Data on lung cancer in radiation workers From Dr Irwin DJ Bross, Director ofBiostatistics and Ms Deborah L Driscoll Roswell Park Memorial Institute Buffalo, NY 14263, USA Sir, In'Epidemiological studies of radiation workers: preliminary communication'(June Journal, p 450), Dr Schofield cites the Mancuso study of Hanford workers in the state of Washington, but omits reference to the National Institute of Occupational Safety and Health (NIOSH) study of nuclear submarine workers at the Portsmouth Naval Shipyard. Particularly with respect to lung cancer risks, the NIOSH study has some parameters which are quite similar to the British Nuclear Fuels Limited (BNFL) study. The NIOSH final report on the retrospective cohort study (Rinsky et al. 1981) made a number of serious analytic mistakes (Bross & Driscoll 1981) which, if noted here, may enable the BNFL study to avoid falling into like errors. In Table 3 of the BNFL study (Schofield 1982), the ratio of'observed'(0) to'expected'(E) deaths for the lung cancer series is 0.77, somewhat less than the NIOSH value. NIOSH erroneously concluded that because O/E was not markedly larger than unity there was no lung cancereffect. However, as the BNFL report correctly notes (and as was true in the NIOSH study), there is a strong'healthy worker'effect.(This is usually the case with these highly selected'elite'nuclear workers.) In addition to the healthy worker effect, there are two otherfactors in the NIOSH (and BNFL) study which tend to obscure a lung cancer effect. One is the latent period: only after 15 years is it likely that the effect would be detectable. The other is that many workers have relatively low exposures. In the NIOSHreport (Rinsky et al. 1981) only about 5% of the person-years involved individuals with appreciable exposures and more than 15 years' latency.

Clear lung cancer effects appeared in the NIOSH data when exposure and latency were taken into account (Bross & Driscoll 1981). There is some indication in the preliminary data that the BNFL study will find similar effects if the statistical dosage-response analyses avoid the errors ofthe NIOSH report. Thus, it would be anticipated that thoseworkers who are still'serving'will be less likely to have had 15 years follow up than the'pensioners'. While this is not a direct measure of latency, it should be correlated with latency. Using it as a surrogate for latency tends to diminish, not increase, the lung cancer effects.