Past occupational exposure to asbestos among men in France

among Objectives This study aimed at reconstructing changes in the frequency and levels of occupational asbestos exposure in France over the past century. Methods Work histories were collected during 11 population-based case-referent studies recently carried out in France, and an asbestos-specific job-exposure matrix including 10 625 jobs was used to estimate indices of past occupational asbestos exposure. The results were estimated from a sample of 4287 subjects, bootstrapped 200 times. Results The distribution of socioeconomic categories within the sample was compared with that of the general population in 1954, 1962, 1968, 1975, and 1982. The proportion of blue-collar workers was similar. The highest proportion of exposed subjects was found between 1950 and 1980. Around 10% of each 10-year age class was exposed to asbestos. For those born in 1930-1939,15.2% was exposed between the ages of 20 and 29 years. For each age class born in 1900-1939, the proportion exposed at least once by 60 years of age ranged from 18.2% to 24.5% and, of those exposed, the cumulative duration of exposure ranged from 11.3 to 15.4 years by the age of 60 years. A population exposure index showed that the heaviest exposure occurred between 1960 and 1970 and that the age classes born between 1920 and 1929 were the most heavily exposed. Time trends showed that the mean value of this index for the men aged 20-59 years reached a peak in the 1960s and then decreased. CO~CIIIS~O~S This study presents data of reasonable validity about occupational asbestos exposure in France and its trends over the past centuly; the data are being used to forecast the development of male mortality from mesothelioma in France.

frequency and levels of asbestos exposure (1 I), predictions indicate that, for Western Europe, the trend toward increased mesothelioma mortality will continue for at least 2 decades, because of the 30-to-40-year latency period (12).
It is thus important to know the distribution of occupational asbestos exposure in each country and how it has changed in recent decades. Accurate estimates of the population's exposure allows more realistic forecasts of changes in the incidence of asbestos-induced cancers than are possible with modeling based only on past mortality. This information can also facilitate the development and assessment of programs to monitor asbestos risk (1 1). Nonetheless, little has been published about asbestos exposure in general populations, and only partial data are available (13). In particular, case-referent studies about the relation of asbestos exposure and lung cancer or mesothelioma show extremely variable proportions of exposed referents, depending on the region studied (4); they do not allow an assessment of the prevalence of exposure among the general population.
The objective of this work was to describe the prevalence and levels of occupational asbestos exposure for the French male population and their changes over the century, according to the work histories of a representative sample of French men.

Subjects
We created a sample from several different populationbased case-referent studies, all of which included a complete work history obtained by a structured interview of the subject himself. Data were collected between 1984 and 1995. The subjects came from nearly all regions of France, from the most to the least industrialized. Their mean age at data collection was 59.2 years. The subjects included in this sample were the referents of all the studies. To increase the size of the sample, we also included cases, but only for studies involving disorders for which no connection with asbestos was known (bladder cancer, leukemia, and nephropathies), to avoid overestimation of exposure to asbestos. The data for each subject included a complete job history, with the starting and ending dates for each separate job or job episode, the occupation being coded according to the International Standard Classification of Occupations (ISCO) (14) and the industry according to the International Standard Industry Classification (ISIC) (15). The study was limited to men because of the paucity of complete work histories for women. We pooled data for 4700 subjects from 1 1 case-referent surveys (table 1).

Assessment of asbestos exposure
A job-exposure matrix (JEM) specific for asbestos and applicable to France was developed by Orlowski et a1 (26). Asbestos fiber types were not distinguished in the matrix. Its job axis combined ISCO and ISIC codes. This matrix provided the following information for each ISCO-ISIC combination: (i) an index of the probability of asbestos exposure (unexposed, <0.1,0.1-0.5,0.5-0.9, >0.9), (ii) an index of the frequency of exposure (<5%, 5-30%, 30-70%, >70% of work time), and (iii) an index of the intensity of exposure, expressed in fibers per milliliter (flml) (<0.1, 0.1-1, 1-10, >10 flml). When significant technical or protective changes occurred that were likely to modify exposure conditions, the corresponding date was included in the matrix, and the exposure indices were modified. Thus the matrix could contain several periods for the same ISCO-ISIC combination. This job-exposure matrix included 10 625 different ISCO-ISIC combinations, including those for different periods.
The work histories of the 4700 subjects included in our sample were linked with the job-exposure matrix. For 413 subjects, no job episode co~l-esponded to a job coded in the matrix. These subjects have been excluded, and the analysis thus concerned a sample of 4287 men, with 12 953 different job episodes.

