« ForrigeFortsett »
sheds. But lor estimating purposes, it was assumed that sand and gravel workers have the same "prevalence rate" as the granite workers, under further assumption of a similar degree and duration of exposure. In the case of the stone, clay, and glass products industry, a specific prevalence rate was selected for estimaling purposes, although special studies showed a range of rates. While the chosen rate - 6.16 percent-appears conser. vative, there is no reason to assume that it is a better estimate than any other within the wide 2- to 20-percent range of individual study results. If prevalence rates had been calculated from sample data, the precision of the estimates could be measured by standard er. rors. But for the data in table 1, the standard error of estimated prevalence is not available, and the precision of the estimates is therefore not known.
Dose-response. Dose-response studies generally attempt to establish a statistical relationship between dose (exposure) and response (onset of disease or death). For simplicity, this will be illustrated by example. Eight of 22 North Carolina textile manufacturing plants were selected for a 1970–71 study of cotton textile workers. ' A dose-response curve was fitted to the resulting data on the prevalence of byssinosis among a group of 1,259 workers in the cotton preparation and yarn area. (See table 2.) From the fitted curve, the byssinosis prevalence rates and their 95-percent confidence limits for workers in the cotton preparation and yarn area were predicted at various cotton dust levels of exposure:
representative of production workers in all yarn manofacturing industries, because the occupational composition of other industry segments would probably not be similar to that of the colton preparation and yarn arca.
Closer scrutiny of the dose-response relationship in the cotton preparation and yarn area makes the selection of the 25.8 prevalence rate even more puzzling, as only about 285, or 23 percent, of the 1,259 workers fell in the categories exposed to median dust levels 0.5 mg/m' or above. According to the data in table 2, the average overall byssinosis prevalence for the entire study was 15.5 per 100 workers, or 195 persons.
For the study of illnesses such as byssinosis, information on duration or years of exposure to the hazard is also crucial. This factor should be taken into account because a worker with 10 or 20 years of work exposure would seem to be more susceptible to byssinosis than a worker with a few years of exposure. Moreover, a worker's exposure level may change over the years, due to changes (not always for the worse) in working conditions, including ventilation, industrial hygiene practices, and so forth. In short, a worker normally experiences different amounts of exposure, of varied duration, over the course of his or her employment. If the dose-response relationship curve is not adjusted for the extent of exposure, its accuracy is diminished. Unfortunately, comprehensive data on the intensity, duration, and fluctation of exposure are rarely available, particularly in retrospective studies of the type used in making the above estimates.
In general, dose-response relationship curves are nonlinear, monotonic (increasing or decreasing), and have lower and upper asymptotes (usually, but not always, 0 and 100 percent). The models tend to operate well in a restricted range of exposure or dosage levels, but not over the entire range; that is, they may be useful in determining the "safe" level of exposure, but they are not suitable for developing national or industrywide statistics on occupational illnesses.
Dust level (mg/m') ... Predicted prevalence
per 100 workers 95-percent confidence
The referenced study suggested that a reasonably safe level of lint-free cotton dust is 0.1 mg/m' in the cotton preparation and yarn area, because nearly 94 percent of the workers exposed at this level had no symptoms of byssinosis.
As indicated in table 3, the authors of the Interim Report used the results of the North Carolina study to calculate the total number of expected byssinosis cases by multiplying 1977 BLS data for production workers in six yarn manufacturing industries by the prevalence rate of 25.8 per 100 workers - the prevalence predicted by the textile worker study for dust level exposure of 0.5 mg/m?. But it is unrealistic to assume that all workers in yarn industries generally are exposed to such high levels of cotton dust. This assumption might not even be true of cotton preparation and yarn workers nationwide, the types of workers among whom the prevalence study was done. Even if it were, however, the results of this special study should not be construed as necessarily
Standardized mortality ratio. The standardized mortality ratio has been widely used as a summary index of mortality in occupational epidemiologic studies. The ratio is a method commonly used to accomplish indirect age adjustment by applying age-specific death rates of a standard population to a study population to yield a MONTHLY LABOR REVIEW August 1982 .
