% of the data and <5% of the person-years). We calculated cumulative exposure estimates for both the calibrated JEM estimate and the job/industry-specific estimate for each cohort subject.
Calibrated JEM vs Job/Industry-Specific Estimates
We compared the calibrated JEM estimates to the job/industry-specific estimates for the year 1980 (median employment year) using the Pearson correlation statistics for the subset of job/industry groups meeting our definition of lead-fume or lead-dust-exposed. We calculated the ratio of the 97.5th to the 2.5th percentile (BGR95) of the predicted job/industry geometric mean (GM) as a measure of the ability of the model to discriminate between high and low-exposed jobs and industries.27 For the subset of exposed study subjects, we calculated the Pearson correlation statistics between the calibrated JEM and job/industry-specific cumulative estimates.
For lead fume, the overall mean concentration was 0.51 mg/m3 (GM=0.03 mg/m3; geometric SD, GSD=11; 95th percentile=1.4 mg/m3; maximum=570 mg/m3). For lead dust, the mean concentration was 1.9 mg/m3 (GM=0.18 mg/m3, GSD=10, 95th percentile=9 mg/m3; maximum=92 mg/m3). The number of measurements generally increased over time, peaked in the late 1980s, and then decreased consistently through 2000 (Figure 1). The median measurement year was 1985 for lead fume (fifth percentile=1965) and 1982 for lead dust (fifth percentile=1963). There were four times more lead-fume measurements than lead-dust measurements. The mean employed person-year in the study cohort was 1980 (fifth percentile=1960).
The distribution of the lead-fume and lead-dust measurements by job and industry JEM intensity ratings are shown in Supplementary Table S2. For lead fume, 47% of the measurements were associated with a job or industry intensity rating of 3. For lead dust, 87% of the measurements were associated with a job or industry intensity rating of 3. Only 14% of the fume and 0.4% of the dust measurements had both a job and industry rating of 1.
The model parameters for the fixed effects and the variance components are shown in Table 1. A third to almost half of the database’s large GSD was due to within-occurrence variance for both lead fume and lead dust. Compared with a model that included the random effects, but no fixed effects (not shown), the time trend and JEM intensity ratings explained 16% and 26% of the between-job variability, 37% and 15% of the between-job/industry variability, and 16% and 5% of the between-occurrence variability for the lead-fume and lead-dust models, respectively.
Overall, the predicted lead-fume and lead-dust concentrations declined over time, except for a slight rise in the mid-1980s (Figure 2). Exposure levels were ∼20-fold and 50-fold higher in 1965 than in 2000 for lead fume and lead dust, respectively. For lead fume, the b-spline parameterization with the lowest -2 log-likelihood fit statistics had two knots and two degrees of freedom. For lead dust, the best parameterization had one knot and two degrees of freedom.
In 1980, the predicted concentrations increased with increasing job intensity rating. The predicted concentrations for job intensity ratings 1 and 2 were 14% and 72% of the predicted concentration for rating 3 for lead fume, respectively, and 38% and 51% of rating 3 concentrations for lead dust, respectively. For lead fume, we observed an interaction effect between the job and industry JEM intensity ratings that occurred only when the job intensity rating was 2 or 3. The predicted concentrations for industry intensity rating 1 and 2 were 20% and 33% of the predicted concentration for industry intensity rating 3. We did not observe an interaction for industry JEM rating for lead dust, which is likely due to sparse data in some combinations (Supplementary Table S2).
Application to the Cohort
The distribution of employed person-years in the cohort study by job and industry JEM probability ratings is shown in Table 2. Based on our assignment criteria, 5% of the person-years were exposed to lead fume, 2% were exposed to lead dust, and 2% were exposed to both lead fume and lead dust. Calibrated JEM estimates based on the fixed-effects terms were assigned to all exposed person-years. Job/industry-specific estimates were assigned based on the fixed-effects terms and both the job and industry BLUPs for 75% of the lead-fume-exposed person-years and 85% of the lead-dust-exposed person-years. The remaining exposed person-years were assigned job/industry-specific estimates based on the job BLUP, but not the industry BLUP, because of sparse data in that job/industry group. Overall, 5920 women (8%) were ever exposed to lead fume and 2762 women (4%) were ever exposed to lead dust.
