Many people who have larynx cancer do not die of it. The five-year survival rate exceeds 50%. Hence, mortality is not a very sensitive outcome for larynx cancer, and incidence is to be preferred. The original mortality study had been conducted primarily to investigate lung cancer, which is usually fatal. Reports regarding larynx cancer and acid mists surfaced after the mortality study had been begun.
A cohort incidence study was one possibility. One difficulty would be determining who among this cohort got larynx cancer. There are no central cancer registries in the United States (there are a number of statewide registries), which would have made possible a rapid identification of cases. Instead, each individual (or their next of kin) in the cohort would have to be contacted.
Another question was the choice of a non exposed comparison group. There was no internal group of non exposed individuals who could be used. The principal difficulty in using any external group would lie in determining their laryngeal cancer rates. There are several statewide cancer incidence registries that go back in time, and there are some data regarding national cancer incidence based on a sample of the U.S. population.
Whether such a study would be worthwhile depends on two judgments, one mathematical and the other more subjective. The first question is whether the sample size of the study is large enough to answer the question posed: does exposure to acid mists cause larynx cancer? This involves a calculation of the "power" of the study to detect an excess risk of a specified size (the "power" to avoid a Type 11 error, the error of accepting the null hypothesis of no association when the alternative hypothesis is true). Let us assume from the prior studies that the postulated true relative risk for larynx cancer due to acid mist exposure is 2.0. A formula exists (Beaumont and Breslow, 1981) for calculating the power of a cohort study if the expected number of larynx cancers is known, and given a specified level of Type I error (the error of rejecting the null hypothesis when it is actually true, usually specificed as 5%). Generally, an 80% power is desired. Knowing the approximate number and age structure of the person-years at risk for this cohort from the already completed mortality study and knowing (at least approximately) the age-specific laryngeal cancer rates for a non exposed population (see below), it was possible to determine that approximately three or four laryngeal cancers were expected. Using the formula cited above, the investigators determined that in this cohort there was sufficient power to detect a twofold risk of larynx cancer. From the standpoint of sample size the study appeared worthwhile.
The question of whether the study is worthwhile must also be addressed from the standpoint of public health. Larynx cancer is rare. However, it is a serious disease, often fatal, and substantial numbers of workers are exposed to acid mists. Furthermore, Iow-level exposure to acid mists is of concern for the general public due to acid rain. Following the cohort might also throw light on the previously observed excess of lung cancer mortality, a far more common disease.
From a scientific standpoint, the association of acid mists and larynx cancer was also of interest in that it might further elucidate possible mechanisms of carcinogenesis (e.g., acidic irritation of epithelial cells, leading to mutation, or inhibition of ciliary clearance mechanisms in the respiratory tract, leading to increased presence of other carcinogens such as cigarette smoke).
References for Answer 1c:
Beaumont J, Breslow N: Power considerations in vinyl chloride studies. Am J Epidemiol 114:725-734, 1981.
The principal known causes of larynx cancer are smoking and drinking, with relative risks on the order of ten and two, respectively. These exposures could act as confounders, distorting the observed association between exposure and disease, if the exposed cohort and the non exposed comparison group differed in their smoking and drinking habits. It would be necessary to control for these confounders (particularly smoking) by collecting information about them for both the exposed and nonexposed cohorts, and then somehow taking this information into account in the analysis.