American Journal of Industrial Medicine 14703-713 (1988)

Validity and Reliability of a Method for Retrospective Eva1 uation of C h lorop henate Exposure in the Lumber Industry

Clyde Hertzman, MD, MSc, FRCPC, Kay Teschke, MSC, Helen Dimich-Ward, MSC, and Aleck Ostry, MSC

This paper describes the validity and reliability of a method to retrospectively assess exposure to antisapstain agents used in sawmills (chlorophenates). The method is based on experienced workers’ estimates of exposure for each job title at the sawmill where they work. At a pilot mill, 10 randomly selected workers estimated the frequency and duration of exposures to chlorophenates for all 59 job titles. The reliability of their mean exposure estimates was very high, with an intraclass correlation coefficient for all raters of 0.91 (based on a calculated index of exposure). To assess validity, urinary chloro- phenate levels were measured for 86% of the workers at the mill during the summer and/or fall, and compared to experienced workers’ estimates of exposure. The correla- tion between workers’ exposure estimates and the urinary chlorophenate levels for each job title were consistently above 0.65 for all analyses and greater than 0.72 when summer and fall urine sample results were averaged. The evidence indicates that the validity and reliability of worker exposure estimates are high enough to justify inves- tigation of the method’s generalizability to other sawmills.

Key words: validation, reliability, epidemiology, chlorophenate, cohort study

INTRODUCTION

Several case control studies [Eriksson 1979; Eriksson et al., 1981; Hardell et al., 1980, 1982, 1983; Hardell and Sandstrom, 1979; Hardell and Bengtsson, 19831 have suggested an association between chlorophenate exposure and soft tissue sarcomas, non-Hodgkin’s lymphomas, and nasal cancers, whereas others [Pearce et al., 1985; Smith et al., 19811 have demonstrated no association. To help resolve this inconsistency, it was proposed that a cohort study of exposed workers in the British Columbia sawmill industry be conducted. B.C.’s sawmill workforce is one of the largest chlorophenate exposed cohorts in the world. However, despite use of the antisapstain agent dating back to the 1940s, virtually no exposure information has been recorded. Since estimation of dose would be a crucial preliminary step to a

Department of Health Care and Epidemiology, University of British Columbia, Vancouver, British Columbia. Address reprint requests to Clyde Hertzman, Department of Health Care and Epidemiology, Mather Building, 5804 Fairview Ave., University of B.C. Vancouver, B.C. V6T 1W5. Accepted for publication July 6, 1988.

0 1988 Alan R. Liss, Inc.

704 Hertzman et al.

cohort study, a method of retrospective reconstruction of exposure histories had to be developed.

Establishing exposures retrospectively is a long-standing problem in studying the etiology of diseases that have latent intervals. The available literature was recently reviewed by Kromhout and coworkers [ 19871, who identified studies using workers, occupational hygienists, plant supervisors, chemists, or occupational health teams as the exposure assessors. The objectives of these assessments varied from identifying sources of exposure [GCrin et al., 19851, to grouping tasks by the level of exposure to a variety of chemicals within a given workplace [Kromhout et al., 1987; Corn and Esmen, 19791. Our task was different. We needed to develop a method for assessing relative exposures to a particular agent using the only dose-related information that would be available from the sawmill records: the workers’ job titles.

This paper describes the development of a method of retrospective exposure assessment in a pilot sawmill, based on estimates made by experienced workers. Because no industrial hygienists were regularly employed over the long term in the large sawmills, only workers were in a position to make retrospective assessments. If this method is proven reliable, valid, and generalizable to other sawmills, it will be used to estimate exposure for the cohort study.

The development and evaluation of the method involved three phases. The first two phases examined the reliability and validity of worker estimates of current chlorophenate exposures, and are the subject of this paper. In the third phase (to be addressed in a future paper), experienced workers were asked to assess past exposure to chlorophenates in the same sawmill.

