Next | ToC | Prev


A Growing Problem:   Retroactive Alteration of the Study

This chapter is arranged in six parts:

  1. Necessity of Basic Rules, p.1
  2. Worst-Case Scenarios with Hypothetical Compound "XYZ," p.2
  3. Constant Cohorts and Credibility, p.2
  4. Criticism of the A-Bomb Study, p.3
  5. A Brief Chronology of the A-Bomb Study, p.3
  6. Summary on the Growing Problem, p.8
    Then a letter from the chairman of RERF.
    Then tables.

1.   Necessity of Basic Rules

          Epidemiologists worldwide appreciate the need for certain fundamental rules of research, four of which are described below. The need exists because scientists are neither saints nor robots. Scientists, like the people who are supposed to trust their work, have personal needs, hopes, and wishes which constantly threaten to undermine their scientific objectivity.

          The rules of research, needed as barriers to both the unintentional and intentional introduction of bias into research, ordinarily provide the basis on which scientists trust each other's results. Scientists who, and institutions which, are casual about such rules can destroy the credibility of major databases and lengthy research projects.

          This chapter documents the current menace to the A-Bomb Study in these respects.

          In order to assure readers that I have direct experience with what is and is not required, in order to maintain the credibility of a research project, I will mention a piece of my personal history. When I was proposing that certain blood-lipid measurements are predictors of human atherosclerosis (Go50a, Go50b), I helped to persuade the National Heart Institute to support a multi-center prospective study to test the hypotheses (Co56).

          Felix Moore, biostatistician at the National Heart Institute, helped to develop protocols, and I will feel forever grateful to him for his explicit insistence that every possible barrier to bias be built into the procedures, so that no one could dismiss the results -- however they might turn out -- by raising the question of possible bias. His toughness paid off. No one raised the question, and although the absence of bias cannot guarantee correct results, the results of that study have stood the tests of time very well.

Rule -- Change No Input :

          Prospective studies, like the A-Bomb Study, have a supreme virtue which normally gives them greater scientific weight than retrospective studies:   Their input is "set in concrete" before any results from the follow-up are known. Since there is supposed to be no later opportunity whatsoever to alter any of the input (for instance, measurements, diagnoses, cohorts), there is supposed to be no chance that unhappiness -- about the study's true outcome -- could affect the results or their meaning via retroactive revisions of the input.

          In short, results are supposed to have no influence upon the input. In prospective studies, the lesser opportunity for bias to operate is a major basis for their greater scientific credibility compared with retrospective studies. For convenience in further discussion, we can refer to the rule against fixing up input, after any results are known, as the "change no input" rule.

Rule -- Use Blinding :

          In the prospective studies where investigators feel compelled to disregard the "change no input" rule, care must be taken so that the changes are never made by anyone who is familiar with the results of the existing follow-up. We can think of this as the "use blinding" rule.

Rule -- Explain Everything :

          Because retroactive alterations of the input weaken the credibility of any study, investigators are expected to justify the scientific need for each change and to show how it could not have introduced any bias. They often do this with great attention to detail. We can refer to this standard as the "explain everything" rule.

          Every UNexplained change in input inevitably raises doubts, including the question:   What motivated the investigators to review one particular aspect of the input in the first place? Was something about the outcome -- absent the alteration -- making them unhappy?

Rule -- Keep Cohorts Intact :

          We must emphasize the importance of keeping a constant structure and keeping the same people together throughout the full course of a prospective study. In order to illustrate why this is so important for the believability of any study, we shall refer again to Chapter 2's hypothetical compound "XYZ," which explicitly excludes ionizing radiation.

2.   Worst-Case Scenarios with Hypothetical Compound "XYZ"

          We shall suppose that, in a study of XYZ carcinogenicity, there are five cohorts of 20,000 persons each:   An unexposed control-group (Dose-Group 1), plus Dose-Groups 2, 3, 4, and 5 which each received progressively higher doses of XYZ, but which otherwise are just like Dose-Group 1. And we shall suppose that, 35 years after exposure to XYZ, the true results are that cancer-rates rise above the spontaneous rate in direct proportion to the dose of XYZ received.

          Now, in order to make the point clear about intact cohorts, we will create worst-case hypothetical scenarios.

          Because these are worst-case scenarios, we will say that the XYZ study is controlled by unscrupulous persons who have an interest in making XYZ look less carcinogenic than they know it to be. Having specified that they are unscrupulous, we can say that they would not hesitate to conceal the carcinogenicity by shuffling the persons with cancer, in the study, until cancer-rates were very similar in all five dose-groups. Before moving a cancer-case to the dose-group where it was "needed," the XYZ partisans would assign the appropriate new dose to the person. And in order to create "noise" in the system, the handlers could assign new doses to a lot of cancer-free persons also. If the handlers of the XYZ study were to establish perpetual alteration of the cohorts and dosimetry as an acceptable policy for their study, they could continue shuffling people, as needed, with each subsequent follow-up.

