The Focus on Two Central Questions
This chapter is arranged in five parts:
The Focus of Our Analyses, p.1 Examination of All Cancer-Deaths Combined, Except Leukemia, p.2 Lifetime versus Minimum Fatal Cancer-Yield, p.3 A Potential Cause of False Answers, p.4 Reasons for Confidence in the A-Bomb Study, p.5
1. The Focus of Our Analyses
Chapters in the previous section have prepared the A-bomb database for answering some questions about the quantitative aspects of cancer-production in humans by ionizing radiation. The analysis begins in the next chapter. Therefore, it is important to be explicit now about which questions will be examined, because different methods are appropriate for different questions.
Two of the Central Questions in This Field :
Over the past twenty years, one question has been so commonly asked by analysts in this field that -- for convenience in discussion -- we have given a name to its answer: Lifetime Fatal Cancer-Yield. Although the question is phrased in many ways, it is always a variant of the following:
"How many extra fatal cancers will be produced among a population of 10,000 exposed persons per rem (or centi-sievert) of whole-body exposure, during the remaining lifespan of the entire group?"
Definition of Fatal Cancer-Yield :
Fatal Cancer-Yield is the number of fatal, radiation-induced cancers which occur among a specified number of irradiated people, per unit of dose, during a specified amount of time after the exposure.
Most reports and papers present Lifetime Fatal Cancer-Yield as a Number, N, per 10,000 irradiated persons (or N x l0^-4 persons), but some express it as N per hundred thousand persons (N x 10^-5 persons) or per million persons (N x 10^-6 persons). The radiation unit varies too. We use "per rem" or centi-sievert (rem^-1, or cSv^-1); others sometimes use rads (rad^-1) or grays (Gy^-1) or milli-grays (mGy^-1). A table of dose-equivalents is located at the end of our Index.
Initial Persons and Cancer in General :
Regardless of the units chosen, the analysts share a common thrust in their question: How many extra cancer-deaths, of all types combined, will occur in a specified number of initial persons per unit of radiation exposure to the entire body, during the post-exposure lifespan of the group?
Analysts addressing this question include, among many, the UNSCEAR and BEIR-3 radiation committees, myself, Hoffman and Radford (Hoff85), and RERF analysts Preston and Pierce (Pr87b; Pr88) and Shimizu and co-workers (Shi88). A directly related and very important second question is:
Is the Fatal Cancer-Yield the same at all dose-levels and dose-rates, or is the risk per rem at low doses and dose-rates different from the risk per rem at high doses and high dose-rates?
The focus of our analyses is on those two questions: The cumulative number of radiation-induced cancer-deaths per 10,000 initial exposed persons, and possible variation in the number of radiation-induced cancers with variation in dose-level or dose-rate.
Additional Questions :
Of course, there are many additional questions which could be asked. How many years of lifespan are lost by an individual who dies early of radiation-induced cancer? (This issue was treated in Go81.) Another example: When thirty years have passed since the bombing and some of the initial persons have died from a variety of causes, at what rate per 10,000 residual persons are people dying of radiation-induced cancer per year?
These and many other questions are perfectly valid, of course, and are not to be disparaged, but they are wholly separate and simply different from the two main questions which we are asking here.
2. Examination of All Cancer-Deaths Combined
Our analyses quantify radiation-risk to populations from all cancers combined, excluding leukemia, as do many of the analyses by others (recently, the RERF reports TR-1-86, TR-9-87, TR-5-88).
There are three main reasons for handling all cancers combined.
(1) Primary Site Misdiagnosis :
First, by keeping all the cancers combined, analysts avoid the errors originating with misdiagnosis of the cancer's primary site -- errors described near the end of Chapter 11. For instance, if lung-cancers have spread beyond the lung and are mislabeled as brain-cancers or liver-cancers, and if misdiagnosis of the primary site is common (see Chapter 11), then some serious errors are going to be made about site-specific rates, site-specific latency periods, site-specific sex-and-age differences, and so forth. Analysts invite error if they ask a database to answer questions which it is not capable of answering, or if they ask a question too early in a follow-up.
(2) Small-Numbers Problem :
Second, by keeping all the cancers combined, analysts avoid creating a small-numbers problem of real severity. As noted in Chapter 4, even the biggest database can be rendered inconclusive by excessive subdivision. The most reliable results come from the least possible subdivision.