Sample validation
We used French census data to verify the comparability of this sample with the French male population in terms of occupation. Age, gender, and occupational data were available from 5 different censuses, 1954, 1962, 1968, 1975, and 1982. For each census and by age class (15-24, 25-34, 35-44, 45-54, and 55-64 years), we compared the proportion of subjects in the sample and in the French male population with a blue-collar job. The occupational classification for the census used the French code of socioprofessional categories (27) in effect at each census, while the sample was coded only according to the 1968 ISCO. Although the detailed codes of these classifications are not completely compatible, it was easy to classify the occupations of the French socioprofessional categories corresponding to those of the ISCO code that began with 7, 8, or 9 (production and related workers, transport equipment operators, and laborers). The comparison concerned only these categories, which included the vast majority of asbestos-exposed occupations: in the sample, among the 4539 job episodes that were considered exposed to asbestos by the job-exposure matrix, 3748 (83%) fell into these ISCO categories. The comparison was carried out according to the defined age classes. Some of the censuses, however, used different definitions for the youngest age class (20-24 years in 1954 and 1982, 17-24 years in 1975), and we grouped the sample data in the same way as the census data for these years.

Statistical analysis
The variables we studied included the proportion of men occupationally exposed to asbestos and the duration of exposure. The exposure indices were calculated for specific periods and cumulatively. We also calculated a population exposure index (PEI), defined as the product of the mean intensity, in fibers per milliliter, the proportion of persons exposed, and the average duration of exposure. Only exposure occurring between the ages of 20 and 59 years was studied (except for cumulative exposure indices, for which exposure was taken into account from the first exposed job), because both the age of entry into the job market and the age of retirement have changed substantially during this century. These changes make age-class or interperiod comparisons for the extreme age classes difficult to perform. To study the changes in asbestos exposure over time, we estimated the principal indicators of exposure distribution by age class (period of birth) and by 10-year age groups; this procedure also allowed us to study the period effect.
Despite the large sample size, some exposure levels and age classes included an insufficient number of subjects for a reliable estimate of the distribution of the exposure variables. We therefore used the bootstrap resampling method (28). The initial sample was boot-strapped 200 times, and for each drawing, each job episode was randomly classified as exposed or not exposed according to the probability assigned by the job-exposure matrix to the corresponding ISCO-ISIC combination. The distribution of the asbestos-exposure variables was estimated from the 200 bootstrapped samples, thereby providing their confidence intervals.

Results
We first examined the validity of the sample as to its occupational distribution, compared with that of the French male population. The comparison of the proportion of sample subjects classified into principal groups 7, 8, and 9 of the ISCO (production and related workers, transport equipment operators and laborers) during the last 5 available censuses, by their age class at that time and the proportion of men in the French general population in the same age class is shown in table 2. No clear trend emerges from the analysis of this table. The chi-square tests performed on 24 cells (and not 25, for the sample did not include any subject older than 55 years in 1954) showed significant differences for 12 cells. When the test was significant, in 8 cases, the proportion of subjects in groups 7-9 of the ISCO was lower in the sample than in the general population, while the reverse was true for 4 cells. The global chi-square test, for all age classes in each census, was significant 3 times (1962, 1968, and 1975) and not significant twice (1954,1982). Accordingly no trend could be deciphered. Most important, the percentages were overall reasonably close in that, in 18 of 24 comparisons, the difference was less than 5%. Most of the cases with a spread greater than 5% (4 of 6) involved the extreme age classes (15-24 years and 55-64 years), which cover entry into and departure from worklife, periods during which it could be difficult to classify subjects in an occupation. It can thus be concluded that, overall, the composition of the sample was as close to that of the French male population as to the distribution of occupations likely to entail workplace asbestos exposure during the period covered by this study.
The sample comprised men questioned between 1984 and 1995. It is censored for the youngest age classes which had not completed worklife at the time of the interview. For these age classes, therefore, we did not have asbestos exposure data for the entire career.
The proportion of men exposed to asbestos in each age class (by period of birth) is shown by 10-year age groups in table 3. The most-exposed men were those aged 20-29 years, and the highest proportion of the male populationmore than 15% -was exposed during the 1960s and 1970s, a period of expansion for both the French economy and the use of asbestos. For each age class, table 4 presents the cumulative proportion of those exposed up to various ages. The age class with a period of birth between 1930-1939 was, overall, the most heavily exposed, with 24.5% of all men exposed at least once during their worklife. Comparing the cumulative proportions of those exposed up to the age of 40 years shows that the age classes which entered the workplace around 1950 have the highest proportion of exposed men. Table 5 presents the mean duration of asbestos exposure by age group among those exposed, by age class. The mean duration of exposure was highest around the 1960s, when the proportion of those exposed was also highest. Examining the cumulative duration by age class up to various ages (table 6) shows that, among those exposed, the men born before 1950 had the longest period of asbestos exposure. Table 7 presents the values of the population exposure index, which takes into account the mean level and duration of exposure and the proportion of exposed subjects, by both period of birth and age group. Table 8 reports the cumulative values of this index, by generation, up to various ages. Again, overall asbestos exposure was highest in the 1960s and 1970s, with a maximum for the men born between 1940-1949 at the ages of 20-29 Table 2. Proportion of male workers (ISCO groups 7, 8, and 9: production and related workers, transport equipment operators, and laborers, respectively) in the sample and in the censuses.     years (PEI 0.2). When the cumulative value of the population exposure index is considered up to the age of 60 years, it appears that the men born in 1920-1929 were the most heavily exposed (PEI 0.4). Only the generations born after 1950 show a substantial reduction in their exposure to asbestos. In other words, in France, it is the men aged 45 years or older today who have been the most massively exposed. The evolution of exposure to asbestos over time in France between 1930 and 1990 (earlier periods were not included, due to the small number of subjects in the sample) is summarized in table 9. The mean values of the percentage of men exposed at least once, of the duration of exposure among those exposed, and of the population exposure index were computed by 10-year periods for men aged 20-59 years. The total percentage of exposed men was highest between 1940 and 1960 (over 12%), and it decreased after this period. However, the mean duration of exposure to asbestos increased strongly from 1950 to 1980, explaining that the population exposure index reached its highest value between 1960 and 1970. This index decreased rapidly after 1970, due partly to the decrease of the prevalence of exposure, and mainly to the lowering of exposure levels.