          Of course, a study's meaning can be fixed, retroactively, by adjustments of its input other than the moving of cancer-cases. For instance, without changing the dose of a single cancer-case, the handlers of the XYZ study could arrange "evidence" even for a safe threshold, just by changing the doses of enough elderly cancer-free participants. Since the "expected" cancer mortality-rate (without XYZ exposure) is higher among older people than younger people, the handlers of the study would benefit by moving some cancer-free elderly participants into Dose-Group 2. This would simultaneously raise the mean age of Dose-Group 2, and lower its cancer-rate per 10,000 persons. The excess cancer-deaths really caused by XYZ exposure in Dose-Group 2 would be hidden, because a residual higher cancer-rate in Group 2 than in Group 1 would appear to be simply the natural consequence of Dose-Group 2's older mean age, compared with Dose-Group 1.

          With variations on these scenarios, handlers of the XYZ study could obscure the true carcinogenicity of XYZ at low doses for decades.

          If the membership of XYZ cohorts is treated as fluid, there are countless ways for the handlers of the XYZ study to obtain whatever results they want, by moving either cancer cases or cancer-free participants -- or both -- from cohort to cohort, after assigning them new doses as needed. By contrast, when cohorts are "set in concrete" early in a study, their constancy becomes a formidable barrier against such behavior.

3.   Constant Cohorts and Credibility

          The discussion of "Worst-Case Scenarios" with XYZ indicates why the scientific community appropriately has more confidence in the results of studies where investigators provide assurance that cohorts are never disturbed (or only minimally disturbed), than in studies where investigators retroactively and repeatedly change the membership of their cohorts. Continuity is the main barrier against the entry of bias, both conscious and unconscious.

          If cohorts are kept intact throughout the course of a prospective study, regardless of retroactive improvement of dosimetry, then everyone knows that (A) the mean age at exposure and the sex-ratio of the cohorts cannot be improperly used as described above in the XYZ study, and (B) a cancer-death occurring in a cohort belongs to that cohort forever, and subsequent follow-ups of that cohort will include every previously recorded case, because cases are permitted to "go" nowhere else.

          When cohorts are kept constant, even retroactive improvements in dosimetry can be handled without damaging the integrity and credibility of a prospective study. However, as we shall see in this chapter and in the next one, revision of doses has created a crisis for the A-Bomb Study because cohorts are not being kept constant.

4.   Criticism of the A-Bomb Study

          In contrast to all the virtues described in Chapter 4, the A-Bomb Study appears to grow increasingly deficient in adherence to the basic and meaningful standards of continuity in prospective research, in my opinion.

          I arrived reluctantly at this view. It arises from events described in this chapter's "Brief Chronology" -- especially events since the mid-1970s. I have said elsewhere that everyone who follows this field -- myself included -- should have spoken up earlier (Go88b).

          Now, with some 40 years of follow-up results on hand for the A-bomb survivors, the dosimetry and the cohorts of the study are being retroactively revised in major ways.

          I do not object at all to possible improvements in the study's dosimetry;   on the contrary, I welcome every genuine improvement. The problem is not the new dose-estimates. But very serious problems arise from the current and planned handling of such changes -- handling which is described in the years 1987 and 1988 of the "Brief Chronology."

          It would be tragic if the credibility of this honestly done study -- a study which has such importance for human health everywhere -- were to be unnecessarily and unfairly undermined by irregular practices which could so easily be unfavorably interpreted.

5.   A Brief Chronology

* 1950 :

          Japan's National Census of 1950 identified 284,000 A-bomb survivors throughout Japan, of which about 195,000 were enumerated in Hiroshima or Nagasaki (Bee71, p.615). Not all 195,000 were enrolled in the follow-up study. One initial criterion for inclusion was that the place of family registration (honseki) be in or adjacent to Hiroshima or Nagasaki;   survivors with more distant honseki were put into a "reserve" (Bee71, p.615).

          Beebe and co-workers state that all eligible persons who were within 2.5 kilometers of a bomb's hypocenter at the time of bombing (ATB) were included, plus "about 20 % of those more distally exposed, the latter selection being made at random, within restraints as to age, sex, and city, that ensured comparability with those under 2.0 km [kilometers] ATB" (Bee71, p.615).

          A "Not in City" (NIC) group of over 26,000 persons was identified and enrolled too;   it consists of persons who were away from the cities ATB or were located farther than 10 kilometers from the hypocenters ATB (Bee71, p.616). This group is not called "A-bomb survivors" and is not merged with the nominal "zero dose" cohort of A-bomb survivors in recent RERF reports.

* Mid-1950s :

          Excess leukemia -- meaning radiation-induced leukemia -- was showing up undeniably in the A-bomb survivors. "An organized dosimetry study was judged necessary," (Kerr87a, p.3).

* 1957 :

          Participants in the A-Bomb Study were assigned doses in the T57D dosimetry (Tentative 1957 Doses). T57D doses were based largely on distance from the hypocenter and on measurements made during post-war bomb-testing, at the Nevada Test Site, for instance.

* 1963 :

          The 1950-1960 follow-up on mortality in the study was published (TR-15-63;   Jab65). It covered a total of 99,389 persons:   About 73,000 A-bomb survivors and about 26,000 "Not in City" (NIC) persons.

