Document text
Machine-read from the scan, so expect some errors. Lines that couldn’t be read are left out.
Page 1
Tenth Midyear Topical Symposium Saratoga Springs, New York Proceedings of Papers Presented at t b e Meeting
Page 2
R43IATION EXPOSURES OF HANFORD WORKERS DYING FROM VARIOUS CAUSE. Under Contract No. E(11-1)-3428 Hanford employees who are concerned with the manufacture of radiation products wear radiation badges and records have been systematically kept of the readings since1944 .(1) ConSined with suitable epidemiological information these data prcvi2e an opportunity to discover whether there is a cancer hazard associated with low level radiation. Linking ofthe epi de nio logical information, which includes such factors as sex,I zte of birth and work history,has been provided by the social security numbers of the workers. These have enabled a follow-up to be maintained to identify any deaths in the monitored population and to obtain death certificates. Yinor errors which accumulate more in the records of c uz rex t lya live employees, because they have no definite cutoff date,as dead persons do, invalidate strict comparability of the records of live and dead employees. Therefore,in place of waiting until all such errors have been strictly eliminated from the data,the current presentation is based only upon comparisons among various categories of dead employees. included3,521 men and 362 women who died between 1944 and 1972 and for whom death certificates were available(table 1). Analysing the annual proportions exposed and the mean dose per exposure in various ways (including breakdowns by calendar year,years since hire,years before death, and age at exposure) showed that the exposure proportion was systematically higher for the cancers than the other causes of death(table2,Fig. 1).
Page 3
I n view of these findings it was decided t o : (i) apply a strict statistical test based on the logistic model(*) to the mean doses cumulated in various ways; (ii) quantify the radiation sensitivity by estimating the.doubling dose bz sed o n - the linear model asa working hypothesis. The radiation doses were cumulated to stated periods before death, and the period before death which showed the maximum statistical significance of the difference between the nean dose of a given type of cancer and that of non-cancers (control or standard group) provided an estiinilte of the latent per: ' 0 2 of the cancer(tables3 and 4).By cum u latin g thera diet ion sensitivity to cancer induction were obtained. By these tests the only cancers showing significant radiation associations were myeloid neoplasms, lung cancer and pancreatic tumors for men and breast cancer for women (table5). These estimates were checked by forming from c h ~ aestinates of the excess mortality ratio due to radiation a;ld con paring these ratios with the observed standardized nor t ali tyrE tio from national statistics(table6). The only disease s:?owing a marked difference in this validity test check was lung cmcer and this may possibly be accounted for by factors such as smoking and other hazards from industrial exposure.Estimates were also made of the actual numbers of radiogenic cancers
Page 4
(table7),with the finding that about 30 cancer deaths(or the results also showed that the ages of maximum sensitivity to radiation were at both ends of the life span for adults. rL ~ P ,c anc ers that were found t o be radio sensitive, and their estimated latent periods,agree in general with what i s known or suspected of adult cancer a etiology ; however, the estimate o f doubling dose were lower(i.e.,estimated radio sens it iv ty was greater)than previous studies had led one to suspect.Further analysis to establish the exact shape of the dose - response curve and to study possible interactions and associations with potential pro not ers of cancer and other factors are clearly nec es sz ry, bckh for this study ar,d in other populations exposed t o siailar levels of dose. the6th Annual Health Physics Society Topical Symposium Vol. 111. Study of the Lifetime Health and Mortality Experience of Employees of AEC Contracts.COO - 3248 - 1,1971. 2 . Cox, D.R. Analysis of Binary Data. Methuen, London, 1970.
