Comments on a paper by Irwin D.J. Bross and Deborah L. Driscoll: "Direct estimates of low-level radiation risks of lung cancer. . . ." Yale J Biol Med 54: 317-328, 1981.

The authors use a statistic on page 320 which assumes zero variance for an estimated Poisson mean E. The estimate of E is based on national age-sex-specific rates for white males. The sample variance of this estimate is negligible, but possible demographic and geographic fluctuations in the lung cancer rate are not reflected in this sample variance. Increased variance of E would lower the value of the statistic difference/S.E. difference. This variability of E has been recognized and a discussion of the possible biases inherent in the selection of a control population to estimate E can be found in a paper by McMichael et al. [1]. The same paper also discusses the variability of lung cancer rates among various states in the U.S. and emphasizes the necessity to control covariables which are associated with lung cancer risk. The authors should therefore comment on smoking habits and exposure to other toxicants (e.g., asbestos or chemicals) which could influence the lung cancer rate in exposed shipyard workers. The selection of "only those cases with a latent period of over 15 years" is odd and not adequately supported or explained. Such an arbitrary selection of cases may underestimate or overestimate possible radiation risks. An explanation why "the full CDC/NIOSH" dosage x latency "table" was not used seems therefore necessary. Simple division by person-years to obtain an estimated lung cancer rate assumes implicitly that the lung cancer rate is constant in time [2]. This is highly unlikely but verification would be possible with the data in the full table. Lung cancer rates estimated from the full table would be more comparable with the BEIR estimates. The estimated rate given in the paper is a conditional rate since cases with latencies of less than 15 years are neglected.