* Mid-1960s :

          Excess solid cancers also had started showing up in the A-bomb survivors (Maki68;   Go69) -- not just excess leukemia.

          A more sophisticated dosimetry was worked out, largely by workers at the Oak Ridge National Lab, and the dosimetry was designated T65D -- Tentative 1965 Doses (Kerr87a, p.3-4).

* 1970 :

          The 1950-1966 follow-up on cancer-mortality was released (TR-11-70;   Bee71). This follow-up introduced the T65D dosimetry. The survivors were divided into four kerma dose-groups:   0-9 rads, 10-39 rads, 40-179 rads, and 180+ rads.

          The report changed the input of persons in a large way. A group of 9,513 A-bomb survivors, who had been ATB within 2.5 kilometers of a hypocenter, was added from the "reserve" of survivors whose honseki was not in or adjacent to either city (see 1950, above). Minor changes in the previous sample of 99,389 persons occurred too:   "56 subjects have been excluded because they emigrated from Japan, and 22 have been dropped for such reasons as duplication, error in identification, and failure to possess a family register" (Bee71, p.615). As a result, the sample of the 1950-1966 follow-up report consisted of:

  A-bomb survivors from the previous follow-up ...  72758 
  A-bomb survivors newly added ...................   9513 
  Combined A-bomb survivors in the report ........  82271
  The number 82271 includes 3791 individuals 
  with unknown doses.

  Not-in-city (NIC) group ........................  26553 

  Total sample in 1950-1966 report ............... 108824 

* 1971 :

          The 1950-1970 follow-up report on cancer-mortality was published (TR-10-71;   Jab72). The number of dose-groups was changed from four to five. About 650 persons were transferred from "dose unknown" into various dose-groups. The "Life Span Study Cohort Extended" was composed as follows:

  A-bomb survivors with known doses
       (see Table 5-A) ...........................  79113 
  A-bomb survivors with unknown doses ............   3131 
  NIC group (20176 Hiroshima;  6347 Nagasaki).....  26523 
  Total sample in 1950-1970 report ............... 108767 

* 1977 & 1978 :

          The 1950-1974 follow-up report on cancer-mortality was published (TR-1-77;   Bee78). The Not in City or NIC group was omitted. Although the number of survivors (82242) hardly differed from the previous report, about 625 persons were transferred from "dose unknown" into various dose-groups. The" number of dose-groups was changed from five to eight.

  A-bomb survivors with known doses
       (see Table 5-A)     .......................  79736 
  A-bomb survivors with unknown doses ............   2506 

Status of Existing Outcome, 1950-74 :

          The supra-linear, concave-downward shape of dose-response for cancer-mortality was already showing up in the 1950-1974 observations (Go81, p.380-85;   Go89a), and the record shows that the radiation community also saw indication of this curvature in the evidence (details in Chapter 22, Part 2).

Call to Re-Examine the Dosimetry :

          George D. Kerr of the Oak Ridge National Lab (ORNL) reported to the Defense Nuclear Agency (DNA) that at Hiroshima, neutron-exposure might have been overestimated two-fold and distal gamma-exposure might have been underestimated by two-fold. "These differences are too important in the investigations of the effects of radiation on man by the RERF to be treated lightly," Kerr told the DNA (Kerr87a, p.5).

* 1979 :

          The Department of Energy funded Kerr to complete an inquiry into the accuracy of the T65D dosimetry. Other analysts were added to the inquiry, partly funded by the Defense Nuclear Agency (Kerr87a, p.5). Work was also underway at the Livermore National Lab on revised estimates of the neutron-exposure.

* 1980 :

          The Livermore analysts, Loewe and Mendelsohn, issued their report entitled "Revised Estimates of Dose at Hiroshima and Nagasaki, and Possible Consequences for Radiation Induced Leukemia (Preliminary)" (Lo80). Thereafter, the literature became well furnished with published preliminary estimates from various sources, and they did not agree with each other in some important aspects (Kerr87a, p.7).

          The different impact which the competing proposals about dosimetry would have, upon estimates of cancer-hazard, was an obvious part of the debate from the beginning (see 1978 above).

The Blinding Issue :

          It would appear impossible to have achieved "blinding" for every physicist who helped to create the new DS86 dosimetry, because a proposed change in a physical parameter (e.g., gamma-ray energies, transport, or transmission-factors;   see Chapter 8) would have predictable consequences for raising and lowering doses, or for changing the shape of the dose-response relationship. For instance, a sufficiently large net increase in gamma dose for the distal exposed people would necessarily lessen the estimated cancer-risk per rem and steep supra-linearity of dose-response at low doses.

          RERF appears (A) not to see any problem in this aspect of "blinding," and (B) to regard the U.S. and Japanese sponsors of such work as disinterested. See the letter dated June 13, 1988 from RERF to the author, at the end of this chapter.

The T65DR Dosimetry :

          The 1950-1978 follow-up on cancer-mortality was released by RERF (TR-12-80;   Kato82). The input was altered in two important ways, which changed the term T65D into T65DR dosimetry (R for revised).