Page 13
authored by Mancuso, Stewart and Kneaie" Excuse me for corning t o the podium t o comment about Dr. Stewart's talk. I will also request the privilege of projecting a few tables and figures to illustrate some points i n my renarks. First,I would like to review briefly the history of the Health and Study under ERDA sponsorship f o r twelve years. has taken great pains to refine his very large body of data as xuch a s possible. In addition, during the past few years, Dr. Barkev Sanders, Dr. Mancuso's former associate i n the study, has anaiyzed the dzti? i n arranged foran independent analysis o f the data by Crs. Stewart 2nd Kneale, the results of which have been prisented today. that 3r. Mancuso's new team o f independent analysts has not had sufficient opportunity t o probe the data in great depth, which we feel has led t o certain errors i n analysis that I wish to discuss. I n presenting this material so soon, they have also been deprived o f the benefit of adequate t o us and t o other interested persons during briefings by Drs. Mancuso and Stewart during the past two weeks, but this has prwided only a limited opportunity t o check out their analyses andto reanalyze the data in a manner that might have been more apprcpriate. My remarks will necessarily be limited t o comments about the tables Dr. Mancuso has conducted the Health a n d Mortality During that period he Recently, Dr. Sanders l e f t the study 2nd Er. Mancuso My reason for making these remarks is that we a t Hanford belizve Certain tables and figures from the p a p e r were nade available and figures orto conc1usions.that flow directly from the tables. Battelle we have had access t o similar computer printouts and can indicate some biases that have become apparent during the last year and a half i n a s t u d y of the Hanford mortality data. The study was authorized by ERDA and carried out by Dr. Ethel Gilbert of the Bat tell e it at is t ical staff, who i s here today and can respond t o questions. based on Dr. Gilbert ' s work. My remarks today a r e
Page 14
To get down to specifics, we believe that there is a consistent bias that results in overestimation of the mortality rates for cancer i n the exposed workers. We feel that this accounts for many of Dr. Stewart's tional mortality. In proportional mortality, one considers the number of deaths due toa cause ora group o f causes, such as cancer deaths, as the proportion of the total number o f deaths. i f one cannot define or study the population i n which the deaths occur. However, for this study the population is well-defined and the follow-up corrplete enough that biases are not likely to occur i n comparisons between groups within the population. Proportional analysis suffers from the' defect thata deficiency o f deaths from certain causes will increase the proportion of deaths from other causes. analysis, which i s now under way, reveals that this i s the case ina comparison o f exposed and unexposed workers. We see i n Table 1thata large percentage of th2 unexposed employees worked a t Hanford for less than two I!E semnd part of the argument i s seen i n Table 2 , which shows that the mrt ali ty rates from causes otkr than cancer are higher inthe short-term, unexposed than i n the long-term, predominantly exposed workers. trast to this, the cancer rate sar2 approximately the same for the two say t h a t , when correction is provided for age, calendar year, sex and race, for 100 deaths ocicurring as a result of circulatory diseases i n the U.S. white male population, 89 deaths from those diseases would occur among the short-term workers having employment periods of less than two years, while only 75 deaths would occur among t h e workers who were employed for more than two years. the fewer deaths from other causes artificially inflate the proportion O f cancer deaths i n the exposed, long-term group. This bias will affect all tables in which doses are compared for cancer and noncancer deaths. The bias may be due t o the steady employment w i t h excellent health surveillance Dr. Stewart has analyzed t h e data i n terms of proporThis method is used widely From a n enployment standpoint they were essentially transient. I n interpreting the population-based mortality rates, one might The significance of this bias is that
Page 15
and good medical care programs t h a t long-term workers in ERDA plants enjoy or t o the process of self-selection that takes workers outof more rigorous employment when they begin to suffer from chronic circulatory or respiraradiation doses i n the employee population of a plant like the Hanford Plant i s markedly skewed and distinctly not a normal distribution. Figure 1 shows the distribution of doses among deceased workers considered i n Dr. Mancuso's and i n our study. orlow doses witha long tail of fewer, substantially higher doses Hill be very familiar t o this audience. cations t o Dr. Stewart's analysis. We feel t h a t Dr. Stewart's use of the mean i n many of her tables is improper. markedly skewed distributions with 9 poverty families i n discussing annual incomes. I t would be m2mingless t o say that 10 families have a n average incme of Sl0,OOO i f cne has an income o f $91,000 and each of the others $1 yOGo2 per year. our analogy is not far fetched. change the values o f the means drastically in Dr. Steuart ' stables. Table 3 , dealing with cancer of the pancreas, gives tne effect of dropping the one or two l a r g e s t values on the mean of 49 values. t rated even more dramatically i n Or. Stewart's analysis of cancer of the breast where her findings appear to rest on a single csse. In Table 4 , we have presented a listing of the cases i n order of msn it ude of total cumulative dose. We see that one case i n which the total dose i s 74.39 rem dominates the means for the entire sequence. Table 5 provides a cotiiparison between Or. Stewart's tabulation of the means of t h e cumulative doses by years before death witha similar tabulation for which the one high dose However, I would hate to think that the implications o f this report might Fly second major point is that the distribution of total cumulative The very large percentage of zero This distribution has two important 'inpliWhen dealing with such the mean does not effectively sumar5ze We might consider the comon analocy o f lunping the millionaire If we look at some of the data analyzed by Dr. Stewart, we find that Occasional high total exposure values d o This is illus - We acknowledge that the or,e case should not be ignored.
