Next | ToC | Prev


The Popularity of Risk Reduction-Factors in the Radiation Community

This chapter is arranged in six parts:
  1. Magnitude and Premise of Risk Reduction-Factors, p.1
  2. A Historical Perspective, p.2
  3. The Exact Statements Which Convert Myth into Consensus, p.7
  4. Unwarranted Conclusions from the Canadian Fluoroscopy Study, p.14
  5. Unwarranted Conclusions from the Holm Radio-Iodine Study, p.15
  6. The Bottom Line, p.25

1.   Magnitude and Premise of Risk Reduction-Factors

          The previous section of this book established that there is no safe dose or dose-rate of low-LET ionizing radiation, with respect to induction of human cancers.

          This section takes up a separate issue: The general practice, by the radiation community, of reducing estimates of cancer-risk if the exposure is low or slow. This is the usual practice, whether or not any suggestion is made of a completely safe dose or dose-rate.

          With the exceptions of breast-cancer and thyroid-cancer, the radiation community has been making its cancer risk-estimates for low-LET, low-dose exposures by asserting that the "effectiveness" or carcinogenic potency per rad is considerably less at low total doses than at high total doses, and also that the risk is considerably less if a dose is received slowly than if the same dose is received all at once.

          In other words, the radiation community reduces estimates of risk-per-rad, observed in humans at acute high doses, by factors which are often called "dose effectiveness factors." There are two possible kinds of such factors: (1) one for the alleged reduction in cancer-risk per rad in going from a high total dose to a low total dose, all delivered acutely, and (2) one for going from a high dose-rate to a low dose-rate.

Definition and Magnitude of "DREFS" :

          The NCRP (National Council on Radiation Protection) has, in essence, combined the two types as a single factor, referred to as the DREF, the dose-rate effectiveness factor. NCRP states that the DREF could also be called a "dose-magnitude effectiveness factor" (Ncrp80, p.9). A DREF is the ratio of two linear slopes: A steeper slope over a lower slope (Ncrp, p.176). The significance of these slopes, demonstrated in the next chapter, need not be considered for now.

          NCRP gives a range for its DREF factors as 2 to 10 for human carcinogenesis by low-LET radiation. Thus, if a human cancer-study provided cancer-risks per rad from observations at very high doses, these risk-rates would be divided by a factor of 2 to 10 to derive what NCRP considers appropriate estimates for cancer-risk per rad at low doses or low dose-rates.

          Some users of DREFS refer to these very same factors as 0.5 to 0.1. They are simply multiplying the observed cancer-risks at high doses or high dose-rates by 0.5 to 0.1 instead of dividing by 2 to 10. The fractional formulation is in better accord with the formal meaning of "factor," which is defined by Webster's dictionary as "any of two or more quantities which form a product when multiplied together."

The Premise of DREFS :

          The underlying premise of DREFS is that human dose-response for acute exposure is likely to have a concave-upward shape, except at extremely high doses. This premise is not only stated clearly in the NCRP report, but also in UNSCEAR 1977 and ICRP 1977 (see Part 3 of this chapter).

          Of course, readers have seen for themselves in Chapters 13 and 14 that this premise is invalidated by the A-Bomb Study (1950-1982), whose dose-response has the opposite curve throughout the full dose-range. And readers are reminded that when RERF analysts, Shimizu and co-workers, examined all the A-bomb data 1956-1985, they too found the dose-response to be either linear or supra-linear (see Chapter 14, Part 2). Even if data in subsequent follow-ups were to produce a dose-response more linear than supra-linear, the underlying premise of a concave-upward shape would still be invalid.

          Readers may wonder, "If the underlying premise of reduction-factors has already been invalidated, then why bother to discuss this topic at all?" Such a question assumes that human evidence will be accepted as the decisive reality-check on expectations concerning humans.

          Our position is that it should be. Therefore, we are critical throughout this chapter whenever human epidemiological evidence of good quality is subordinated to non-human evidence.

          Parts 2 and 3 of this chapter will show that "the general wisdom" of DREFS first emerged in the absence of good human epidemiological evidence, and also that, for many years already, good human evidence has been at variance with the underlying premise of DREFS.

          Some readers will find it intrinsically interesting to see for themselves that, nonetheless, DREFS have been overwhelmingly popular and well embraced in the radiation community. This is demonstrated to mid-1989 by the statements provided in Part 3. Although we focus, in Part 2, on only three of the DREF documents, readers will find both earlier and later documents quoted in Part 3. (The quotations in this chapter may not be fully understood by readers who are unfamiliar with the linear-quadratic hypothesis of dose-response in this field. They may wish to study Chapter 23, Parts 1 and 2, before this chapter.)

          In Chapter 23, we will show our own quantitative approach to risk-estimation for acute-low doses and for slow-low doses -- an approach which is consistent with the existing human evidence.

          We call attention to our emphasis above on low doses: Acute-low and slow-low. As for high total doses delivered slowly ("slow-high doses"), this topic is explored separately in Part 6 of Chapter 23.

2.   A Historical Perspective

          Although good human evidence on the shape of dose-response exists today, a couple of decades ago, the human evidence was very thin on this issue. Some animal experiments suggested that the dose-response relationships for tumorigenesis and certain other biological end-points were concave-upward, meaning that the cancer-risk per rad (cGy) could be higher at high doses than at low doses. If this were true for humans, then it would mean that extrapolation from high doses to low doses in a linear fashion would overestimate the cancer-risk at low doses.

          If one were to rely on some of the experimental animal data, and assume the human dose-response relationship to be concave-upward, then reduction-factors would seem reasonable in trying to assess cancer-risks at low doses. And this was done. Over and over, one finds variants of the statement that "Radiobiological reasons exist for making this assumption in the absence of direct human data." (Allusions to radiobiology are allusions to evidence from other species and cell-studies, and to hypotheses derived therefrom.)