Occupational Illness Estimates
Table 3. Expected numbers of byssinosis cases, selected yarn manufacturing industries
Assumed Industrial Exposed Yarn manufacturing
prevalence Classification worten
fin percent) Code
2211 2257 2281
1 000 34.400
256 256 258
Broad woven labnc mills
cotton Cucular in tabric mails Yamn somng malls Tentunang. throwing, west
ng, and winong malls Thread Mills Tre cord and lanc
2282 2284 2296
13.900 3.500 1,600
258 25 8 25 8
3 590 930 410
Sanct US Department of Labor, An hierm Report to Congress on Ocupational Dis. Bases, June 1980.table A-1.P. 126
number of “expected" deaths. It is defined as:
total observed deaths in a study population
total expected deaths in the population A ratio greater than I means that more deaths have been observed in the study group than would be expected on the basis of rates for the standard population; conversely, fewer deaths than expected are indicat. ed by a ratio less than 1. A test, such as chi-square, is generally used for determining the level of significance of the results.
If the focus is on mortality from hazards in the workplace, an ideal standard population would include all workers in the Nation. But because mortality data are not available in this detail, the total U.S. population, or the male population, is generally selected as the standard. Consequently, misunderstanding sometimes arises in applying standardized mortality ratios to estimate the total number of deaths caused by certain diseases in industry.
In the Interim Report, the ratio was used to derive estimates of work-related lung cancers." Data from three sources were used in the computations: (1) The number of workers exposed to beryllium com
pounds and oxides in end-user processes, estimated at 50,000, was obtained from the NIOSH 1972
National Occupational Hazard Survey. (2) The mortality ratio of 1.6 for lung cancer among
dardized mortality ratio of 1.6 does not mean that the mortality rate of the study population is 1.6 times that of the standard population and can thus be used as a multiplying factor to obtain the number of deaths in a broader population. Even when exposed to the same health hazards and having the same age-specific death rates for all age groups, different study populations will yield different values of the standardized mortality ratio if their age distributions differ. Obviously, then, the ratios should not be used in estimating the number of deaths due to disease in populations which differ in composition from a study population.
Second, a high ratio does not imply thai all deaths from lung cancers, for example, are caused by occupational exposure; it tells us only that the study population has an unusually high mortality risk. We do not know what percentage of deaths actually resulted from exposure, in this case to beryllium. Even if the standardized mortality ratio were interpreted to mean that the number of deaths of the study population is 1.6 times that of the standard population, 58 (that is, 93/1.6) deaths per year from lung cancer would have occurred in the study population irrespective of any exposure to beryllium. That leaves 35 deaths per year (or about 38 percent of total estimated deaths) which might be attributed to, or aggravated by, exposure to beryllium compounds. At most, the estimate of 93 deaths depicts the total cancer toll among the occupational group, not the excess cancer resulting from beryllium exposure.
beryllium workers was taken from epidemiologic
studies. (3) The 1976 U.S. age-adjusted incidence for lung
cancer of 116 per 100,000 males over age 20 was used.
Relative risk. Because incidence or other direct measures of occupational disease are generally lacking, epidemiologic study of occupational morbidity often relies on a measure of excess risk of a disease among workers in specified working environments to determine the association between certain occupational factors and the incidence of disease. One such measure is relative risk.
As we will see, relative risk also is subject to misuse in making estimates of occupational disease in industry, perhaps because of confusion about its definition. This can be illustrated again by an example from the Interim Report. From the National Occupational Hazard Survey, 98,090 workers were estimated to be exposed to chromates in chromate pigment production and, as noted, the 1976 U.S. male age-adjusted incidence rate of lung cancer for those over age 20 was 116 per 100,000. Based on these two figures and a chosen relative risk of s, the estimated number of lung cancers per year among the 98,090 workers was 570 cases: (exposed population) x (incidence rate of male population) x (relative risk)
(98,090) x (116/100,000) X 5 = 570 cases.
The number of expected deaths from lung cancer among workers in end-user industry processes was then calculated as: 50.000 X 1.6 X (116/100,000), or about 93 lung cancer deaths. Is this a valid estimate of all lung cancer deaths in this industrial population which were due to exposure to beryllium compounds?
First, it is important to understand that a stan
The report stated: “Based on three studies reporting relative risks from 2.3 to 38, a relative risk of S will be used for workers exposed to chromate compounds. ... "?
