An evaluation of several methods for assessing the effects of occupational exposure to radiation

Several methods for the analysis of occupational radiation exposure data, including procedures based on Cox's proportional hazards model, are presented and evaluated. Issues of interest include the contribution of an external control, the effective handling of highly skewed exposure data, and the potential for detecting effects in populations occupationally exposed to radiation. Expressions for evaluating the power of various procedures are derived and applied to data from the Hanford population in order to determine power curves for detecting leukemia effects, with both additive and multiplicative linear models being used. It is found that the introduction of an external control can increase power, although not when an overall adjustment factor must be estimated from the data or when death rates for the study population are substantially lower than those for the control population. It is also found that very little power is lost if exposures are grouped. Finally, the power calculations indicate, as expected, that in analyses of occupationally exposed populations, such as the Hanford workers, there is very little chance of detecting radiation effects at the levels of our current estimates. However, power is reasonably good for detecting effects that are 10 to 15 times larger.