When we look at the entries for cancer-deaths in Column N of Tables 11-B (males) and 11-D (females), we are impressed with the very small numbers of cases in many of the rows -- and these entries are for all types of cancer combined. This may surprise some people, who may have assumed that a study with 91,231 participants and a follow-up already lasting 37 years post-irradiation would not present analysts with the small-numbers problem.
What we all need to recognize is that 66,028 out of the 91,231 initial persons are in Dose-Groups 1 and 2, and they received almost no dose. When analysts attempt to subdivide the exposed groups by specific cancer-sites (or by age and sex), it is not surprising at all that we are still fighting the small-numbers problem, and the statistical instability inherent in such data. Column N in Tables 11-B and 11-D stands as a reminder that we cannot ask more from the A-Bomb Study than its database is presently able to answer.
(3) Scientific Believability :
Third, by keeping all the cancers combined, analysts may provide what many people desire: A believable estimate of how many people die from radiation-induced cancer of any type, per rem of whole-body exposure. Until this question can be answered in a scientifically credible way, it would be a mistake to attempt something inherently less credible: A site-specific analysis.
Moreover, there is no pressing social need to subdivide cancers into sites or classes, because when the issue is dying from radiation-induced cancer, few people care very much whether the fatal disease arises in one organ or in a different one (unless the time from exposure to death is very different -- an issue which remains clouded by the problems of misdiagnosis and small-numbers).
Exclusion of Leukemia :
It is customary to exclude radiation-induced leukemia from a combined analysis, however, because its distribution in time is clearly much earlier than the other radiation-induced cancers.
Leukemia's exclusion from the combination does not seriously diminish the total number of malignancies in the database, as Row 158 in Table 11-H can show. For the entire 1950-1982 follow-up, Table 11-H, Row 158, shows 233 cumulative deaths from leukemia (normalized data) compared with over 6,000 cumulative deaths from the other cancers combined.
The comparison in no way denigrates the importance of leukemia as a cancer, of course. However, the comparison does illustrate the fact that, when single cancers are considered in isolation, there can be a drastic drop in numbers and an associated drop in reliability for site-specific analyses. Readers might remember the numbers 233 and 6000. Nonetheless, some segments of the radiation community have been giving greater or equal weight to leukemia data than to all cancers combined (Chapters 22,25; recently Mu89).
Exclusion of Non-Fatal Cancers :
Evaluation of Fatal Cancer-Yield, by definition, excludes non-fatal radiation-induced cancer. We would like to make it clear that exclusion of non-fatal radiation-induced cancers from this book is not meant to trivialize the misery and expense caused by non-fatal cancer-cases -- including a high proportion of thyroid-cancers and skin-cancers. The exclusion here is due to the greater reliability, currently, of the available mortality data in the A-Bomb Study.
3. Lifetime versus Minimum Fatal Cancer-Yields
Cancer-induction by ionizing radiation occurs in an exposed population over years and decades. Although, by late 1982, the population of A-bomb survivors has already been observed to a time which is 37 years beyond the bombings, cancer-production by the exposure is definitely not yet finished -- it is persisting (Chapter 17).
At the present time, no analyst can know whether or not a limit exists on the post-irradiation time-period during which radiation-induced cancer-deaths (excess cancer-deaths) will continue to occur. If radiation-induced cancers continue to occur for the full lifespan of those who were only age 5 at the time of the bombings, then about another 37 years will be needed to ascertain with certainty the lifetime Fatal Cancer-Yield in this population.
The Minimum Fatal Cancer-Yield :
Meanwhile, along the way to the final count, there is great value in ascertaining how many radiation-induced cancer-deaths per 10,000 initial persons have already occurred at various times.
We call an interim count the minimum Fatal Cancer-Yield, because every radiation-induced cancer which occurs after an interim count can only increase the interim value. Radiation-induced cancers which have already occurred cannot be undone, so in common parlance, they are "in-the-box."
The cases which are "in-the-box" as of 1982 are not subject to dispute. They are not hypothetical, predicted, or dependent on any choice among competing models of radiation carcinogenesis (for instance, the "absolute" and "relative risk" models). The radiation-induced cases are known to have occurred by a straight-forward method. One can compare real-world events (cumulative cancer death-rates per 10,000 initial persons) in groups whose cancer-risk should be the same except for radiation exposure, and then the difference is the radiation-induced rate. We shall return, below, to the issue of the groups' comparability (see "A Potential Cause").