The most commonly reported jobs and job/industry combinations that were exposed to lead fume or lead dust in this cohort are shown in Table 3, along with the calibrated JEM and job/industry-specific estimates. For lead fume, “install/assemble electric/electronic equipment” was the most frequently reported exposed job. For lead dust, “other electric/electronic equipment install/maintenance” was the most frequently reported exposed. Exposure estimates for job/industry combinations are shown in Supplementary Table S3 (available online). For lead fume, “foundry workers in agricultural machine industry” was the most highly exposed job/industry (job/industry-specific estimate: 0.227 mg/m3in 1980). For lead dust, “workers engaged in packing in miscellaneous electronic equipment industry” was the most highly exposed job/industry (job/industry-specific estimate: 0.355 mg/m3 in 1980).
Calibrated JEM vs Job/Industry-Specific Estimates
For lead fume, the BGR95 was 7.3 for the calibrated JEM job/industry estimates and 10.4 for the job/industry-specific estimates. For lead dust, the BGR95 was 2.7 for the calibrated JEM estimates and 7.5 for the job/industry-specific estimates. The higher BGR95 for the job/industry-specific estimates indicates that we gain more discrimination than when the JEM estimates are used. This greater discrimination is shown in Figure 3, which shows the distribution of the job/industry-specific estimates for each intensity ratings for the year 1980. The calibrated JEM estimate for each intensity ratings is represented in the figure as a point estimate.
The Pearson correlation coefficients between the calibrated JEM estimates and the job/industry-specific estimates in 1980 for lead fume and lead dust were 0.81 and 0.70, respectively. Similarly, the correlation between the cumulative calibrated JEM estimates and cumulative job/industry-specific estimates for the cohort subjects was 0.79 for lead fume and 0.84 for lead dust. There was little correlation between the lead fume and lead dust cumulative estimates (based on calibrated JEM estimates, r=0.06; based on job/industry-specific estimates, r=0.15).
In this paper, we combined Shanghai-specific job and industry JEMs with a database of inspection measurements to estimate retrospective exposure levels of lead fume and lead dust for future epidemiologic analyses in a prospective cohort of Shanghai women. To do so, we calibrated the JEM intensity ratings for jobs and job/industry combinations across time using measurements associated with the same job and job/industries. We also calculated job/industry-specific estimates based on those same measurements that refined the JEM estimates when there were sufficient data.
Lead exposure concentrations decreased substantially over time, with predicted exposure levels ∼20 times and 50 times lower in 2000 than in 1965 for lead fume and lead dust, respectively. This observed decline in exposure levels is generally consistent with previous lead exposure studies in China,20, 21, 22, 28, 29, 30 although the slight rise from the mid-1970s to mid-1980s is less common. This slight rise in exposure levels could be due to increased production or an increased emphasis on evaluating lead exposures, so that more measurements were taken (Figure 1) with a greater focus on high exposures. In China, exposure levels in several lead storage battery plants decreased from 9 mg/m3 in the 1950s to 0.2 mg/m3 in the 1980s, reflecting a 50-fold exposure decrease.20, 21, 22, 28, 29, 30 This substantial exposure decline has a large influence on the cumulative exposure levels and will lead to exposure misclassification in the epidemiological study if treated as constant. Our estimated time trends, however, are not likely to have captured potential job or industry-specific differences in time trends. We were unable to evaluate such differences because of insufficient data and thus, we may have introduced exposure misclassification for specific jobs and industries for which time trends differ.