          First, Kato and Schull tell us (Kato82, p.396):   "Recently, as a result of further analysis of the physical evidence, the hypocenter in Nagasaki has been relocated 37 m [meters] to the southwest of the location used in computing the original T65 dose (3) and the dose estimates based on this new hypocenter have been used here."

          Second, Kato and Schull tell us that the rounding process used in the calculation of individual doses was changed to the method of half-adjustment (rounding to the nearest rad), instead of rounding down (dropping fractional values of a rad). They say (Kato82, p.396):   " . . . as interest has mounted in the possible consequences of exposure to doses of less than 10 rad, this practice [of rounding down] could obscure an effect through the systematic underestimation, admittedly small but possibly nonnegligible, of the exposure of individuals in the low-dose groups."

          [We would like to warn readers against inferring, from the statement above, that small dose-distinctions -- like a fraction of a rad -- can ever be taken seriously in the A-Bomb Study;   see Chapter 8, Part 4.]

Discontinuity of Cohorts :

          The two changes described above, though small, required RERF to recalculate all the individual doses. Of course, as a result, each of the eight cohorts of survivors necessarily acquired a slightly altered mean dose. However, instead of maintaining continuity by keeping the 1974 cohorts intact -- while using each cohort's new mean dose in calculations -- RERF chose to shift thousands of persons from from one cohort to another before computing the new mean doses. Table 5-A shows the net changes in the cohorts (by comparison with the 1950-1974 cohorts).

Status of Existing Outcome, 1950-78 :

          The supra-linear dose-response for cancer-mortality, seen in the previous follow-up, was persisting (Go88c, Go89a).

* 1981 :

          DOE convened a symposium to review what further work was needed on a new dosimetry, and several teams of analysts were established under DOE, NAS, and the Japanese Ministry of Health and Welfare to reassess the dosimetry for A-bomb survivors. Kerr87a lists the membership of each team;   some of RERF's leading analysts of the cancer-results were included, which necessarily involves the "blinding" issue again.

* 1983 & 1984 & 1985 :

          In the book which introduces the new DS86 dosimetry (Roes87), a chapter by Woolson and co-workers describes the very close coordination between those who developed the dosimetry and those who intimately know the results of the follow-up.

          Such practices certainly raise the "blinding" issue again. Indeed, in 1984, RERF inspected "the impact of the new dosimetry" on a segment of the survivors, and this "initial look" guided "further developments in dosimetry." To avoid any distortion of these phrases, we provide their exact context below (from Woo87, pp.405-406):

          "In May 1983, the authors proposed a dosimetry system for use by the Radiation Effects Research Foundation (RERF) that would incorporate the new findings and calculations of the joint United States-Japan working groups on the reassessment of A-bomb dosimetry. The proposed dosimetry system evolved from extensive discussions with RERF personnel, numerous meetings of the scientists from Japan and the United States involved in the dosimetry reassessment research, and requirements expressed by epidemiologists and radiobiologists on the various review panels . . . These discussions and our own experience indicated that, in the light of the expansion of computer and radiation technologies and the desire for more detail in the dosimetry, an entirely new approach to the dosimetry system was appropriate. This resulted in a complete replacement of the T65D system as distinguished from a simpler approach involving a renormalization of T65D parameters to reflect the new dosimetry.

          "The proposed dosimetry system for RERF and the plan for implementation was accepted by the Department of Energy (DOE) Working Group on A-bomb Dosimetry chaired by Dr. R.F. Christy. The dosimetry system plan was also presented to the binational A-bomb dosimetry review groups for critical comment and was discussed at joint US-Japan workshop. [sic]

          "A prototype dosimetry system incorporating preliminary dosimetry estimates and applicable to only a limited set of A-bomb survivors was installed on the RERF computer system in the fall of 1984. This system was successfully operated at RERF and provided an initial look at the impact of the new dosimetry [1]. [The reference is to Dale L. Preston;   see Pr85.] The experience gained by the use of this prototype paved the way for an improved system called Dosimetry System 1986 (DS86), which incorporated further developments in dosimetry and treated a more extensive set of survivors in the RERF data base.

          "The fourth joint dosimetry workshop, held in Hiroshima on 16 and 17 March 1986, reviewed the results and findings of the research to assess the A-bomb dose estimates and their incorporation into DS86. As a result, the US-Japan A-bomb radiation dosimetry committees formally approved replacement of T65D with DS86 for use by RERF for computation of doses to A-bomb survivors. The purpose of this chapter is to provide a description of DS86."

* 1986 :

          The 1950-1982 follow-up on cancer-mortality, by Preston and co-workers, was released by RERF (TR-1-86;   Pr87a). The study has additional names and numbers in the RERF system:   "Life Span Study, Report 10, Part 1," and also "Studies of the Mortality of A-Bomb Survivors, Report 8."

          The 1950-1982 report used the T65DR dosimetry exclusively.