In phase 1, a random sample of senior workers currently employed in the mill was interviewed about exposures in all current job titles. Interobserver agreement between the workers’ exposure estimates was then evaluated as an index of reliability. Phase 2 involved measuring the urinary chlorophenate levels of all the workers in the mill and correlating them with the exposures estimated by the experienced workers in phase 1. This tested the validity (or accuracy) of the workers’ estimates of exposure for each job title. Since chlorophenates as used in sawmills are largely skin absorbed [Kauppinen and Lindroos, 1985; Braun et al., 1977; B.C. Research, 19861, and excreted in the urine [Wyllie et al., 19751, the urinary concentration of chlorophe- nates has commonly been used as a measure of dose, and was therefore chosen as the ‘‘gold standard” against which to compare worker estimates.

The purpose of this paper is to assess the ability of experienced workers to accurately and reliably estimate chlorophenate exposures in current sawmill jobs. If they are successfully able to do so, then workers familiar with past jobs might also be able to estimate exposures retrospectively.

METHODS Sawmill Selection

The selected sawmill has used chlorophenates since the early 1950s and maintains two methods of chlorophenate application: a diptank and a spray booth. It currently employs approximately 200 workers (all males) on three shifts. The greatest exposures to chlorophenates likely take place at the diptank and in the planermill where treated lumber is processed. However, some diptank-treated wood also returns to a part of the sawmill where resawing and chipping of treated lumber occur. There

Chlorophenate Exposure Evaluation in Lumber Industry 705

may also be exposures in the maintenance garage where lumber carriers used at the diptank are repaired, and in the squeezer area where the finished lumber is packaged.

Phase 1 : Interviews Twenty workers were randomly sampled from among those with more than 5

years seniority. Eight were excluded because they were unavailable for interview, had prior knowledge of the study, or had experience in only one area of the mill. The remaining 12 were interviewed in a controlled manner to ensure that there was no interaction with those still waiting to be interviewed. Two of the 12 did not respond to the complete list of job titles (10 and 43 job titles out of 59). Their responses were excluded from the analysis, leaving a total of 10.

The workers (hereafter known as raters) were asked to categorize the frequency of exposure for a typical individual in each job title as: daily, several times a week, several times a month, several times a year, or never. If the frequency of exposure was rated greater than “never,” they were then asked to estimate the duration of exposure on a typical day using the following categories: more than 7 hours per day, 5-7 hours per day, 3-5 hours per day, 1-3 hours per day, and up to 1 hour per day. For analysis purposes, each of these was expressed as a continuous variable. Estimates of frequency were converted to “days of exposure per year,” wherein “daily exposure” was rated as 250 days per year; “several times a week” was rated 150 days; “several times a month,” 36 days; “several times a year,” 3 days; and “never” was rated zero. Duration of exposure was rated zero if frequency was “never.” Thereafter, it was expressed as the midpoint of the stated range: .5 hours, 2 hours, 4 hours, 6 hours, 8 hours.

Phase 2: Urine Sampling Two sets of urine samples were collected from all workers in the sawmill: one

set in the summer over a 4-week period beginning in mid-August 1986, and the other in the fall beginning in mid-October. The object was to obtain repeat samples from workers at different times of the year representing common weather conditions in coastal B.C.: the drier warm weather of the summer, and the wet cool weather common to most of the other seasons. Samples were taken over a 4-week period to reduce the sample storage time prior to analysis, and to maximize the number of workers able to give samples.

Urine samples were collected on Thursdays to allow several recent working days of exposure. Grab samples were collected rather than the more complete 24-hour samples because of the greater likelihood of compliance. Since chlorophenates have an excretion half-life of 4 days to 10 weeks [Kalman and Horstman, 19831, daily fluctuations in exposure would tend to be averaged such that midweek grab samples should represent the average chlorophenate exposure over several weeks prior to sample collection.

All samples were collected in silanized wide-mouth amber glass jars to mini- mize the effects of photo-decomposition and adsorption to the sides of the container [Drummond et al., 19821. Extractions and analyses were based on methods developed for total penta- and tetrachlorophenates [Drummond et al., 1982; Workers’ Compen- sation Board of B.C., 19841, using acid hydrolysis, extraction with toluene, and quantification using a Varian 6000 gas chromatograph with an electron capture detector. Specific gravity was also measured to indicate if any samples were unusu-

706 Hertzman et al. 19

17

Number o f Job Titles

never several times several times several times daily exposed per year per month per week exposed

Fig. 1. Raters’ average estimate of frequency of exposure to chlorophenates.

ally dilute or concentrated. For certain excreted chemicals, grab samples corrected for specific gravity have been found to give a better correlation with 24-hour excretion. Because the exact mechanism of chlorophenate filtration in the kidney is not known, however, the uncorrected urine concentrations were used in the data analysis.