Exposure concentrations increased monotonically with increasing JEM job intensity ratings for both lead fume and lead dust and with JEM industry ratings for lead fume. These differences provide evidence that overall, the industrial hygienists who rated lead exposure intensity were able to identify broad differences in lead exposure. These broad differences were not consistent with the category cut points used by the industrial hygienists. For lead fume, the categories were 0, >0–0.003, 0.003–0.03, and >0.03 mg/m3; however, the predicted GMs from the model for the year 1980 (the median person-year) were 0.03, 0.04 and 0.06 mg/m3. This difference suggests that the industrial hygienists may have underestimated exposure levels but more likely it reflects, at least in part, the influence of the higher exposures in 1980. It is also likely that the model overestimated the lower-exposed categories because inspection measurements are generally collected where overexposure is suspected. Thus, the jobs selected for measurement may typically be low exposed, but the specific jobs measured may have been more highly exposed, and thus, the jobs may not necessarily represent the average profile of all jobs. Given that industrial hygienists generally try to select the highest-exposed situation for any exposed job, this may not have resulted in a substantial differential bias. The between-group ratio (BGR95) provides an estimate of the relative difference between high and low-exposed groups. For lead fume, high-exposed combinations were approximately one order of magnitude higher than the low-exposed combinations (BGR95 of 7.3 and 10.4 for the calibrated JEM and job/industry-specific estimates, respectively). In contrast, for lead dust, only a 3–7-fold difference between high and low combinations was observed (BGR95 of 2.7 and 7.5 for the calibrated JEM and job/industry-specific estimates, respectively). The lower between-group ratio for lead dust may reflect that lead-dust-exposed job/industry groups were more similar in exposure concentration, but it may also reflect the inability of the models to capture this difference because of the smaller number of measurements or reflect the above-mentioned potential for selective sampling of jobs and industries. For these reasons, we anticipate greater exposure misclassification in the lead dust estimates than in the lead-fume measurements.
A JEM intensity rating assumes the same exposure level for all jobs/industry combinations with that rating. We were able to refine the calibrated JEM intensity ratings by calculating job/industry-specific estimates using a framework previously used to estimate historical benzene levels.19 The magnitude of the additional refinement, for example, in the year 1980 was, however, much lower for lead than for benzene. Lead fume and lead dust had a between-group ratio that was 1.4 and 3 times higher for the job/industry-specific estimates than for the JEM estimates, whereas for benzene the between-group ratio was nearly seven times higher for the job/industry-specific estimates than for the calibrated JEM estimates. This difference reflects a reduced exposure discrimination that is also reflected in the higher correlation between the annual calibrated JEM and job/industry-specific estimates for lead fume (r=0.8) and lead dust (r=0.7) than for benzene (r=0.6). The smaller influence of the job/industry-specific estimates for lead may be because fewer lead measurements were available (fume: 20,084; dust 5383), in contrast to the substantial number of benzene measurements (over 60,000), from which to derive stable BLUPs to modify the JEM estimates. For the cumulative exposure metrics, the correlation between the calibrated JEM and the job/industry-specific estimates was high for all three agents (lead fume, r=0.79; lead dust, r=0.84; benzene, r=0.88), which suggests that the BLUPs were less influential once duration of exposure was incorporated.
Overall, 8% and 4% of the subjects were identified as ever exposed to lead fume and lead dust, respectively. This exposure prevalence is higher than the 1% lead exposure prevalence reported in Chinese female controls in a separate study7 and lower than the 9–17% exposure prevalence previously reported in lead-exposed controls in mixed gendered population-based case–control studies in other countries.12, 15, 31 The striking differences between studies may reflect gender differences in exposure. For instance, common lead-exposed jobs, such as construction-work,32 were not commonly reported by the women in our study.