Discussion
A sizeable sample of men and the use of a specific job-exposure matrix for asbestos have enabled the   reconstruction of occupational exposure to asbestos and its development in France since the beginning of the century. This reconstruction provides information, by age class (10-year periods of birth) and age group, about the principal characteristics of exposure associated with cancer risk: frequency, duration, and level. These data could be used to model the trends in mesothelioma mortality among French men over the next decades. However, various problems related to the validity of the estimates presented merit discussion.
The representativeness of the sample in relation to the overall French male population was analyzed in terms of occupations likely to entail asbestos exposure, and the sample used for this study was very close to the general population for the various periods that were considered. This similarity is probably explained by the fact that the sample was drawn from population-based case-referent studies in a variety of regions covering most of France. One might nonetheless have expected a selection bias since, for each age class, it was those who survived into the 1980s and 1990s who were interviewed. Because mortality is higher among workers than among more well-to-do classes, this "healthy survivor effect" should thus have tended to underestimate asbestos exposure in the general population. Nonetheless, since the mean age of the sample at interview was only 59.2 years, this bias was probably minimal. In particular, the comparison of the frequency of men in ISCO groups 7-9 in the sample and the general population showed no significant differences for the most recent census (1982).
Despite the size of this sample, only a few subjects represented the oldest and youngest age classes. This representation reduced the precision of the estimates. Using the bootstrap method showed, however, that the confidence intervals were fairly narrow and hence that the estimates of the exposure indices were relatively precise. Another problem is that the sample included the different age classes interviewed during the same period; it is accordingly censored for the youngest generations. Because we were concerned only with past exposure, however, this censoring did not affect our results.
The characteristics by which asbestos exposure has been assessed come from a job-exposure matrix. Hence, the validity of the estimates of past exposure and of their trends are closely linked to the accuracy of the job-exposure matrix and especially to its ability to describe the changes that have occurred over time. The job-exposure matrix was constructed using knowledge from different sources of data and expertise, mainly trends in imports of asbestos, modifications in technology and working processes in the various economic sectors, and changes in regulations about the protection of workers and asbestos use. The design of the matrix allowed changes that were likely to modify exposure conditions to be taken into account by changing the probability, the level, or the frequency of exposure. Cut-off points were chosen according to different criteria, and they were specific for each ISCO-ISIC combination. For instance, the job-exposure matrix contains 476 different entries associated with the ISIC code 5000 (construction), which is a key sector regarding asbestos exposure. They correspond to 272 different ISCO codes, 66 of them being divided into 2 different time periods, 7 into 3 periods, 18 into 4 periods, and l l into 5 or more periods. The cut-off points varied from 1930 to 1980, depending on the reason for changing the value of one or more exposure indices. This feature of the matrix allows for more-detailed exposure assessments over time for different occupations than what would have been derived only from gross asbestos import figures.
Variations in asbestos exposure are indeed linked to asbestos imports, since almost all the asbestos used in France was imported. How close the correlation is in time between the quantity of asbestos in a country (reflected by cumulative import figures) and the way the workers are exposed to asbestos is nevertheless not clear. Before 1937 (first year for which data are available), the quantity of asbestos imports, which had begun at the end of the last century, was probably low. Figure 1 shows the evolution from 1937 to 1994 of annual asbestos imports (Association Fran~aise de l'Amiante, personal communication). After a break during World War 11, imports increased steeply, reaching their peak in 1975, and then decreased, ending