Summary of Reviews of " Mancuso Report " (Final Report No. 13)

Page 1

This document isa summary of the written evaluations made by a number of outside reviewers who are experts i n the epidemiologic study of radiation effects i n human populations. demiologists has also been selected by the Assistant Administrator f o r Environment t o prepare for hima statement regarding the scientific An ad hoc panel of epimeritsof the Mancuso Report. This panel i s scheduled t o meet on the mortality pattern of Hanford workers. The analysis concentrates on those men who have died. The authors' comparisons of proportions o f men dying from various types of cancer, and of mean doses i n cancer deaths and other deaths, purport to show a relationship between "lowlevel '' occupational radiation exposure and the development of cancer. All reviewers were critical of the methods used by Mancuso e t a l . i n handling this potentially valuable source of data. The scientific merit of the findings presented in the Report was seriously usual method applied t o longitudinal studies of populations. The method used did not allow conclusions t o be drawn a b o u t risk of death from cancer or specific types of cancer. relationships were based on inadequate case-control study designs , without sufficient consideration of possible confounding variables. I t was noted that the data had not been handled in the 1 Cohort-type designs would have been appropriate, as i n other epidemiologic studies of radiation effects (e.g. , atomic bomb survivors, uranium miners). Until such analyses are performed on t h i s population,

Page 2

the findings cannot be meaningfully compared t o those from other studies on radiation effects. The two types of study design used in epidemiologic research are the "cohort" and "case-control"; each has i t s own problems and limitations i n examining hypotheses about causation o f disease. A case-control study design involves comparison of "cases" witha particular disease (e.g. , a type o f cancer) and appropriate "controls" without the disease. The variable of interest is usually past exposure (or level of exposure] t o an agent believed t o be causally related t o the disease. Ina cohort study, a cohort (group) of persons exposed t o a given agent and a cQhort of similar nonexposed persons are followed-up over time. in the two cohorts. Standard statistical methods have been developed The frequency of illness or death i s compared f o r analysis of the results from each of these two study designs. The Mancuso study has some features of a case-control study, in that radiation dose i s compared between "cases" (e.g. , cancer deaths) a n d "controls" (other deaths). The methods used, however, are , inadequate t o assess the effects of factors other thail those of interest (.i .e. , radiation dose and mortality). Some consideration i s given i n the Report t o the factors of calendar time of employment, age at first employment and employment years t o death. Other factors are ignored, such as the type of work, duration of employment, a n d disability retirement which may influence both the cause of death

Page 3

and the poss-ible level of exposure to radiation. should have been taken into account. The analysis o f case-control studies must either enter all relevant factors into a multivariate analysis framework, o r must control for these factors i n the selection of cases and controls. factorial analysis which might have been done. isolated individually as in this case, itis impossible t o determine how much of the difference in each factor can be attributed t o multiple characteristics o f a few individuals. All of these factors The manuscript does n o t discuss any multiWhen each factor i s Mean annual radiation doses are presented for cancer and nonThe differences i n meancancer decedents among "exposed" workers. cannot be analyzed statistically without standard deviations and distributions of doses. This information isnot provided. Mean cumulative doses are given for selected causes of death a n d ages a t death. The only striking findings are the h i g h mean doses for "RES neoplasms" i n certain age-at-death groups. Mean cumulative doses for one needs t o know the distribution of doses within the (small) group of RES cancers, and specifically for multiple myelomas. If proper case-control study designs had been used i n this Report, the magnitude of the relative risk of death from cancer associated with radiation exposure could have been estimated by standard methods. Relative risk (exposed vs. not exposed, or various exposure groups vs. not exposed) could have been estimated for certain types of cancer.