Seemingly, this is a conservative choice, but in reality there is no way to tell whether the relative risk of 5 is ap. propriate or not, because we have no information on its precision. More basic issues are whether relative risk should even be used to estimate ihe incidence of lung cancer cases, and what is involved if one does so.
Relative risk is a measure of the strength of the association of the disease with a certain factor, such as exposure to a specific chemical, and thus is an important statistical tool in retrospective epidemiological studies. It is defined as the ratio of the incidence rate of those exposed to a factor to that of those not exposed. Conversely, relative risk can be used. 10 compare groups of subjects diagnosed as having a disease to determine if the groups differ in the proportion of persons who had been exposed to the specific factor or factors. However, because retrospective study entails looking at the historical frequency of the suspected cause in a diseased group and a control group, the incidence rates of the diseased among the exposed and unexposed cannot be estimated directly but only approximated by relative risk, an odds ratio (or risk ratio).
Consider the following tabulation, in which the total T of the ith group of workers in a study population T (T = § T.). where K equals the total number of groups, is divided as:
lest is commonly used to determine whether the relative risk is significantly different from 1. Using a risk ratio from a particular study and an incidence rate for the general male population to estimate the extent of occupational disease in a larger population exposed to a specific factor assumes similarity among the age and sex distributions of all groups. However, in the Interim Report's estimate of disease from exposure to chromate compounds, the composition of the general male popu. lation and the exposed population may not be similar to that of any special study population. Therefore, use of the relative risk of 5, selected from a range, may produce biased results.
In brief, relative risk is a measure only of the strength of the association between the disease and the exposure factor. If significantly different from I, is indicates only that the disease is strongly associated with the exposure factor, not that the factor necessarily causes the disease. Any firmer conclusion would require further study. Rel. ative risk is surely a critical measure for assessing the etiologic role of a factor in disease, but it is not suitable for estimating the incidence of disease. Cancer related to occupational factors
What fraction of cancer incidence in this country is attributed to occupational exposure to carcinogens in the workplace? An unpublished 1978 report prepared jointly by several research institutes indicated that about 20 percent of all cancers are occupationally related and stated: “If recent evidence is considered and if the full consequences of occupational exposures in the present and recent past are taken into account, estimates of at least 20 percent ... may even be conservative."" The report concluded that earlier estimates that only 1 to 5 percent of all cancers in the United States were attributable to occupational factors had not been scientifically documented and that Dr. Philip Cole's 1977 estimates of less than 15 percent for men, and less than 5 percent for women, contained a large element of uncertainty." The results from the joint report have been cited in numerous publications, and questions have been raised concerning their validity.
The 20-percent overall estimate resulted from a twostep merger of the results of several separate studies. The first step developed estimates of the fraction of cancers due to asbestos exposure, while the second compared the risks from asbestos exposure with those from five other high-exposure substances, with the final result based on that comparison. Details of the estimation procedure follow.
According to the report, about 8 to 10 million workers have been exposed to asbestos since the beginning of World War II, and approximately 4 million have had heavy exposure. On the basis of a longitudinal study of a cohort of insulation and shipyard workers, the report
Data classified in the table may be obtained from prospective, cross-sectional, or retrospective studies. According to the tabulation above, the proportion of workers exposed to a factor and having the disease is A/(A + B;), while the corresponding proportion of unexposed workers with the disease is C/(C: + D.). Thus, the relative risk of disease for exposed workers is: A (C: + D,)IC (A + B). But because the incidence of a specific disease in a population tends to be low, the calculation (A,D)/(B,C.)—that is, (A/B) (C/D.) – provides a close approxi tion of the relative risk (but not of the incidence of disease) for the ith group of individuals. For an overall relative risk, a commonly used formula is:13
R = Ž (A,D.IT)! Ž (BCIT)
Because age is an important factor affecting incidence of disease, it should be accounted for in measurements of overall relative risk. In such an age-adjusted risk ratio. T, is the total of the ith age group. The chi-square
MONTHLY LABOR REVIEW August 1982 •
Occupulionul Illness Estimates
indicated that, of deaths of heavily exposed workers, 20 10 25 percent were from lung cancer, 7 to 10 percent from pleural or peritoncal mesothelioma, and 8 10 9 percent from gastrointestinal cancer. Summed up, this suggests that 35 10 44 percent of cohort deaths were altributable to cancer diseases. Accordingly, the joint study group concluded that, over the nexe 30 years, at least 1.6 million (about 40 percent of the 4 million) heavily exposed workers would die from the asbestos-re. lated cancers listed above. Based on an assumption suggested by data from a second source, the excess risk to the remaining less heavily exposed workers (4 to 7 million) was estimated to be one-fourth of that for the heavily exposed (a 10-percent risk, obtained as % x 40 percent), yielding a cancer estimate for this group of 0.4 to 0.7 million. This raises the total to between 2.0 and 2.3 million cancer deaths over the next three decades, with expected averages of 58,000 to 75,000 deaths per year associated with asbestos alone. The joint study indicated that this excess number of cancer deaths would account for roughly 13 to 18 percent of all expected cancer deaths.