Each worker was asked to complete a self-administered questionnaire at the time of the urine sampling. Its main purposes were to detail which job(s) each worker had held prior to sampling, to record any time off, to allow the worker to give his own estimate of his chlorophenate exposure, and to determine what personal protective equipment he wore.

RESULTS AND DISCUSSION Phase 1 : Reliability

The 10 raters who gave complete information on all job titles had an average of 15 years of work experience at this mill (range 6-32 years) and an average of 18.2 years of experience in the sawmill industry. They had held an average of 7 jobs in multiple locations within the mill, with a range of 2 to 18 jobs.

Figure 1 shows the distribution of the raters’ average frequency and ratings for the 59 job titles under assessment. It demonstrates that only 4 jobs were unanimously considered to involve no exposure, whereas 19 jobs were rated as involving daily, or almost daily, exposure. Thirty-six jobs involved exposures ranging from several times per year to several times per week.

Figure 2 shows the estimated daily duration of exposure to chlorophenates by job title. Of the 55 jobs involving some exposure, 33 were estimated to involve 3 hours or less on those days when exposure occurred. For both the frequency and duration of exposure, the variation between raters’ estimates was least for those jobs at either end of the spectrum and greatest for those jobs in the middle.

Three measures were used to assess reliability: components of variance, intra- class correlation coefficient between individual raters, and intraclass correlation coefficient for all raters [Winer, 1971; Bartko, 19661. The components of variance approach indicates the proportion of overall variance within the data set that can be explained by a specific factor. Reliability is maximized as the proportion of variance explained by job title differences increases and that explained between raters de- creases. The residual term is unexplained variance and is a composite of rater/job title interaction and pure error.

The intraclass correlation coefficient between individual raters measures how an

Chlorophenate Exposure Evaluation in Lumber Industry 707

Number o f Job T i t l es

0 up to >I-3 hr >3-5 hrs. >5-7 hrs. >7 hrs 1 hour

Fig. 2. Raters’ average estimate of daily duration of exposure to chlorophenates.

individual rater’s estimate of frequency or duration of exposure compared with those of other raters. However, the purpose of this study was not to assess individual raters but rather to find out whether or not the group’s mean estimate of exposure was reliable. That is, had we selected many different groups of raters, would the group means of their estimates have been in agreement? The measure that best estimates this form of reliability is the intraclass correlation coefficient for all raters. In this calculation, the residual term includes between-rater variability as well as rater/job title interaction and pure error, in order to allow generalizability of the results to other samples of raters.

Table I shows the reliability analysis for frequency and duration of exposure as well as for a combined index of the two. In the case of frequency of exposure, 51% of the variance was explained by job titles, 6% by raters, and 43% remained in the residual. The intraclass correlation coefficient for a single rater is 0.54. According to Fleiss [ 19751, the intraclass correlation coefficient can be given the same numerical interpretation as its analog used for categorical data, the Kappa statistic. The levels of agreement have been graded [Landis and Koch, 19771 as excellent (0.75 or greater), good (0.40-0.75), and poor (less than 0.40). Thus the agreement among single raters for frequency of exposure may be considered good. The intraclass coefficient for all raters in the study is 0.92. This is excellent and suggests that had we selected several samples of 10 raters to estimate frequency of exposure by job title, each group would have given very similar mean ratings.

The components of variance estimates and intraclass correlation coefficients are very similar for duration of exposure to what they were for frequency of exposure. These two variables (frequency and duration) are highly correlated, partly because those raters responding “never” to the frequency of exposure question would automatically have a zero score for the duration of exposure variable. The combined index of exposure, which is the multiple of the frequency and duration estimates for each job title, represents an estimate of hours of exposure per year. The intraclass correlation coefficient between individual raters is approximately the same as in the previous two analyses, and the component of variance due to job titles is approxi- mately seven times that due to raters. The intraclass correlation coefficient for all raters is 0.91, similar to the separate analyses for frequency and duration.