Page 4

As noted above, multifactorial methods would have been needed i n such Further adjustment f o r age a t death and year of death (by other investigators) suggests the possibility of an association between radiation dose and death from only two types of cancer (multiple meyloma, and cancer of the pancreas). small numbers of deaths, with relatively high radiation doses (i.e., over 10 rads).Thus, the statement that "there isa definite relationship between low level ionizing radiation a n d cancer" ( i n the summary o f the Mancuso Report) i s questionable. These findings are based on The cohort (vs. case-control) design should have been used t o analyze these data. In a cohort design, use of the person-years method of follow-up*, from year of employment or f i r s t exposure, would be appropriate. cross-classification would be needed t o take into account confounding variables (-length of employment, year of First employment, U. S. would be used, and person-year distributed according t o sex, age, and time. Multiplying these person-years by the U. S. annual death rates provides "expected" numbers of deaths, which can be compared with the numbers actually observed i n the occupational group. Of Various types of subclassification and *For example, for a given 5-year period of calendar time, a person under observation for the entire5 years contributes 5 person-years, while a person who dies a t the mid-point of the period contributes 2.5 person-

Page 5

particular importance would be careful analyses by length-of-employthe person-years method o f follow-up, e.g., in using groups that b-y definition had t o survive, 5 , 10, or 15 years after first employCare must be taken here t o avoid biases resulting from Comparisons of mortality by dose group could be made within Some of the analyses i n the Report deal with proportional mortality. This is simply the proportion of all deaths that are attributed toa given cause (or causes). of cancer deaths among several radiation dose groups, and also compares The Mancuso Report compares the percentage the percentage o f cancer deaths in the Hanford workers with the percentage expected on the basis of U. S. white males dying in 1960 Although proportional mortality can be used i n cohort studies, the limitations of this mehtod are well-known t o epidemiologists. These limitations are not mentioned by Mancuso etal. gives no indication o f actual risk o f death or disease associated with a given factor (in this case, radiation dose). lation of a mortality rate, the denominator of the rate is all individuals exposed t o a given dose. Proportional mortality must be used with great caution, t o avoid groups are lower than those i n the general population, since workers are selected on the basis of past and present health. This is known For example, actual --death rates i n most occupational

Page 6

to differ from long-term workers i n health and other factors relevant with larger cumulative radiation doses. of proportional mortality by length of employment. proportional mortality analysis can only suggest possible association which then should be explored by more meaningful methods. In the Mancuso study, longer work periods are associated This complicates any comparison Other inadequacies of the Report go beyond the specific methods of analysis, and indicate a lack of critical judgment. limitations of death certificates,in terms of accuracy of diagnosis of disease (e.g., cancer of the pancreas) are n o t discussed. Report fails to mention that i n studies involving a large number o f statistical tests or comparisons, there isa probability some will turnout to be statistically signifiaant b~ chance alone. identify with certainty those results that are real and those that are chance findings. Statistical "proof" of a real o r causal association cannot be obtained. statistical association are based on the strength of the association, its biological plausibility, and its agreement with the findings o f Judgments about the reality of a Mancuso etal. apparently did not note the discrepancies between t h e i r findings and those on Japanese A-bomb survivors and other findings of the study with regard t o multiple myeloma, but negative findings for leukemia,should have been noted by Mancuso etal., and For:iexample, the purportedly postive

Page 7

should have led them t o be suspicious of their methods and results. The evidence for a dose-related increase i n multiple myeloma in the extensive studies of Japanese A-bomb survivors i s weak and equivocal, while the association between radiation exposure a n d increased risk of leukemia is well-documented. study should not be made t o stand alone. from the nature of statistical tests and what they can tell us. The findings of any -one epidemiologic This should be apparent Final Comment -- Current Progress of Studies on Hanford Workers Studies of these populations have n o t been discontinued. must be carried out, hGwever, with designs and methods that avoid (as much a s possible) misinterpretation o r misleading presentation o f the larger body of literature on radiation effects on humans. These studies must allow comparisons of their findings with work i s now in progress, and a draft of a report has been prepared by has compared numbers o f deaths observed from specific causes in the Hanford workers w i t h numbers of deaths expected on the basis of U. S. Dr. Marks' analyses, using the cohort study design, Briefly, the total number o f observed deaths in the cohort o f workers was 85% of the number expected on the basis of rates for U . S. white and i s cornpatible with the "heal thy worker effect" mentioned above. This low percentage isnot unexpected i n occupational groups, *These death rates are specific f o r sex, age, calendar year, a n d cause

Page 8

Excess mortality has n o t been observed for leukemia. multiple myeloma) was greater than expected, but the finding was not statistically significant. More detailed analyses are i n progress, including consideration of job classification, duration o f employment and radiation dose.