It is highly unlikely that cases already "in-the-box" are going to climb out of "the box." The only way for the final lifetime Fatal Cancer-Yield to turn out lower than the interim 1950-1982 Fatal Cancer-Yield would be for the irradiated survivors to start having fewer cancers per 10,000 initial persons than the Reference Group in subsequent follow-ups. But the evidence in Chapter 17, on the duration of the radiation-effect in the A-Bomb Study, is a powerful indication that excess cancers will continue to occur beyond 1982 in the irradiated groups.
But Just Suppose That . . .
Although the odds are heavily against the following scenario, we wish to describe it and to comment on its public health implications.
Let us suppose that at the end of the lifespan follow-up of the A-bomb survivors, when all 91,231 initial persons have died of one cause or another, the cancer death-rates per 10,000 initial persons are the same in the Reference Group and in the exposed groups. By definition, the lifetime Fatal Cancer-Yield would be zero. On the other hand, we already know (not a speculation) that interim counts along the way show that excess cancer-deaths have been occurring in the exposed groups. Such a combination of findings would probably mean that radiation accelerated fatal cancer in the people who were going to die of cancer anyway, but at some later time.
And we must comment on the meaning of advancing such cancer-deaths by 5, 10, 20, 30, or 40 years through radiation. Dying of cancer at age 20 years, or at age 30, 40, 50 or 60 years, is sharply different for an individual than dying from cancer "anyway" at age 70 years. Acceleration of cancer by radiation would almost make irrelevant whether there were any "extra" fatal cancers or not.
Periodic Examination of Minimum Fatal Cancer-Yields :
By examining the Minimum Fatal Cancer-Yield periodically, we are necessarily moving closer and closer to the ultimate Lifetime Fatal Cancer-Yield. If we reach a time when successive interim counts show no further increase in the Minimum value, this would suggest that we should not expect the final Lifetime value to exceed the most recently found Minimum value.
On the other hand, if successive interim evaluations of Cancer-Yield were to show a steeper rise than expected, or even a fall, predictions of the Lifetime Fatal Cancer-Yield could be appropriately revised up or down, prior to the final counts.
Therefore we think it would be useful, in this field, for analysts to report not only their current predictions about the lifetime Fatal Cancer-Yield, but also to state explicitly their findings about the current minimum Fatal Cancer-Yield.
4. Potential Cause of False Answers
We have emphasized, in Chapter 11 and above, the importance of comparing Dose-Groups whose cancer-risk, at any interim time during the lifetime follow-up, would be alike (except for sampling variation) in the absence of exposure to the bombings. Otherwise, Cancer-Yields could be falsely high or falsely low.
Our normalization process in Chapter 11 ensured that all the Dose-Groups were perfectly matched for age and sex at the outset of the study in 1950. The nature of the study itself makes the presence of additional confounding variables less likely in this study than in some others. In the A-Bomb Study, for instance, dose is related to the survivor's distance from the bomb's hypocenter. Since the radius from the hypocenters extends in all directions (except into the sea), and since two separate cities are involved, Dose-Groups are going to be composed of persons from many neighborhoods and occupations.
It is a reasonable approximation -- but an approximation nonetheless -- that the radiation-exposed and the unexposed groups will have the same exposure to all kinds of hazards which can affect death-rates along the way to the final count.
An Extreme Scenario Involving a Single Dose-Group :
The importance of this approximation can be readily appreciated by considering an extreme and extremely unlikely scenario. Suppose that one of the exposed groups moves into a single neighborhood sometime after the bombings, and suppose that a disastrous epidemic occurs in this one locale early in the follow-up. Suppose that the epidemic removes half of this particular cohort.
This scenario means that this one Dose-Group is no longer like the others in cancer-risk. Sorely depleted by the epidemic, it simply has many fewer residual persons, per 10,000 initial persons, available to develop cancer. So if we just count cancers, we could find that this one exposed group has fewer cancers per 10,000 initial persons at the end of the study than the Reference Group.
The scenario illustrates how fallacious answers could arise if initially comparable groups do not remain comparable during a follow-up.
Guidance from Non-Cancer Death-Rates and Person-Years :
Readers will have noticed, in Tables 11-B through 11-H, that Column P speaks directly to the central question of cancer death-rate per 10,000 initial persons, whereas Columns Q, R, S and T do not. The function of Columns Q-T, however, can be clarified by reference to the epidemic scenario above.
For instance, if the epidemic scenario had really occurred, the problem would show up almost immediately in Columns R and S, "All Deaths per 10,000 Initial Persons" and "All Deaths Minus All Malignancies per 10,000 initial persons." Both these rates would suddenly soar compared with the rates in the other Dose-Groups. (As noted in Table 11-G, "All Malignancies" include leukemia.)
