916 Joa. Retssland ct al. of all radiation-induced deaths will be among cx-workers. Alore significantly, we see that in table 3 we have a background of all-cancers of 2462 which seriously reduces the chance of detecting the cffect of radiation on the death rate. In our estimates of the time necessary for the survey, we optimistically work with table 2, that is assuming a 5% leaving rate. In the nuclear power industry at least, this is a realistic figure. Since the nuclear power industry is only 25 ycars old-—i.c. less than the latent period—we are not yet in the steady state situation. Hig. 3 shows the growth period of the effect of radiation, the numberofinduced deaths reaching the steady state valuc aftcr 30 years because of the use of Ayy. The ex-workcr numbers require 80 years to reach steady state. In table 4 we sec the accumulated number of radiation-induced deaths at 5-year intervals following the start of radiation work. We haveincluded in this the corresponding figures for a working population of 3000 and also the total number(in and »x) of workers at that time. So we sec that in the first 25 years we would expect considerably less than the 2 radiation-induced cancer deaths predicted in a 3000 work foree— table 4 shows 1 death but that is for A,y—taking a 50-year latent period (Aggy) we would expect 0-6 deaths duc to radiation in the first 25 years. Since we are interested in conservative estimates we shall restrict further discussion to the steady state situation. Table 5 shows how manyyears are necessary before a survey on our work force has a 50% chance of confirming a positive risk from radiation exposure to the 5 and 20% significance levels. It is clear from tables 2 and 3 that followup studies are essential and while table 5 gives emphasis to this it also clarifies the relative merits of looking at specific groups. ‘The most obvious deduction from table 5 is that the analysis of the survey data should be restricted to those workers and ex-workers below retirement age, The large background of natural cancer deaths above the age of 65 serves only to spoil the resolution. ‘The optimum on thesefigures is an analysis of 16-55 for which the time reeaured is ll years (20% significanec) and 44 years (5% significance); however, the influence of the assumedlatent period becomes important and 16-65 is probably safer. Comparing Aggy ancl By, in table 5 we sce only marginal differences overall and although C,, requires significantly shorter survey times for the lower age groups it is not a realistic form for the risk-time relation, In table 6 we show the magnitude that the risk cocflicient must be before it can be detected as positive against statistical flucluations, ‘The second part of table 6 shaws the range ofrisk cocfiicients compatible with the observed mean annual numberof radiation-induced deaths being those predicted in column If of table 2. Until the observation time is greatcr'than that in table 5, a negative risk is compatible with the observation and since we do not permit this possibility, we show the residual probability that the risk is not positive (expressed as a percentage). ee m2 Cau feb toad anes ‘Tables 5 and 6 taken together demonstrate the difficullics to be faced in analysing the results of a survey of the causes of death of occupationally exposed radiation workers.