New Discontinuities in the Cohorts :

          This follow-up report retroactively changed the input of persons in a big way:   It added 11,393 Nagasaki survivors who were exposed at a distance of 2500 to 9999 meters from the hypocenter "and for whom complete follow-up during 1950-1982 was available" (Pr87a, p.153). Preston and co-workers say:

          "These survivors, about half of whom have T65DR dose estimates of 0 centi-gray (cGy), were added to the cohort to increase the precision of the background (0 cGy) mortality rate estimates and, consequently, the excess risk estimates in Nagasaki. Although a comparable group of survivors is available in Hiroshima, it was felt that, since the Hiroshima 0 cGy group was relatively large already, little improvement in precision would result from the addition of these survivors to the LSS cohort" (Pr87a, p.153). Elsewhere (TR-1-86, draft), they say that the comparable group of Hiroshima survivors consists of 23,000 persons.

A Special "Reserve" of Input :

          It seems as if RERF has been conducting one public study, with 80,000 survivors on view (see Table 5-A), plus another study with over 34,000 additional A-bomb survivors in reserve, who are followed-up and selectively added to the public study as needed. These reserves are clearly not from the "Not in City" group, which was beyond 10,000 meters ATB from a hypocenter (see 1950).

          In addition, some minor changes of the cohorts occurred. According to TR-1-86 (draft):   "A total of 21 subjects were lost to follow-up due to emigration. In addition, a small number of subjects were reassigned among city, sex, age ATB, and radiation exposure categories on the basis of information received since the preparation of the last report."

          As a result of the changes described above, the expanded Life Span Study cohort (now designated LSS-E85) is composed as follows:

  A-bomb survivors with known doses
       (see Table 5-A) ...........................  91231 
  A-bomb survivors with unknown doses ............   2384 
  Not-in-City (NIC) group ........................  26517 
  Total sample in the 1950-1982 report ........... 120132 

Status of Existing Outcome, 1950-82 :

          The radiation-induction of fatal cancer from mean internal organ-doses as low as 11 rems was evident, the supra-linear shape of dose-response for cancer-mortality became unmistakable, and the widely suggested concave-upward shape became provably in error (see Chapters 14, 27, 29, and 30 of this book).

* 1987 :

          In 1987, RERF and NAS began revealing DS86, "the new dosimetry." [As stated earlier, the National Academy of Sciences (NAS) handles the U.S. Department of Energy's share of RERF sponsorship.] The main publications were Roes87, Elle87, TR-9-87, and TR-12-87.

          Using those publications, we have provided (Chapter 8, Parts 3 and 4) an overview of how DS86 differs from T65DR, and of some important questions which may never be settled, like the explosive yield of the Hiroshima bomb and the magnitude of unmeasured doses from fallout and activation products. In Table 5-B (this chapter), we provide an overview of how many A-bomb survivors do not yet have any place in the DS86 database.

Assignment of New Doses :

          From a NAS publication (Elle87, p.22-24), we learn that individual cases can receive very close attention, from someone, before they receive a new DS86 dose:

          "The desired end result of the dose reassessment is reliable estimates of the doses delivered to the various organs within a specified survivor. This aspect of the reassessment was performed with particular care . . . Three anthropomorphic models were selected . . . Several different postures e.g., standing, kneeling, etc., are considered in applying these models. Because dose to some organs, e.g., the breast, varies considerably depending on the orientation of the survivor relative to line of sight to the weapon, a special effort was undertaken by RERF to recover this information from the survivor shielding history . . ." (Elle87, p.24).

          Woolson confirms that cases receive close, individual attention (Woo87, p.430):   "Since complete and detailed drawings exist in RERF files depicting the survivor's location and surrounding neighborhood (see Appendix 7-2) [2], the largest reduction in the uncertainty of the dosimetry from DS86 can be accomplished by a better description of the survivor configuration in a new computer data base."

The "DS86 Subcohort" (Table 5-B) :

          Cases which were "difficult," in terms of estimating a new DS86 dose, have been suspended from reports, at least temporarily. Table 5-B shows that suspensions of Nagasaki survivors are high in all dose-groups, but are concentrated in Hiroshima in Dose-Group 2, where an astonishing 31 % is "missing in action" (our term). In all, 15,240 persons out of the total 91,231 persons (about one-sixth of the study-sample) were transferred into suspension -- another reserve of potential input.

          RERF calls the remaining 75,991 persons the "DS86 subcohort."

          TR-9-87 also reports, " . . . no efforts have yet been made by RERF to develop ad hoc procedures to assign environmental transmission factors to survivors shielded by terrain or in factories. However, as part of the continuing work of the reassessment committee special consideration is being given to dose estimation procedures for Nagasaki survivors in these important categories" (Pr87b, p12).

          We have another indication that suspension of 15,000 persons from the study is temporary. In a letter (August 11, 1988) to myself from the co-author of TR-9-87, Dr. Donald Pierce, we learn:   " . . . it appears that in the near future the DS86 subcohort will be very substantially extended, by computing doses for a large fraction of the remainder of the T65D cohort" (Pier88).

          Under the circumstances, it is far from clear what accounted for RERF's rush to introduce the DS86 dosimetry when so much of the study-sample still lacked doses in the new system.

Perpetual Enhancement of the Dosimetry :

          In both Elle87 and TR-9-87, we learn that the DS86 dosimetry is not only incomplete, but also provisional:

          TR-9-87, p.9:   "Work on the new dosimetry system is still in progress. It is certain that various aspects of the basic dosimetry system and the way in which this system is used will be modified over the next few years."