Somewhat later, the problem would become evident in Column T, the ratio "Person-Years per Initial Person." Every person in a study contributes one person-year for each full year he(she) is alive during the follow-up period. If the epidemic scenario were really to affect one Dose-Group and not the others, the increment in person-years per initial person, in the affected group, would be far lower than in the other groups during each subsequent follow-up. On a cumulative basis (like Column T), the ratio in the group which was depleted by the epidemic would necessarily fall behind the rising ratios in the undepleted groups. (Under some other circumstances, however, a detectable reduction of person-years per initial person can be the result of enough extra cancers, which also cause people to be removed early from a follow-up.)
Eventually Column Q, "Cancer-Deaths per 10,000 Person-Years," may also show anomalies. But as a detector of an "epidemic" catastrophe, it is trickier because the group affected by the epidemic would have both fewer cancers and fewer person-years.
We have studied Columns Q-T, in Tables 11-G and 11-H for entire Dose-Groups, and in Tables 11-C and 11-E for separate age-bands. In examining the age-bands, one gives particular attention to the two oldest, because they account for the overwhelming share -- 77 % -- of the cancers observed so far in the 1950-1982 follow-up period (see Table 4-B).
Of course we see some anomalies, but this is to be expected on the basis of random differences in sampling (sampling variation). Column F provides a reminder of the modest magnitude of the numbers from which such rates arise. We see no indication, in the 1950-1982 follow-up data, that any calamity has occurred which would make it unsuitable to compare cancer-deaths per 10,000 initial persons in the Reference Group with such rates in the various exposed groups.
"Competition" from Other Radiation-Induced Deaths :
Above, we have discussed defenses against obtaining false Cancer-Yields due to an undetected and competing cause of death which affected one Dose-Group much more than the other Dose-Groups. So we should mention that there is one situation in which false Cancer-Yields would not result from a difference across Dose-Groups in non-cancer death rates per 10,000 initial persons.
Suppose that fatal cancer were not the only delayed (beyond 1950) cause of death resulting from exposure to ionizing radiation. For the sake only of illustrating this point, suppose that ionizing radiation also increased or accelerated heart disease. If it truly did so, we would expect to see Column S entries in Table 11-H already rising with dose, somewhat the way the cancer entries in Column P rise with dose. The extra deaths from heart disease would be competing with cancer for potential victims, and therefore we would expect a lower Fatal Cancer-Yield than we would see without this radiation-induced competition.
But it would not be a falsely low Cancer-Yield. Radiation would be "entitled" to do whatever it does, and if it caused sufficient extra deaths from non-malignant disease to lower the number of extra cancer-deaths per 10,000 initial persons, so be it. It would not cause a false evaluation of the cancer effect.
5. Reasons for Confidence in the A-Bomb Study
In epidemiology, there can never be a guarantee that analysts have detected every confounding variable which is important enough to invalidate the results.
For instance, suppose that all of the Reference Group in the A-Bomb Study has taken up residence next to a "cancer-factory," but no one realizes it. If this highly unlikely situation were real, it would be causing an underestimate of the Minimum Fatal Cancer-Yield from radiation, because the situation would be reducing the observed difference in cancer-deaths, per" 10,000 initial persons, between the radiation-exposed groups and the Reference Group. On the other hand, if all of RERF's Dose-Group 3 had taken up residence next to the "cancer-factory" while the other Dose-Groups had not, the result would be an overestimated Minimum Fatal Cancer-Yield at low doses.
As noted earlier, one of the great virtues of the A-Bomb Study is that its very nature protects it quite well (though never perfectly) from such hazards.
There is additional reason for confidence in the Cancer-Yields which will be obtained from the A-Bomb Study. Our normalization process in Chapter 11 ensured that the Dose-Groups were completely comparable in age and sex distributions at the outset of the study, in 1950. Moreover, we have an objective basis for believing that they have remained comparable, between 1950 and 1982. Table 11-H, Column S," certainly affirms that no catastrophe has yet affected one Dose-Group and not the others. So far, "All Deaths Minus All Malignancies, per 10,000 Initial Persons" remains approximately constant, within sampling variation, from one Dose-Group to another.
However, it should be noted that only about a third of the study's initial population has died so far (31,299, normalized, out of 91,231 initial persons). By the time that everyone has died, every excess cancer-death occurring in an exposed Dose-Group will necessarily mean a reduction in that group's non-cancer death-rate per 10,000 initial persons. There can, of course, be only 10,000 total deaths per 10,000 initial persons.
In the end, everything must add up.