Lead-fume and lead-dust exposure can co-occur. Making assessment of lead exposure even more complicated is that often lead-fume coagulates to eventually form a particle that meets the definition of dust. For example, in lead battery manufacturing both lead-fume and lead-dust exposure can occur in the pasting and assembly processes. In the inspection database, 13% of the unique job/industry combinations with lead-fume measurements also had five or more lead-dust measurements. Similarly, 57% of the unique job/industry combinations with lead-dust measurements had five or more lead-fume measurements. Some of this overlap may, in part, reflect coding errors for the substance in the database. For example, the most frequent lead-fume-exposed job in Table 3 is “Install/Assemble Electric/Electronic Equipment”, which is the second most frequent lead-dust-exposed job. As both are in the electrical/electronics industry, the source of exposure could be the same (e.g., solder). After applying our exposure criteria to high probability lead-exposed jobs, there was less overlap for lead fume than for lead dust: 38% of lead-fume-exposed person-years were considered also exposed to lead dust and 82% of lead-dust-exposed person-years were considered also exposed to lead fume. Despite this overlap, we developed exposure prediction models separately for lead fume and lead dust because both types of lead particles have different properties that affect their toxicity. Lead-dust particles are much larger than lead-fume particles.3 When inhaled these larger particles are captured in upper respiratory tract, where they are cleared, swallowed and moved to the gastrointestinal tract, where lead absorption efficiencies are much lower.33 The smaller and lighter lead-fume particles are more biologically available than a comparable mass of the larger, more chemically diverse dust particles, making a direct comparison between the two sample results challenging.11 The separate and joint health effects of lead fume and lead dust will be considered in future epidemiological analyses within this cohort.
Many of the limitations related to the use of the short-term, area, inspection measurements were previously described for the benzene measurements in this database.19 These limitations include potential sampling of worst-case or higher exposed exposure scenarios and factories, which may overestimate the average exposure levels, and the use of area measurements rather than personal measurements, which may overestimate or underestimate exposure depending on where the monitor is stationed in relation to work tasks. Generally, full-shift exposures are evaluated in epidemiologic studies, so that using measurements of short-term exposures would presumably result in a higher bias than had the measurements been full-shift samples. In addition, we assumed that the exposure measurements in the database represented women’s exposures in the workplace, which is reasonable because the database includes measurements collected both on men and women and because women represent ∼50% of the manufacturing workers in Asian countries.34 However, exposure misclassification may be introduced with this assumption because previous studies have shown that men and women may have different exposure concentrations within the same job; unfortunately, the direction of gender differences cannot be predicted in advance.35
There were also limitations specific to the lead measurements in this study. First, we were unable to capture differences in exposure levels that might be due to changes in sampling and analytic methods, because the information on sampling and analytic methods was not available within the database. Second, lead exposure measurements collect the “total dust” distribution of particle sizes, but for some health outcomes, different particle size fractions, such as respirable particles that reach the gas-exchange regions of the lungs, might be more relevant.36 In general, respirable lead particle concentrations are positively correlated with total lead-dust concentrations within an industry.3, 37 However, striking differences in the proportion of respirable particles were found across industries. For instance, the proportion of respirable particles was 43–49% in lead smelters and radiator soldering and 11–17% in lead battery and lead powder manufacturing.3 Third, we assumed a constant exposure level pre-1965 based on the 1965 exposure estimates because the data were too sparse to estimate a stable time trend in that period. This could affect the slope of exposure-response associations, as it would underestimate the exposure levels of subjects who worked in the early years. Fourth, three fume measurements exceeded 100 mg/m3, which is an unrealistically high concentration. However, predicted job/industry-specific estimates from models that both included and excluded these outliers were highly correlated (Spearman ρ=0.99) and the predicted concentration differences with and without the outliers were <0.001 mg/m3. As their negligible influence, we chose not to remove these potential outliers from the inspection data set.
In summary, we developed exposure prediction models for lead fume and lead dust for the SWHS cohort by combining information from both a JEM and a database of inspection measurements. Our models allowed us to describe time trends in lead-fume and lead-dust exposure in Shanghai over nearly five decades and to estimate job- and job/industry-specific lead dust and fume estimates, which were then used to estimate quantitative historical lead exposure levels to the study subjects. These historical lead estimates will be used in future epidemiologic analyses to enhance our ability to detect associations between lead exposure and cancer risk in this cohort.
NRC. Lead in the Human Environment. Committee on Lead in Human Environment, National Research Council, National Academy of Science: Washington, DC, 1980.
ATSDR. Toxicological profile for lead. Agency for Toxic Substances and Disease Registry, US Department of Health and Human Services, Public Health Service: Atlanta, 2007.
