HARVEY J. HILASKI AND CHAO LING WANG

Incidence rates of occupational disease, published each year by the Bureau of Labor Statistics, understate the total impact of the work environment on workers' health.' This is so because the statistics virtually exclude chronic types of illnesses, as well as illnesses having a long latent period whose relationship to the job often surfaces only after retirement or death.

Alternative methods of measurement confirm that an undercount exists, but differ concerning its magnitude. This article examines some of the alternative methods of estimating occupational diseases and suggests that a consensus on the adequacy and reliability of the estimates is not likely.

One of the first studies to highlight the scope of occupational disease in this country was a pilot study sponsored by the National Institute for Occupational Safety and Health (NIOSH). Confined to cross-sectional samples of workers in designated small industries in Oregon and Washington, the study was designed to determine the usefulness of a set of medical procedures for diagnosing occupational disease, and to ascertain how much new data on occupational illnesses would be generated by this method. The results of the study, published in 1975, underscored the issue of a large undercount in current occupational illness statistics, primarily those of BLS.'

Harvey J. Hilaski is an economist in the Office of Occupational Health and Safety Statistics, Bureau of Labor Statistics. Chao Ling Wang is a biostatistician in the same office. Mary Kay Rieg of the Review staff provided special editorial assistance.

When analysts compared counts of cases of "probable occupational disease" from the pilot study with those from employer logs maintained under regulations of the Occupational Safety and Health Administration (OSHA), and with employer workers' compensation records, they found that nearly 90 percent of the cases (approximately 400) uncovered in the pilot study were not listed in employers' files.' In effect, the findings indicated that the procedures normally followed by employers in recording and reporting occupational illnesses in fulfillment of legal requirements result in a gross underestimate of occupational disease in the United States.

It should not be surprising that most of the "probable occupational disease" cases found in the pilot study were not included in employer records. Consider, for example, that of the 346 cases discovered in the study, hearing loss was most prominent, amounting to about 28 percent. This condition usually has a gradual onset, with the result that the worker may be unaware of any defect in hearing unless he or she undergoes a hearing test such as that administered in the study. Moreover, hearing loss is often part of the aging process; without baseline data and subsequent periodic testing of the work environment and resulting effects on the worker, the occupational relationship can be seriously challenged.*

The NIOSH study results may be questioned in several aspects, including possible bias inherent in the special procedures used to assess the health status of surveyed workers. However, the design is a feasible method for

dealing with the detection of occupational diseases which are not readily apparent. Unlike the BUS annual survey. which measures the incidence of occupational disease, or the number of new cases recognized by employers as occupational in keeping with regulatory criteria, the NIOSH study measured the prevalence of occupational illness, or the level of occupational illnesses existing at a given time. For some purposes, prevalence is the more meaningful measure, in that it reflects the universe of persons with disease whose medical and economic needs may warrant special attention. However, high prevalence does not necessarily indicate high risk; instead, it may reflect an increase in survival, perhaps due to improved medical care. Conversely, low prevalence may reflect either a rapidly fatal process or cure of disease. The incidence measure is better adapted to keep a "running tab" on the health effects of workplace hazards, except for that component of occupational disease which is chronic in nature or of long latency.

Indirect estimates

Other studies and reports on occupational health problems have also pointed to an undercount in current illness estimates, but they are largely based on indirect evidence. The lack of reliable measures of occupational illnesses nationally has necessitated use and manipulation of surrogate data from epidemiologic or other studies to produce rather specific estimates of occupational diseases. On the basis of such data, a 1979 Labor Department report to Congress claimed that a conservative estimate of the prevalence of byssinosis among nearly 560,000 workers exposed to cotton dust at current levels was 83,610.5 In 1980, another congressional report by the Department (hereinafter referred to as the Interim Report) provided several estimates of respiratory disease prevalence or deaths from worker exposure to asbestos, silica, beryllium, cadmium, chromium, arsenic, nickel, coal tar products, and diisocyanates. Among other things, the Interim Report stated that, of 1 million workers currently exposed to silica, an estimated 59,000 will develop "some level" of silicosis; it also predicted 43,230 lung impairments as a result of exposure to diisocyanates, chemicals used to produce plastic products.

Much like the NIOSH study results, these estimates suggest a substantial undercount in the regularly published national statistics, which for 1980 showed 2,200 cases of all "dust diseases of the lungs."'' (The BLS survey does not ask employers to specify recognized occupational illnesses such as asbestosis, byssinosis, and silicosis and thus provides no direct estimates of the incidence of these diseases.) It seems worthwhile, therefore, to examine more closely the major methodologies and techniques commonly used in deriving indirect estimates of occupational disease."

Prevalence of disease. Prevalence rates were used to derive the estimate for silicosis cited above These rates express the proportion of the population having a disease at a given time, regardless of time of onset. But because epidemiological studies, the major source of prevalence estimates of occupational disease, are con fined to a particular population with a specific exposure, extrapolation of the findings to other, larger populations requires caution. Table 1. excerpted from the Interim Report, will be used to illustrate some of the limitations in the use of prevalence rates for this purpose.

The exposed population figures, obtained from a variety of sources, including the Mine Safety and Health Administration and NIOSH, are not representative exclusively of the exposed work force. The more important figures in this table are the prevalence rates, which were obtained from various special studies. Estimates of the numbers of disease cases were calculated by multiplying the exposed populations by the prevalence rates.

The crucial problem in applying prevalence rates derived from special studies to an entire production work force is the lack of assurance that the composition and levels of exposure of the larger population are the same as those of the worker group studied. The group of workers selected for epidemiological study is usually not a statistical sample of all production workers in the industry; therefore, the prevalence rates from such a study cannot be generalized to the larger population.

Use of the specialized prevalence rates in deriving the silicosis estimates probably led to biased results. For example, the 10-percent prevalence rate for the granite industry, which was adapted from an article published in the Archives of Environmental Health in 1974, was not a prevalence measure of silicosis or other respiratory disease, as would be expected. The source data related only to current dust exposures in the Vermont granite