Page 16
be affected by one case more or less of a disease that is as common as cancer o f the breast. for believing thata radiation effect on breast cancer can be seen i n this popul ati on. The same table points u p another difficulty. Drs. Stewart and Kneale have used at statistic in deciding whether the comparisons o f the neans of radiation doses for different disease categories are sign i - fi cant or not. the t statistic as it was on the mean. With a distribution of radiation doses a s nonnornal and skewed as thatI showed’in Figure 1 , very large s a ~ p ’ l e swould be required for the use of this test of significance of t statistics thc lt are calculated for very small samples and then used t o indicate levels of significance. use of statistical methods are not valid. Dropping t h a t case would remove Dr. Stewart’s basis The effect of a single large observation is a s great on Ne can see that there are many examples inthe tables The conclusions based on such misTo show where these methods can lead one, l e t us consider the table i n which Dr. Stewart concluded that5.9 radiogenic breast cancer deaths occurred ? nthis population among 7 exposed females, which came out t o 84.3% o f the exposed cases. return t o ti?? listing of doses, we are obliged t o conclude one of three things. Either the one woman w i t h 1 4 . 4 rems died 6 times o r the second alternative is that breast cancer resulted from radiation doses as low a s 90 millirems, which i s well below the dose that the female worker would have received from natural background i f she had moved t o Denver instead of remaining i n Richland. The third alternative is that the methods used i n the analysis have led toa ridiculous conclusion. The data i n Table 6raise another issue of less significance from the standpoint o f the overall analysis but of some interest in itself. Dr. Stewart has combined myeloma deaths with myeloid leukemia deaths. Although she believes that these are closely related diseases, the great majority of hematologists regard them as distinct entities.In Table 6 , we have listed separately the cases of these two diseases. We find that If we nay round the 5 . 9 t o 6 deaths and
Page 17
there are three workers with unusually high total radiation dose values among the eleven multiple myeloma cases. This is of considerable i n - te rest and bears further investigation. We have some reservation about inmediately concluding that this is a radiation effect because previous large s e r i e s of cases of radiation exposure such as the survivors of the atomic bombings i n Japan and the ankylosing spondylitis cases in Great Britain have failed to show excessive numbers of multiple myelomas. However, we do feel that this finding warrants our further attention. On GULL,the other hand, t h e mortality from myeloid leukemia, when separated i s not remarkable, and the exposure values are not unusual for this limited number of cases. Before closing, I would like to present some results of Dr. Gi5ert ' s analysis. She reviewed the various occupational categories among t h e total employee population and analyzed those who were classified as craftsmen and operators because they included the vast na j or it y of high categories with greater and others with substantially less potential the percentage for each group that had total doses greater than 5.3 rems, which is the 95th percentile i n the distribution of doses. this comparison provides standardized mortality ratios for mal …
The text continues in the original scan (28 pages).