Park DU, Paik NW . Effect on blood lead of airborne lead particles charaterized by size. Ann Occup Hyg 2002; 46: 237–243.
Hodgkins DG, Robins TG, Hinkamp DL, Schork MA, Levine SP, Krebs WH . The effect of airborne lead particle size on worker blood-lead levels: an empirical study on battery workers. J Occup Med 1991, 1265–1273.
Moline JM, Landrigan PJ . Lead. In: Rosenstock L, Cullen MR, Brodkin CA, Redlich CA, (eds). Textbook of Clinical Occupational and Environmental Medicine, 2nd edn. Elsevier Saunders, Philadelphia, 2005 pp 967–978.
IARC. Inorganic and organic lead compounds. IARC monographs on the evaluation of carcinogenic risks to humans, vol. International Agency for Research on Cancer 2006; 87: 228–237.
Fu H, Boffetta P . Cancer and occupational exposure to organic lead compounds: a meta-analysis of published data. Occup Environ Med 1995; 52: 73–81.
Jemal A, Graubard BI, Devesa SS, Flegal KM . The association of blood lead level and cancer mortality among whites in the United States. Environ Health Perspect 2002; 110: 325–329.
Steenland K, Boffetta P . Lead and cancer in humans: where are we now? Am J Ind Med 2000; 38: 295–299.
Bhatti P, Stewart PA, Hutchinson A, Rothman N, Linet MS, Inskip PD et al. Lead exposure, polymorphisms in genes related to oxidative stress, and risk of adult brain tumors. Cancer Epidemiol Biomarker Prev 2009; 18: 1841–1848.
Boffetta P, Fontana L, Stewart P, Zaridze D, Szeszenia-Dabrowska N, Jaar B et al. Occupational exposure to arsenic, cadmium, chromium, lead and nickel, and renal cell carcinoma: a case-control study from Central and Eastern Europe. Occup Environ Med 2011; 68: 723–728.
Cocco P, Ward MH, Dosemeci M . Occupational risk factors for cancer of the gastric cardia. Analysis of death certificates from 24 US states. J Occup Environ Med 1998; 40: 855–861.
Golabek T, Darewicz B, Borawska M, Markiewicz R, Socha K, Kudelski J . Lead concentration in the bladder tissue and blood of patients with bladder cancer. Scand J Urol Nephrol 2009; 43 (6): 467–470.
Rajaraman P, Stewart PA, Samet JM, Schwartz BS, Linet MS, Zahm SH et al. Lead, genetic susceptibility, and risk of adult brain tumors. Cancer Epidemiol Biomarkers Prev 2006; 15: 2514–2520.
Rousseau MC, Parent ME, Nadon L, Latreille B, Siemiatycki J . Occupational exposure to lead compounds and risk of cancer among men: a population-based case-control study. Am J Epidemiol 2007; 166: 1005–1014.
van Bemmel DM, Boffetta P, Liao LM, Berndt SI, Menashe I, Yeager M et al. Comprehensive analysis of 5-aminolevulinic acid dehydrogenase (alad) variants and renal cell carcinoma risk among individuals exposed to lead. PLoS One 2011; 6: e20432.
van Wijngaarden E, Dosemeci M . Brain cancer mortality and potential occupational exposure to lead: findings from the National Longitudinal Mortality Study, 1979–1989. Int J Epidemiol 2006; 119: 36–44.
Zheng W, Chow W-H, Yang G, Jin F, Rothman N, Blair A et al. The Shanghai Women’s Health Study: rationale, study design, and baseline characteristics. Am J Epidemiol 2005; 162: 1123–1131.
Friesen MC, Coble JB, Lu W, Shu X-O, Ji B-T, Xue S et al. Combining a job exposure matrix with exposure measurements using an empirical Bayes framework to assess occupational exposure to benzene in a population cohort in Shanghai, China. Ann Occup Hyg 2012; 56: 80–91.
Gu XQ . Experience in controlling lead poisoning in the People’s Republic of China. Scand J Work Environ Health 1985; 11 (suppl 4): 16–19.