Phase 2: Validity

In the summer, urine samples were collected from 150 out of 200 employees representing 53 of the 59 different job titles. In the fall, samples were collected from

708 Hertzman et al.

Table I. Reliability Analysis

Components of Variance (%)

Source of Variance

Frequency of Duration of Index of Frequency Exposure Exposure and Duration

Between job titles 51 49 51 Between raters 6 11 I Residual 43 40 42 Intraclass correlation .54 .55 .55

coefficient between individual raters

coefficient for all raters Intraclass correlation .92 .92 .91

154 employees with 51 different job titles. Most of the fall samples were from workers who also participated in the summer, which yielded a total of 132 repeat samples. The 22 fall samples from employees who had not participated in the summer meant that a total of 172 sawmill employees (86% of the workforce) gave at least one sample.

The total urinary penta and tetrachlorophenate levels found in the summer samples ranged from 19-1252 kg/l, with a mean of 239, and a median of 108 pg/I. The levels in the fall samples were significantly lower (p << O.OOl), ranging from 5-893 pg/l, with a mean of 111 pg/l and a median of 52 pg/l. In both cases, the chlorophenate levels were approximately log-normally distributed. In order to nor- malize the distribution, concentrations were expressed in logarithms (base 10) for the analyses.

Since the formulation used in this sawmill consisted mainly of tetrachloro- phenate, the urinary concentrations of that chemical were usually higher than pen- tachlorophenate, particularly when the total urinary chlorophenate concentration was high. However, the ratio of tetra to pentachlorophenate tended to decrease with decreasing total urinary chlorophenate concentration. This may reflect a more rapid excretion of tetrachlorophenate, as was found previously [Kalman and Horstman, 19831.

Pearson correlation coefficients were calculated to test the association between the raters’ exposure estimates from phase 1 and the logarithm of the measured urinary chlorophenate levels from phase 2. Three estimates of exposure were taken from phase 1: frequency, duration, and the index combining the two. In all cases, the average of the 10 rater’s estimates was used.

The phase 1 exposure estimates were applied to the job titles held by each volunteer who gave a urine sample, creating a personal exposure estimate for each worker. First, the job title listed by the worker as his “usual” job was assigned a rating for frequency, duration, and index of exposure. This would be the only job title that would likely be available from sawmill records, and was therefore the crucial one for the purposes of the planned cohort study. However, it might not cover the activities actually done over the short-term.

Table I1 lists the Pearson correlation coefficients for the association between the urine data for both summer and fall and the raters’ exposure estimates for the usual job title. For both the summer and fall, the correlation coefficients for frequency, duration, and the index of exposure were similar. For the index of exposure, they were

Chlorophenate Exposure Evaluation in Lumber Industry 709

TABLE 11. Pearson Correlation Showing the Association Between Loglo of Urinary Chlorophenate Levels and Three Types of Exposure Estimates for “Usual Job”*

Freauencv Duration Index of ExDosure

Usual jobsummer (n = 150) .666 ,678 ,687 Usual iob-fall (n = 154) ,677 ,670 ,668 ~

*p < 0.001 for all correlations.

0.687 (47% of variance explained) and 0.688 (45% of variance explained), respec- tively. These similarities persisted across all analyses, so the remainder of the results are presented for the index only.

In addition to his usual job, each worker listed all the “actual” jobs he worked at during the 8 weeks prior to urine sample collection. In order to determine which period of exposure before sampling best correlated with urinary chlorophenate levels, phase 1 exposure estimates for actual jobs were averaged in several ways: the jobs held in the week of urine sampling; those held in the week prior to sampling; those held in the 2 weeks prior to that; those held in the 4 weeks before that; and those held over the entire 8 weeks prior to sampling.

The left side of Table I11 shows the correlation between the summer and fall urine data and the index of exposure for actual jobs held over the various time periods prior to urine sampling. Coefficients for the usual job titles are given for comparison. Correlations were slightly higher for the summer data, at which time there was a wider range of urinary chlorophenate levels than in the fall. Also, the exposure estimates for actual jobs correlate slightly better with the urine values than do those for usual job titles. This observation tends to support the validity of the rater estimates, since an estimate for the actual job a worker held should be more closely associated with urinary chlorophenate levels than an estimate for a usual job that a worker did not consistently perform during the period before sampling. The differ- ence between the correlations for actual and usual jobs appears to be greater in the summer than in the fall. This may be due to increased job rotation in summer to provide vacation coverage.