          TR-9-87, p.40:   "It is certain that over the next few years the DS86 system will be modified and enhanced. In addition, there will be changes in the ways in which this system is used at RERF to provide dose estimates for individual survivors."

          TR-9-87, p.46:   "It should be kept in mind that the DS86 dosimetry and the way in which it is used at RERF will change with time. These changes will arise because of enhancements and extensions to the original system and RERF's further development of procedures for computation of indirect estimates."

          Perpetual revision of DS86 is also suggested by NAS (Elle87, p.27):   "The DS86 dosimetry system is well conceived to fulfill the needs of the RERF. It is amenable to future development and should serve as the basis for any future amendments to the A-bomb survivor dosimetry that may be desirable."

          The U.S. review panel on the dose-reassessment project apparently expects future changes too:   " . . . the panel recommends that DS86 be reviewed periodically so that it does not become obsolete . . . It is not unlikely that additional dose estimates for A-bomb survivors will made by RERF to develop ad hoc procedures to assign be put forward from time to time as new information becomes available" (Elle87, p.30).

Oblivion Coming for the T65DR Database :

          R.J.M. Fry of the Oak Ridge National Lab (ORNL), and Warren K. Sinclair of the National Council on Radiation Protection (NCRP), presented an "Occasional Survey" in the October 10, 1987 issue of Lancet. Entitled "New Dosimetry of Atomic Bomb Radiations," their survey states that the DS86 dosimetry will replace the T65DR dosimetry:   "DS86 completely replaces the T65D system;   it is not merely an adjustment, and there have been major changes in dose estimates" (Fry87, p.847). Their statement is consistent with the "replacement" statements by Woolson (Woo87), already quoted in the 1983 segment of this chapter.

          Oblivion for the T65DR dosimetry is suggested also by a NAS report. On page 1 of An Assessment Of The New Dosimetry For A-Bomb Survivors, the editor, William H. Ellett writes:   "The T65D estimates are now being replaced, but they have had a long and useful life that amply justifies the considerable intellectual and financial effort that went into their development" (Elle87, p.1).

          These statements all refer to replacing the T65DR doses, instead of keeping them and simply providing each cohort with an additional entry, called mean dose in the DS86 system.

          The statements do not hint that the entire T65DR database (doses, cohorts, structure, and all) has been directed toward oblivion, but the direction seems confirmed in 1988.

* 1988 :

          In 1988, RERF released Technical Report TR-5-88 (Shi88). It is "based on the recently revised dose system, called DS86, that has replaced previous estimates of individual exposures . . . Here, the focus is on cancer mortality among the 76,000 A-bomb survivors within the LSS sample for whom DS86 doses have been estimated . . ." (Shi88, p.1). Some 15,000 LSS members were omitted because, although they had T65DR doses, they had no DS86 doses while TR-5-88 (Shi88) was in preparation.

A Completely New Architecture :

          In TR-5-88, which covers 1950-1985, the doses are not the only things which are new.

          Many of the remaining 75,991 persons have been moved into new cohorts (see Tables 10-A and 10-B). This means new groupings of the cancers which had been previously enumerated (1950-1982) among those 75,991 persons, and of course, it means new mean ages ATB and new sex-ratios in each cohort, too. TR-5-88 also introduces the subdivision of Dose-Groups 2 and 3, and changes the age-bands from five to six. Even the follow-up periods from 1950 onward are treated in units of 5-years instead of 4-years.

          In summary, the architecture of the A-Bomb Study is now completely altered -- not just the doses -- and the T65DR database has been removed from the public's sight.

6.   Summary on the Growing Problem

          The circumstances, events, and plans associated with "the new dosimetry" and described above would be an extraordinary threat to the credibility of any prospective study, no matter how disinterested its sponsors might be in the study's outcome.

          We have no evidence that bias has been introduced into the A-Bomb Study up to this point, and one of our main objectives is to help see that the future handling of the study is such that no one will ever be able to introduce a criticism of bias against it. It is self-evident that as the crucial second half of this study is unfolding over the next 10, 20, and 30 years, new personnel will take over its handling, and -- unless current practices are changed -- they will inherit a study in which continuity will be lacking and in which the changing of dosimetry-inputs may be more the rule than the exception.

          At the beginning of this chapter, I objected NOT to genuine improvements in A-bomb dosimetry, but to the way in which retroactive changes of many types are unnecessarily disrupting the continuity of this study.

          The chapter's title called it "a growing problem." It is growing not only because of all the recent disconnections between the study's past and its future, but also because of the explicitly expressed intention (see 1987) to continue altering the study's input, even as more and more of the cancer-outcome is at hand in the future. Such practices would unavoidably create a real threat to the scientific credibility of the whole study.

          In research, enthusiasm for improved accuracy -- in dosimetry, for instance -- is natural, and I share it too. However, we must not allow this type of enthusiasm ever to blind us to unintended, but nonetheless devastating, consequences of using the new insights in ways which would be certain to undermine the credibility of the entire effort.