In the second step, the study presented data on carcinogenic risks to workers found to be exposed to five substances in addition to asbestos during a 1972-74 National Occupational Hazard Survey." Table 4, adapt. ed from the study, shows selected results. The risk ratios were either standard mortality ratios or risk ratios selected from a range of values obtained from other epidemiological studies. The report indicated that the figures were not precise estimates, but reasonable ones for comparison purposes, because they were all derived by the same method.
Other conclusions pertaining to the second step of the joint study:
• Excess cases for the other five substances combined
are about 33,000 cancers per year, versus 13,900 for asbestos alone (table 4). The data show that these five agents together pose hazards similar to or greater than those posed by asbestos. The projected numbers of excess cancers are only for the 1972–74 groups of N-size workers. But because of workplace turnover, the actual number of workers ex
posed over time will be several times larger than N. • Consequently, the excess number of cases from asbes
tos (13,900) among the 1972–74 group underestimates the annual expected number of cancer deaths related to asbestos — 58,000 to 75,000, as derived in
step 1 - by a factor of 4 to 5. • Because the data for the five other substances listed
in table 4 were derived in the same way as those for asbestos, the estimates may likewise underestimate the number of cancers attributable to these substances.
SOLACE National Cancer Institute, National Institute of Environmental Health Services National Institute for Occupational Salety and Health Estimates of the Fraction of Cancer in me Unted States Related 10 Ocaupational Factors. September 1978, (unpublished). table 2,
• According to results from the first study step, asbes
tos alone will account for between 13 and 18 percent of all cancer deaths over the next 30 years. The data for the five other substances suggest at least 10 to 20 percent additional cancer deaths. Hence the study conclusion that occupationally related cancers may make up 20 percent or more of cancer deaths in forthcoming decades.
Closer examination of the joint study findings indicates that they may not be fully supported by data from the various studies used in their development, a conclusion corroborated in a report prepared for the Office of Technology Assessment, U.S. Congress."
First, a study of causes of nearly 2,300 deaths among a cohort of 17,800 asbestos insulation workers contributed the finding that 35 to 44 percent of workers heavi. ly exposed to asbestos died of cancer. The joint study group selected 40 percent as an approximation of the cohort's fatal cancer risk, and applied this percentage to a national population of 4 million workers considered to be heavily exposed to asbestos. This extrapolation was based on the unstated and probably unjustified assumption that the cohort of asbestos insulation workers was representative of the worker population of mixed industries. The mixed industry population might well differ from the cohort group not only in levels and length of asbestos exposure, but also in terms of population factors, such as age, sex, race, and percentage of smokers, which play significant roles in risk assessment. From the statistical point of view, application of the results from a study population to other populations of different composition and exposure experience usually produces biased estimates. The assumption that the risk of cancers for the less heavily exposed group of 4 to 7 million workers is one-fourth that of the heavily exposed group is similarly questionable. Consequently, the re
excess lung cancer deaths due to asbestos exposure alone (10,400) amount to 85 percent of all lung cancer deaths in this 1.6 million exposed population — that is, 10,400/(1.856 + 10,400). Add to this calculation the niesothelioma cases, and the figure becomes 88 percent. a rather astounding share considering all other cancer
sulting total estimate of 58.000 to 75.000 asbestos-related cancer deaths per year is highly suspect.