Correlation coefficients of 0.758 and 0.716 were achieved for the summer and fall, respectively, based on the average index of exposure for the previous 8 week’s jobs. Thus for the actual jobs, the worker estimates predicted between 51% and 57% of the variance in the urine values.

In phase 1, the experienced workers were asked to estimate chlorophenate exposure over the previous year for each job title. It was anticipated that these ratings would correlate more highly with the average of summer and fall urine samples than with either round of urine samples individually. To test this idea, the data for the 132 workers who gave urine samples in both the summer and fall were used. The two urine values for each worker were averaged, then Pearson correlation coefficients were calculated for the association between the logarithm of the average urine value and the index of exposure estimates for usual and actual jobs. The right side of Table I11 shows that, as predicted, the correlations for the combined summer and fall data were higher than those for either the summer or fall separately. The percentage of variance explained increased to 53% for usual jobs, and to between 61% and 64% for the actual jobs held prior to urine sampling.

The workers who gave urine samples also estimated their own exposures. The

710 Hertzman et al.

Table 111. Pearson Correlation Showing the Association Between Loglo of Urinary Chlorophenate Levels and Exposure Estimates*

Summer Fall Average of Summer Fall Actual jobs held during: (n = 150) (n = 154) Summer and Fall (n = 132)” (n = 132)a

Week of sampling ,779 ,686 .788 ,780 ,684 Week prior ,731 ,728 ,780 ,729 ,761 Weeks 3-4 .735 .695 .768 ,734 ,714 Weeks 5-8 .770 .692 .797 ,769 ,717 All 8 weeks ,758 ,716 .786 ,759 ,717 Usual job ,687 ,668 .726 ,684 ,696

*p < 0.001 for all correlations.

“Eighteen individuals who gave a summer sample did not give a fall sample; 22 individuals gave a fall sample but not a summer sample. This explains the differences in correlation between the values in the respective pairs of columns by season.

correlations of their estimates with those of the raters’ from phase 1 were 0.745 for summer jobs and 0.748 for fall jobs. To see if they were more accurate than those produced by the raters in phase 1, correlation coefficients were derived for the association between the urine data and the volunteers’ self-reported exposures. For the summer data, the correlation coefficient was 0.667 and for the fall, it was 0.575. The workers’ correlations are lower than any of the comparable ones achieved by the phase 1 raters as shown in Table 111. The fact that the phase 1 raters estimated exposures for all 59 jobs titles in a series, whereas each volunteer rated his own without considering its comparison to others, may partially explain the greater success of the phase 1 raters. Another explanation might be that averaging 10 raters’ assessments reduced the influence of individual erroneous judgements.

Finally, a multiple regression analysis examined whether any other predictor variables could explain the remaining variance in the urine samples. Table IV presents both summer and fall data, showing the marginal contribution of each variable to the total correlation and to the total variance explained. Only one added a statistically significant increase (‘ ‘p to enter” < 0.05) in the variance explained: self-reported exposure, for the summer only. Although the use of personal protective equipment like respirators, gloves, and aprons was expected to explain a large proportion of the variance, the small effect of this variable may be due to the fact that workers doing the same job tended to wear similar equipment. The recent history of days off added only 2.2-2.9% to the explained variance. Domestic use of chlorophenates produced the least increase (0.5-2.1%), probably because few workers reported using the chemicals at home.

The fact that a large proportion of the variance in urine values remains unexplained by the raters’ estimates is partly due to differences in exposure between individuals with the same job title. It might be expected that some of these differences would average out over time as each individual performed all the duties of that job title. If this were true, one would expect to see continuing improvement in the correlation between the raters’ estimates and urine values as more urine samples were taken and averaged over time. However, there would still be within job-title variance that could not be predicted by any model that uses job title as the basic exposure unit.