Wang YL . Industrial lead poisoning in China over the past 33 years. Ecotox Environ Safe 1984; 8: 526–530.
Ye X, Wong O . Lead exposure, lead poisoning, and lead regulatory standards in China, 1990–2005. Regul Toxicol Pharm 2006; 46: 157–162.
Lubin JH, Colt JS, Camann D, Davis S, Cerhan JR, Severson RK et al. Epidemiological evaluation of measurement data in the presence of detection limits. Environ Health Perspect 2004; 112: 1691–1696.
Robinson GK . That BLUP is a good thing: the estimation of random effects. Stat Sci 1991; 6: 15–51.
Casella G . An introduction to empirical Bayes data analysis. Am Stat 1985; 39: 83–87.
Kromhout H, Vermeulen R . Application of job-exposure matrices in studies of the general population: some clues to their performance. Eur Respir Rev 2001; 11: 80–90.
Rappaport SM . Assessment of long-term exposures to toxic substances in air. Ann Occup Hyg 1991; 35: 61–121.
Christiani DC . Occupational health in the People’s Republic of China. Am J Public Health 1984; 74: 58–64.
Tang HW, Liang YX, Hu XH, Yang HG . Alterations of monoamine metabolites and neurobehavioral function in lead-exposed workers. Biomed Envrion Science 1995; 8: 23–29.
Wang YL, Lu PK, Chen ZQ, Liang YX, Lu QM, Pan ZQ et al. Effects of occupational lead exposure. Scand J Work Environ Health 1985; 11 (suppl 4): 20–25.
Gustavsson P, Plato N, Hallqvist J, Hogstedt C, Lewne M, Reuterwall C et al. A population-based case-referent study of myocardial infarction and occupational exposure to motor exhaust, other combustion products, organic solvents, lead, and dynamite. Epidemiol 2001, 222–228.
Henn SA, Sussell AL, Li J, Shire JD, Alarcon WA, Tak S . Characterization of lead in US workplaces with data from OSHA’s Integrated Management Information System. Am J Ind Med 2011, 356–365.
Fischbein A, Hu H . Occupational and environmental exposure to lead. In: Rom WN, (ed). Environmental and Occupational Medicine, 4th edn. Lippincott Williams & Wilkins, Philadelphia, 2007 pp 955–990.
ILO. Women in labour markets: measuring progress and identifying challenges. Geneva, 2010.
Kennedy SM, Koehoorn M . Exposure assessment in epidemiology: dose gender matter? Am J Ind Med 2003; 44: 576–583.
Froines JR, Liu WCV, Hinds WC, Wegman DH . Effects of aerosol size on the blood lead distribution of industrial workers. Am J Ind Med 1986; 9: 227–237.
Tsai CJ, Shih TS, Sheu RN . Characteristics of lead aerosols in different work environments. Am Ind Hyg Assoc J 1997; 58: 650–656.
Vlaanderen J, Fransman W, Miller B, Burnstyn I, Heederik D, Hurley F et al. A graphical tool to evaluate temporal coverage of occupational history by exposure measurements. Occup Environ Med 2010; 67: 636–638.
This research was supported by the US National Institutes of Health (grant R37 CA70867) and the Intramural Research Program of the National Institutes of Health (contract N02 CP1101066).
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
Development of Quantitative Estimates of Wood Dust Exposure in a Canadian General Population Job-Exposure Matrix Based on Past Expert Assessments
Annals of Work Exposures and Health (2019)
Peaks, Means, and Determinants of Real-Time TVOC Exposures Associated with Cleaning and Disinfecting Tasks in Healthcare Settings
Annals of Work Exposures and Health (2019)
Use and Reliability of Exposure Assessment Methods in Occupational Case–Control Studies in the General Population: Past, Present, and Future
Annals of Work Exposures and Health (2018)
Annals of Work Exposures and Health (2017)
Evaluating Exposure–Response Associations for Non-Hodgkin Lymphoma with Varying Methods of Assigning Cumulative Benzene Exposure in the Shanghai Women’s Health Study
Annals of Work Exposures and Health (2017)