More on direct estimates of low-level radiation risks.

In an epidemiologic study of mortality at the Portsmouth Naval Shipyard (PNS), the National Institute for Occupational Safety and Health (NIOSH) found no evidence of excess deaths due to leukemia or other cancers among workers exposed to low levels of ionizing radiation [1]. In a subsequent analysis, Bross and Driscoll [2] identified excess lung cancer mortality in PNS workers with lifetime radiation dose of 1 rem or greater and with more than 15 years' latency since first radiation exposure. Although that observation may be important and is currently being examined through case-control analyses, it must be recognized that Bross and Driscoll extracted their observation from matrices of over 4,000 data cells apparently by recombination of innumerable possible permutations of dosage and latency intervals. For that reason, their finding can be regarded as no more than a suggestion for further study. It certainly does not represent a proper scientific conclusion. Bross and Driscoll's analysis illustrates the hazard of performing multiple statistical comparisons on complex data sets in the absence of a priori hypotheses.