UMPLASSIFIEA dismiss them as likely due to chance, even in the many cases in which the numberof workers understudyis so small that statistical significance would be difficult to studies are unduly shaded toward reassurance rather deaths was extreme. Nuclear weapons workers and the public alike may be falsely reassured by the DOE’s follow-up of such leads and other methods ofanalysis of the same data can yield important findings that would otherwise not come tolight but may be viral. The Task Force summarized reported trends or sug- achieve unless the excess of observed over expected emphasis, in repeated statements, that total death rates and cancer death rates among nuclear weaponsplant workers are usually lower than in the U.S. population at large. There are, however, alternauve ways of examining the data, though they are notreflected in DOEstudies. Comparisons of Standardized Mortality Ratios (SMRs) for cancer among nuclear weapons plant workers with the SMRs for other diseases or for total deaths may suggest excess cancer risks. (An illustrative example is presented in Table 2, Appendix F.) If specific data on radiation exposures of workers were available, more sophisticated analyses of this type would be possible. A second problem is presented by the almost exclusive reliance of the DOE studies on death rather than iliness, mortality rather than morbidity, as the health outcome examined. Mortality studies are admittedly cheaper and easier than studies of disease incidence. Yet many adverse health outcomescan be ascertained far sooner duringlife; mortality studies eliminate from consideration virtually all adverse health effects which may berelated to radiation exposure but which will not or have not yet caused death. This is especially true in the case of cancer; many cancers are now treatable, and some curable, and if life is prolonged or the disease cured, mortality studies of nuclear plant workers will not give a true picture of the frequency with which cancer appears in this group. Furthermore, because of the way in which death certificate informationis frequently coded, cancer deaths may be miscounted or falsely attributed to some other disease category. “Statistical Significance” and Fragments of Knowledge “All too often,” one researcher has noted, “investi- gators disregard a positive association between exposure and disease . . . because thefindingis notstatistically significant .. . . A consequenceis that negative findings can be guaranteed simply by doing studies of small populations or bystratifying data so finely that it be- comes impossible to obtain ‘statistically significant findings’ unless an extremely strong exposure effect is present.” Another has pointed out that “a small insensitive study may rule out very strong effects.” Repeatedly, our reviewers described studies in which DOE investugators have dismissed findings because they were notstatistically significant even if more than the expected numbers of total cancer deaths, or deaths from specific cancers, had occurred. Often the num- bers in any one study were too small to test for meaningful effects. Consequently,the interpretationsin these 12 than toward vigorous, inquisitive exploration ofclues, recognition of potential “sentinel” events or warnings, and growing magnitudes of effect over time. Careful gestions of excess rates of cancer (typically mortality rates) associated with working in the nuclear weapons industry at 14 sites, 11 in the U.S. and three in the U.K. (See Table 1.) We identified findings where there was cither a standardized mortality or incidence ratio over one (and the occurrence ofat least 5 cases), or a standardized ratio that was significandy higher than expected,or a statistically significant increase in cancer with increased radiation exposure. Table 1 shows an increase in deaths from all lym- phatic and hematopoietic cancers, non-Hodgkins lymphoma, brain and central nervous system cancer, prostate cancer and lung cancerin five or more of the populations. In addition, there were four sites with increases in bladder cancer deaths. These findings, in our view, do not justify a policy of under- interpretation, reassurance or premature dismissal. The epidemiologic research on the nuclear weapons industry lends itself to meta-analysis, a method involving the aggregation of results from similar but independentstudies. The lack ofstatistical power associated with studying one small group of workers can be overcome by combining theresults from several other studies. Meta-analyses may thus produce findings which were not apparentin any ofthe individual studics. Two recent meta-analyses have been published by non-DOE investigators. One combinedtheresults of seven previously published DOE and U.K. studies (only four DOE studies had sufficiently specific radianon dose data to be included) and identified a 50 to 80 percent increased risk of leukemia mortality among higherexposed workers; the other found a consistent 15 per- cent excess risk of brain cancer among 8 of 10 nuclear weaponsplant worker cohorts compared with the U.S. general population. DOEresearchers have begun to conduct studies pooling data from different sites, but continue to conclude that there is not “clear evidence of adverse effects of low-level radiation by external exposure.” Secrecy, Monopoly and Power From theearliest moments of the development of the nuclear weapons production complex, secrecy has been the most dominant and unvarying characteristic of the process, “National security” has been invoked to justify secrecy not only for the design of weapons, the processes of manufacture and theresults of testing but also for the data on radiation exposure and health DEAD RECKONING URCLASSIFIEE