A Reasonable Solution :

          A major thrust of this book is to propose and demonstrate a very different approach to handling the A-bomb database and to using the new DS86 dosimetry. We will show how an improved dosimetry can be added to the A-Bomb Study while preserving the study's continuity and valid prospective structure.

RERF's Point of View :

          We have, of course, expressed our concerns to RERF, and we present the gracious response from RERF's chairman at the end of this chapter's text. His letter makes three major points:

  1. The subject we raise is of great importance, and our concerns are scientifically legitimate, because retroactive alterations of a study's input can create opportunities for existing results to influence such alterations.

  2. But RERF feels the procedures associated with design and application of the new dosimetry are adequate to protect the study against bias and to maintain its credibility.

  3. Nonetheless, RERF will think over the need to maintain the T65DR database.


Telephone: 082-261-3131
Facsimile: 082-263-7279             5-2 Hijiyama Park, Minami-Ku
Cable: RERF HIROSHIMA                    Hiroshima, 732 Japan

  A Cooperative Japan - United States Research Organization

                      13 June 1988

John W. Gofman, M.D., Ph.D.
102 Donner Laboratory
Department of Biophysics
   and Medical Physics
University of California, Berkeley
Berkeley, CA 94720 U.S.A.

Dear Dr. Gofman:

          Your letter of March 31, 1988 was recently forwarded to me, as it raises an important policy issue to which Mr. Usagawa is unable to respond. The issue that you discuss relates to the recently concluded dose reassessment effort, resulting in the DS86 dosimetry system. It appears to be your feeling that this system should not be used exclusively, as a replacement of the T65D dosimetry system, but that any future evaluations of dose-response relationships on the Hiroshima and Nagasaki databases should be performed using both systems so as to avoid bias to enter the study.

          We fully endorse the view, of course, that it is essential for the dosimetry system to be designed and operated in such a way that there is no chance that the cancer data can affect the exposure estimates. This is true whether one applies T65D or DS86. The concerns expressed in your letter, however, suggest that you may not be fully aware of the way the dosimetry systens operate. First, the design of the new system was done entirely outside this Foundation, by U.S. and Japanese scientists, both groups with their own scientific oversight committees, and with no knowledge of, or access to individual survival information. Second, as far as calculation of individual exposures at RERF is concerned, this is done by well-defined computerized calculations, involving only the original shielding histories from which the T65D estimates were computed. There are no individual judgements made in which cancer results, or other biological endpoints, might come into play. This includes decisions regarding survivors for whom DS86 estimates cannot yet be computed, as these are made on classes of survivors, not individuals, defined by shielding histories alone.

          The history of the entire effort is contained in the final report, entitled "US - Japan Joint Reassessment of Atomic Bomb Radiation Dosimetry in Hiroshima and Nagasaki", a report that was published by this Foundation as a matter of convenience, not because it had any rights of authorship. I enclose a copy of the relevant section of this report for your information.

          I believe that, given the careful way in which the dose reassessment was carried out, with publication in the open scientific literature and continuous oversight as its significant features, your concern about bias does not appear to be justified. We shall, however, be sure to consider the necessity, and if so, the feasibility of dual analyses of our data. Please

John W. Gofman, M.D., Ph.D.     -2-                  13 June 1988

note that many of our recent Technical Reports have contained analyses involving both T65D and DS86.

          Thank you very much for your interest in our studies, and for your thoughtful remarks on a subject that is of great importance to both of us.


                                     Itsuzo Shigematsu, M.D.



Table 5-A
Number and Distribution of A-Bomb Survivors:   Changes of Input in Four Consecutive RERF Reports.

|                             Col.A   Col.B   Col.C   Col.D   Col.E   Col.F   Col.G  Col.H   Col.I  |
|                             Dose-   Dose-   Dose-   Dose-   Dose-   Dose-   Dose-  Dose-  Unknown |
|                             Group   Group   Group   Group   Group   Group   Group  Group   Doses  |
|                               1       2       3       4       5       6       7       8           |
| 1950-1970 Follow-up           Grps 1 + 2                               Grps 6 + 7 + 8             |
|      HIROSHIMA Persons           43730      10707    2665    1677           1460             1670 |
|      NAGASAKI Persons            11404       3700    1231    1229           1310             1461 |
| BOTH CITIES Persons              55134      14407    3896    2906           2770             3131 |
|      Persons with Doses:                                                                          |
|                    79113                                                          Source:  Jab72. |
| 1950-1974 Follow-up                                                                               |
|      HIROSHIMA Persons      29943   13796   10761    2718    1721     656    370     505     1441 |
|      NAGASAKI Persons        4700    6706    3759    1314    1391     748    270     378     1065 |
| BOTH CITIES Persons         34643   20502   14520    4032    3112    1404    640     883     2506 |
|      Persons with Doses:                                                                          |
|                    79736                                                          Source:  Bee78. |
| 1950-1978 Follow-up                                                                               |
|      HIROSHIMA Persons      27577   15933   10911    2783    1740     659    369     510     1429 |
|      NAGASAKI Persons        4004    7140    4031    1442    1388     722    270     377      957 |
| BOTH CITIES Persons         31581   23073   14942    4225    3128    1381    639     887     2386 |
|      Persons with Doses:                                                                          |
|                    79856                                                         Source:  Kato82. |
| 1950-1982 Follow-up                                                                               |
|      HIROSHIMA Persons      27569   15931   10909    2783    1740     659    369     510          |
|      NAGASAKI Persons        9604   12924    4034    1442    1388     722    270     377          |
| BOTH CITIES Persons         37173   28855   14943    4225    3128    1381    639     887     2384 |
|      Persons with Doses:                                                                          |
|                    91231                                       Source:  Tables 26-H, 26-I, Pr87a. |
NOTES   -----