Second, the analytic results by "causes" of death in the cohort group of asbestos insulation workers showed only the percentage distribution by disease, irrespective of cause; they did not indicate what fraction of cancer was induced or aggravated by asbestos alone or by any other specific exposure. Although it may be highly abnormal to find 20 to 25 percent of lung cancers and 8 to 9 percent of gastrointestinal cancers as causes of death among a group of workers, the actual percentage of cohort-group deaths specifically associated with exposure to asbestos remains uncertain.
Third, the method of estimating excess cancer cases or deaths for each exposure substance shown in table 4 may result in either overestimates or underestimates. Risk ratios do not in themselves provide the magnitude of risk, because they are greatly affected by the composition of the study population. In that the risk ratio is a ratio of observed to expected disease cases or deaths, any small increase in observed cases will greatly increase the risk ratio if the expected number of cases is small. This is especially true when the study population is small. Assume, for example, that study results indicate that 1,400 deaths were observed when only 1,000 were expected, yielding a risk ratio of 1.4. The chisquare value for the level of significance will be: X? = (observed-expected)?/expected=160. To achieve the same level of significance for an expected number of 10, the observed number would have to be 50, and the resulting risk ratio would be 5. Thus, a small study population has a better chance to yield a large risk ratio than a large study population, if both experience the same hazards and have the same population composition. It is not surprising, therefore, to find that risk ratios vary widely among epidemiologic studies of workers exposed to the same chemical substance. It is more important to know whether the value of the risk ratio is significantly different from I, which indicates that the risks for the exposed and nonexposed groups are not identical, than to determine the absolute magnitude of the ratio itself.
Fourth, the joint report considered 13,900 excess cases per year due to asbestos exposure to be underestimated by a factor of 4 to S under the assumption of high turnover of the work force. Because the report had already estimated 58,000 to 75,000 asbestos-related deaths per year in an exposed population estimated at 8 to 11 million, further inflation by this factor results in an incredible number of occupational cancers. Based on the age-adjusted incidence shown in table 4, the number of expected lung cancer deaths, excluding mesothelioma, among a 1.6 million male population over age 20 would be 1.6 million x (116/100.000) or 1,856, in the absence of any exposure to asbestos. It follows, then, that
Estimates of the excess number of predominantly respiratory type cancer cases due to the five other substances for the given groups of exposed workers also are indicated in table 4. For these groups, the expected numbers of cancer cases in the absence of exposure would be: 1.970 (arsenic), 360 (benzene). 1,970 (chromium), 1,830 (nickel), and 4,520 (petroleum products), by direct application of the corresponding incidence rate. This means that the proportion of total cancer cases associated with each of these five chemicals would amount to 79 percent, 80 percent. 90 percent, 80 percent, and 67 percent, respectively. These proportions appear unreasonably high.
Finally, the joint study group considered the excess number of cancers attributed to the five other substances to be underestimated also, based on the twostep findings for asbestos and applying the same logic. However, this inference is not justifiable, because estimates for the other substances were obtained independently, and the magnitudes of the estimates were greatly affected by the value of the risk ratio chosen for each. Disability-impairment data
Three types of disability data are available to the occupational health analyst: Social Security Administration data from the various surveys of disabled adults; the social Security Administration's disability applicant files; and the National Center for Health Statistics' Health Interview Survey data. These data sets also suggest that there are greater numbers of health problems with some occupational connection than published data from employer-based surveys would indicate, but they do not necessarily establish a causal connection between work and disease. Therefore, it may be beneficial to discuss briefly the conceptual framework of these data bases.
The disability study. This study was an analysis of disability data obtained from an interview survey of 18.000 persons age 20 to 64, who had been selected from the 1970 5-percent census sample." The survey was conducted by the Bureau of the Census for the Social Secu. rity Administration. Of the 18.000-member sample, 11.700 had been selected from among those persons who had a disability prior to October 1969 as indicated on the 1970 census questionnaire. A mail screening in 1971 resulted in selection of another 1.200 recent onset cases and 5,100 nondisabled persons. Disability in the study was defined "severe," if it precluded work alto