finally in 1996 when asbestos use was banned by the French government. Almost no information is available about the type of asbestos imported (data are known only for a small fraction of total imports and not for all years; see figure I), but chrysotile seems to have been the most common. Estimates of the exposure indices in this study are not directly comparable to the quantities of annual imports of asbestos in France. Table  9 shows that the mean duration and level of exposure were low during the 1930s when imports were probably not so important, but that prevalence was already high. This situation can be explained by the fact that the jobexposure matrix overestimated the frequency of exposed jobs in the remote past. However, mesothelioma mortality in France increased sharply from the 1950s on (29). Due to the latency period, this trend would be consistent with a significant exposure prevalence during the 1930s. Table 9 also shows that the mean population exposure index has been high since the 1940s, namely, at the beginning of the rise in asbestos imports. Again, this phenomenon could be due to an overestimation of exposure by the job-exposure matrix. Another explanation may be that, even if the annual quantity of imports was still low, the cumulated quantity of asbestos was already high. Moreover, exposure levels increased rapidly in the post-World War I1 period, due to the expansion of the French economy and to the absence of efficient protective regulations. Finally, yearly asbestos import figures do not correlate directly with trends in population exposure, since they do not take into account changes in some important determinants of the workers' exposure, and the main features of exposure among workers cannot be simply derived from import figures.
ij Total imports I! H Fiber type known It must also be considered that a job-exposure matrix is an imperfect tool that can result in individual misclassification (30). Various epidemiologic studies of the effects of asbestos exposure have nonetheless used this method with satisfactory results (26,31,32). The "epidemiologic performance" of this matrix can also be more formally assessed, by comparing its results with those obtained for the same subjects with a reference exposure assessment method (33). The epidemiologic performance of the job-exposure matrix used in this work was assessed by the comparison of 2 series of odds ratios. Both derive from the population-based case-referent study of mesothelioma recently carried out in France by Iwatsubo et a1 (9). This study included 460 cases and 421 referents. One series was obtained for asbestos exposure as assessed by the job-exposure matrix, and the other for exposure assessment by a group of industrial hygienists on a caseby-case basis (Iwatsubo Y, Orlowski E, personal communication). The comparison of the odds ratios for each of the methods showed that the values obtained with both were, on the whole, consistent. For example, the odds ratios associated with an estimated cumulative exposure of 0.5-0.99 fibers per milliliter per year were 4.7 [95% confidence interval (95% CI) 2.5-8.71 for the case-bycase expertise and 2.8 (95% CI 1.4-5.6) for the job-exposure matrix assessment. They were respectively 5.4 (95% CI 2.7-10.8) and 10.1 (95% CI 2.9-35.5) when the cumulative exposure was 2.5-4.9 fibers per milliliter per year, 8.5 (95% CI 2.3-30.8), and 3.0 (95% CI 0.7-12.1) when the cumulative exposure was 10-19.9 fibers per milliliter per year, and 10.2 (95% CI 4.8-21.9) and 14.0 (95% CI 3.2-62.1) when the cumulative exposure was over 20 fibers per milliliter per year. Thus the use of the job-exposure matrix for assessing exposure probably did not lead to any major distortion in the estimates of the population's exposure to asbestos.
Beyond the figures concerning the production, importation, or use of asbestos in the principal industries of various countries, little is cussently known about the distribution of asbestos exposure among various national populations. The proportion of Inen occupationally exposed to asbestos has been estimated to vary from 10% to 36% in Sweden and Norway, whereas 18% of Finnish men 35 years or older have been significantly exposed (13). These data are consistent with those presented in this report. Despite the imprecision inherent in the limitations of available data and tools, this study brings information of reasonable validity about occupational asbestos exposure in France and its trends over the past century. These data are currently being used to model the changes in male mortality from mesothelioma in France in the decades to come.