Other potential sources of error may be found in the validity and reproducibility

Chlorophenate Exposure Evaluation in Lumber Industry

TABLE IV. Additional Variance in Urinary Chlorophenate Levels Explained bv Variables Derived From the Questionnaire Data

711

SUMMER DATA FALL DATA R R2 AR2 R R2 AR2

Raters’ exposure estimate ,687 ,472 - ,668 ,447 - for usual job title

plus self-reported exposure ,751 ,565 ,093 ,693 ,480 .033 plus personal protective .783 ,613 ,048 ,712 ,506 .026

plus days off ,798 ,638 ,025 ,729 ,532 .026 plus use of chloro- ,801 .642 ,004 ,730 ,532 .OOO

equipment and hygiene

uhenates at home

of the urinary chlorophenol measurements. In one between-laboratory evaluation of the analytical method conducted by us, results agreed within lo%, 70% of the time. No between-laboratory bias was detected. Nonetheless, random measurement error appears as unexplained variance in Table IV. In addition, although chlorophenol is reported as being 80% excreted through urine [Wyllie et al., 19751, individual metabolic variability will also contribute to unexplained variance.

CONCLUSIONS Phases 1 and 2 have demonstrated that in one sawmill, groups of experienced

workers are capable of giving estimates of current exposures that explain more than 50% of the exposure variance with excellent reliability. It is possible that further gains in accuracy might occur if “intensity” of exposure could be assessed by the exposure raters. Whereas this variable was contemplated for inclusion in the questionnaire, it was felt that intensity would be correlated with frequency and duration, highly variable on a day-to-day basis, and, perhaps, more abstract than the latter two variables. The fact that frequency and duration were so highly correlated provides indirect evidence to support its exclusion.

Phases 1 and 2 will be repeated at two additional sawmills in order to assess the generalizability of these results. The reliability of raters’ estimates of past exposures will be tested in phase 3. If phases 1 and 2 prove generalizable, and if phase 3 shows that exposures can be reliably estimated retrospectively, then the methodologic groundwork will have been laid for the cohort study.

When considering the cohort study, the most important test of our method of exposure estimation is its potential for minimizing misclassification errors by job title. Random misclassification will lead to a conservative bias in an epidemiologic study since evidence of a genuine dose-response relationship will be obscured, when in fact it exists. In Table V, a correlation analysis for the workers’ usual job has been converted into a 3 X 3 classification table. The logarithm of the average summer and fall urinary chlorophenate concentrations are compared to the exposure index for the usual job for those 132 workers providing urine samples in both summer and fall (from Table 111). Each variable has been divided into units of high, medium, and low exposure at the points that would minimize potential misclassification. The Table demonstrates that 100 of 132 points have concordant job scores and urinary chloro- phenate levels. Two points are ranked high by urinary chlorophenate level but low by job score. None are ranked in the opposite way. The other 30 points are all medium-high or medium-low combinations, that is, misclassified by one class.

712 Hertzman et al.

Table V. Potential for Misclassification of Exposure Levels by Replacing Urinary Chlorophenate Levels With the Exposure Index

Loglo of average summer and fall urinarv chloroDhenate level

Low Medium High

1.332-1.655 1.656-2.200 2.201-2.945

low

0-294 30 6 2 Average of summer medium

and 295-1093 6 36 11 exposure index for high usual job

1094-2000 0 7 34

It is not proposed that the exposure index, which is continuous, be converted to a categorical variable for analysis of the cohort study. However, the categorical style of presentation is a simple representation of the potential impact of misclassification on a dose-response relationship. The evidence presented in Table V demonstrates the degree to which raters’ estimates of current chlorophenate exposure are successful at minimizing misclassification. For example, if the low, medium, and high urinary chlorophenate groups represent valid exposure subgroups with relative risks of 1 .O, 3.0, and 5.0, respectively, then the raters’ estimates would result in subgroups with relative risks of 1 .OO, 2.60, and 3.53 in ascending order of index score. Thus, in this example, the dose-response relationship is highly sensitive to misclassification of those heavily exposed. Still, the monotonic increase in the underlying dose-response distribution is preserved and should be acceptable for the purposes of dose estimation in the cohort study.