  1. There is substantial continuity of cohorts in these four follow-ups, although the continuity is certainly not perfect. Altered cohorts in each follow-up mean that analysts are dealing with new mean doses, new mean ages ATB, new male-female ratios, and new distributions of previously reported cancers. Such changes are a serious matter, for the reasons discussed in the text.

  2. Table 5-B discusses the 1950-1985 cohorts for the T65DR database.

Table 5-B
Persons Missing from the DS86 Dosimetry, As Introduced in 1987, Plus New Alteration of T65DR Cohorts.

| Col.A |   Col.B   Col.C   Col.D   Col.E  Col.F  |   Col.G   Col.H   Col.I   Col.J   Col.K |
|       |                                         |                                         |
| T65D  |     LSS     LSS     LSS 1950-85 Percent |     LSS     LSS     LSS 1950-85 Percent |
| Dose- | Persons Persons Persons Persons 1950-85 | Persons Persons Persons Persons 1950-85 |
| Group | 1950-82 1950-85 Omitted    with     LSS | 1950-82 1950-85 Omitted    with     LSS |
|       |   T65DR   T65DR by DS86    DS86 Omitted |   T65DR   T65DR by DS86    DS86 Omitted |
|       |                           Doses by DS86 |                           Doses by DS86 |
|       | ------------  HIROSHIMA  -------------- | ------------  NAGASAKI  --------------  |
|   1   |   27569   27568    2324   25244     8.4 |    9604    9603     172    9431     1.8 |
|   2   |   15931   16573    5155   11418    31.1 |   12924   13194    3218    9976    24.4 |
|   3   |   10909   10363    1209    9154    11.7 |    4034    3805    1068    2737    28.1 |
|   4   |    2783    2707     113    2594     4.2 |    1442    1417     485     932    34.2 |
|   5   |    1740    1725      99    1626     5.7 |    1388    1375     718     657    52.2 |
|   6   |     659     657      63     594     9.6 |     722     723     295     428    40.8 |
|   7   |     369     369      41     328    11.1 |     270     267      95     172    35.6 |
|   8   |     510     506      74     432    14.6 |     377     376     108     268    28.7 |
|       |                                         |                                         |
| SUMS  |   60470   60468    9078   51390         |   30761   30760    6159   24601         |
LSS = Life Span Study.
All the entries in this table were generated by RERF:
Col.B entries come from Table 26-H in this book;   Col.G entries come from Table 26-I.
All the other columns (except A) come directly from TR-9-87, Tables 3 and 4, p.11, for 1950-1985.
      At page 7, TR-9-87 says that it means T65DR when it writes T65D.

      It is evident from Columns F and K that the DS86 dosimetry was far from complete at the time of its first use by RERF. Indeed, one-quarter to one-half of the Nagasaki cohorts were "missing in action", and an astonishing 31 % of Dose-Group 2 in Hiroshima was in suspension.

      Thus the initial "DS86 Subcohort" consists of only 75,991 persons who have both a T65DR and a DS86 dose-assignment (51,390 persons from Col.E + 24,601 persons from Col.J).

      The full cohort of 91,231 persons, in the 1950-1982 follow-up (T65DR database), is described by Columns B and G. Columns C and H describe the full cohort for 1950-1985 (T65DR database), according to TR-9-87. However, TR-9-87 does not explain the puzzling differences, which are rather substantial in Hiroshima's low-dose groups.

      For instance, Dose-Group 2 increases by (16,573 - 15,931) or 642 persons. Dose-Group 3 decreases by (10,909 - 10,363) or 546 persons. Dose-Group 4 decreases by (2,783 - 2,707) or 76 persons. Since all that anyone can observe here is the final net change in the size of various T65DR cohorts, it is conceivable that the number of affected persons was many thousands. Net changes occur in Nagasaki too. The consequence is that, in Dose-Groups 2, 3 and 4, the T65DR full cohorts in 1985 no longer have the same membership, mean age ATB, mean dose, or sex-ratio in Dose-Groups 2, 3 and 4 as the T65DR full cohorts had in 1982. Changes like this are a serious matter, for the reasons discussed in the text.

      However, we think that the disruption in continuity of the T65DR cohorts may be accidental and therefore temporary. A similar discontinuity within the DS86 dosimetry became evident in comparing TR-9-87 (p.11) with TR-12-87 (p.8). Those disparities were explained in a footnote in TR-12-87 (p.8) as the result of different RERF analysts using different rounding procedures for doses. The footnote states that "a consensus" among analysts there has now been reached.

Next | ToC | Prev
back to RIC | CNR | radiation | rat haus | Index | Search