An evaluation of different methods in epidemiologic studies of radiation risks

The probability of detecting possible late effects of ionizing radiation in human populations depends on the size of exposure in the study population and on the epidemiologic evaluation method used. This is not only due to the mathematical or statistical properties of the method, but also to the choice of the control population, which may be either external (usually large) or internal (usually not so large). Moreover, the use of a large external control population often results in an increasing influence of selection factors expressing themselves e.g. through the 'healthy worker' effect. Three different methods usually employed in follow-up studies are reviewed: the calculation of standardized mortality ratios (SMR), a contingency table analysis based on person-years (PY) and the comparison of cumulative mean doses (CMD) within different subcohorts. The methods are illustrated with a simple radiation risk model; nevertheless, most of the conclusions apply to non-radiation risk studies as well. The CMD method is shown to be heavily selection sensitive. (author)