REFERENCES

BC Research, Division of Environment and Health (July 1986): Analysis of chlorophenols in samples of urine, fabric patches, gloves, or air from occupationally exposed persons (sawmill workers). Phase 11.

Bartko JJ (1966): The intraclass correlation coefficient as a measure of reliability. Psycho1 Rep 19:3-11. Braun WH, Young JD, Blau GE, Gehring PJ (1977): The pharmacokinetics and metabolism of

Corn M, Esmen NA (1979): Workplace exposure zones for classification of employee exposures to

Drummond PB, Van Roosmalen PB, Kornicki M (1982): Determination of total pentachlorophenol in the

Eriksson M (1979): A case-control study of malignant mesenchymal soft-tissue tumors and exposure to

Eriksson M, Hardell L, Berg NO, Moller T, Axelson 0 (1981): Soft tissue sarcomas and exposure to

Fleiss JL (1975): Measuring agreement between two judges on the presence or absence of a trait.

G&n M, Siemiatycki J, Kemper H, Bkgin D (1985): Obtaining occupational exposure histories in

Hardell L, Bengtsson NO (1983): Epidemiological study of socioeconomic factors and clinical findings

pentachlorophenol in rats. Toxicol Appl Pharmacol 41:395-406.

physical and chemical agents. Am Ind Hyg Assoc J 40:47-57.

urine of workers. Int Arch Occup Environ Health 50:321-327.

chemical substances. Lakartidningen 76:3872-3875.

chemical substances: A case referent study. Brit J Ind Med 38:27-33.

Biometrics 31:651-69.

epidemiologic case-control studies. J Occup Med 27:420-425.

Chlorophenate Exposure Evaluation in Lumber Industry 713

in Hodgkin’s disease, and re-analysis of previous data regarding chemical exposure. Br J Cancer

Hardell L, Sandstrom A (1979): Case-control study: Soft-tissue sarcomas and exposure to phenoxyacetic

Hardell L, Axelson 0, Rappe C (1983): Nasal cancer and chlorophenols. The Lancet 8334:1167. Hardell L, Eriksson M, Lenner P (1980): A case control study: Malignant lymphoma and exposure to

chemical substances, particularly organic solvents, chlorophenols and phenoxy acids. Lakartid- ningen 77:208-210.

Hardell L, Johansson B, Axelson 0 (1982): Epidemiological study of nasal and nasopharyngeal cancer and their relation to phenoxy acid or chlorophenol exposure. Am J Ind Med 3:247-257.

Kalman DA (1986): University of Washington Environmental Laboratory, personal communication. Kalman DA, Horstman SW (1983): Persistence of tetrachlorophenol and pentachlorophenol in exposed

Kauppinen T, Lindroos L (1985): Chlorophenol exposure in sawmills. Am Ind Hyg Assoc J 46:34-38. Kromhout H, Oostendorp Y, Heederick D, Boly JSM (1987): Agreement between qualitative exposure

estimates and quantitative exposure measurements. Am J Indust Med 12:55 1-562. Landis JR, Koch GG (1977): The measurement of observer agreement for categorical data. Biometrics

33:159-174. Pearce NE, Smith AH, Fisher DO (1985): Malignant lymphoma and multiple myeloma linked with

agricultural occupations in a New Zealand cancer registry-based study. Am J Epidemiol 121:225- 237.

Smith AH, Pearce NE, Fisher DO, Giles HJ, Teague CA, Howard JK (1981): Soft tissue sarcoma and exposure to phenoxyherbicides and chlorophenols in New Zealand. JNCI 73: 11 1-1 17.

Winer BJ (1971): Statistical Principles in Experimental Design, 2nd ed. Montreal: McCraw-Hill. Workers’ Compensation Board of BC (October 10, 1984): Pentachlorophenol and tetrachlorophenol in

Wyllie JA, Gabica J, Benson WW, Yoder J (1975): Exposure and contamination of the air and employees

48:217-225.

acids or chlorophenols. Br J Cancer 39:711-717.

woodworkers. J Toxicol-Clin Toxicol 20:343-352.

urine. Laboratory Services Analytical Methods.

of a pentachlorophenol plant, Idaho 1972. Pesticides Monitoring J 9: 150-153.