          But by 1980, when NCRP produced its widely cited risk reduction-factors, human data were no longer absent.

1980 -- What the Record Shows :

          While the NCRP was preparing its 1980 report, its colleagues in the radiation community were concurrently preparing the BEIR-3 Report, with heavy reliance on the A-Bomb Study 1950-1974 (TR-1-77; Bee77; Bee78). For the reasons described in our Chapter 4, this is the key study. With respect to the A-Bomb Study, NCRP and BEIR-3 made five important admissions.

(1) Hiroshima -- Cancer Data :

          As early as 1973, Baum (Baum73) noted that there was evidence in some of the Hiroshima-Nagasaki data for a decreasing slope for cancer-risk per cSv with increasing dose -- just the opposite of what would be expected from some of the animal data. The radiation community suggested that the decreasing slope could be ascribed to cell sterilization (or killing) at very high doses.

          In 1980, NCRP conceded that the effect was not limited to very high doses: "Such an effect may be seen at relatively low doses in the Hiroshima data" (Ncrp80, p.160). In other words, the A-Bomb Study was warning that, in the human, the curve for cancer-risk versus dose might be supra-linear throughout the dose-range -- the opposite of the concave-upward expectation.

          Besides the concave-downward, supra-linear curvature in the Hiroshima dose-response (just mentioned), what were NCRP analysts able to see in 1980 ?

(2) Breast-Cancer Incidence :

          In Ncrp80 (p.144, text and Table 10.3), the NCRP authors also acknowledged that breast-cancer incidence in the A-bomb survivors showed a supra-linear dose-response (a highly significant negative Q-coefficient in an L-Q model).

          As we shall see below, BEIR-3 analysts were also unable to find any support in the A-bomb evidence for the concave-upward hypothesis.

(3) Nagasaki -- Cancer Incidence :

          When the Nagasaki data for cancer-incidence were examined alone, the dose-response for all cancers combined (leukemia omitted) also warned against the use of risk reduction-factors for acute doses: "In the Nagasaki Tumor Registry data, the relationship between the radiation dose and the total incidence of all major cancers except leukemia is highly significant, and the observed dose-response relationship appears linear, with no suggestion of upward curvature" (Beir80, p.181).

(4) Both Cities -- Cancer-Deaths :

          When all cancers (leukemia omitted) in Hiroshima and Nagasaki combined were analysed by BEIR-3, using the LQ-L model (the solitary L for the linear neutron-component), the dose-response was also found to be linear. The Q-term was zero, as shown in Beir80, p.186, Table V-9, "Regression Analyses for LSS Mortality Data, 1955-1974 (excluding Leukemia)."

          In the fine print, one discovers that BEIR-3 constrained the equation so that the quadratic term could not turn out negative. The Q term was "constrained to be nonnegative" (Beir80, p.186). We have noted elsewhere in this book that a negative Q-term means that a linear-quadratic model has a supra-linear (concave-downward) shape. Thus, when the BEIR-3 Committee constrained its equation to produce a nonnegative Q-term, the Committee had decided that a supra-linear dose-response must be ruled out. With the constraint upon it, the quadratic term turned out as zero -- the lowest value it could be, without being negative.

(5) Leukemia Registry Data :

          With respect to dose-response for human leukemia in the A-bomb survivors, Beir80 gave enormous weight (as we shall see, in Part 3) to its concave-upward appearance in Nagasaki in the LSS sample. Because leukemia is only a single cancer among many, the data were exceedingly thin -- especially when Nagasaki was examined alone. There were a total of 46 cases in all Dose-Groups combined.

          RERF analysts (Bee78, p.198) had explicitly warned that, "In the face of the paucity of cases (or deaths) in the Nagasaki LSS sample in the low dose range, and the suggestion that the dose-response pattern for the entire Registry may be different, it would seem best not to invest too heavily in the nonlinear appearance of the LSS data."

          By contrast, the Leukemia Registries for survivors in both cities contained far more data (Beir80, p.341). For Nagasaki, the Registry increased the cases from 46 (in the LSS sample) to 231, and for Hiroshima, from 120 (in the LSS sample) to 323 cases. BEIR-3 itself showed that the Registry data for leukemia were not concave-upward in either Hiroshima or Nagasaki (Beir80, p.343 Figure A-5).

Beside the A-Bomb Study :

          The point is that, when the 1980 NCRP and BEIR-3 reports were issued, direct human data were certainly no longer lacking on the shape of dose-response for malignancies, and none of the data -- except the inappropriate leukemia sample -- supported the predicted concave-upward shape.

          Indeed, NCRP itself described several minor human studies, at low doses, in which dose-response appeared concave-downward (Ncrp80, pp.160-166). By the term "minor studies," we simply mean studies which inherently lack the scientific power of the A-Bomb Study (see Chapter 4). No disparagement of the work is implied.

          And with respect to induction of human breast-cancer by low-LET radiation, there was already a succession of studies additional to the A-Bomb Study. Such studies pointed to a human dose-response which is linear, not concave-upward (Boi77, Boi79, Land80, My69, Sho77. It is proper to describe Land80 as an available analysis, because Land was a member of the BEIR-3 Committee -- see Chapter 37).

          Breast-cancer is one of the two most prominent cancers in women (in the USA), and accounts for about twenty percent of all their cancer-mortality, as already noted in Chapter 21, Part 3. There is every reason to generalize from the breast-cancer data to dose-response for less important cancer-sites, in the absence of any contrary human evidence or contrary logic.

1980 -- The Evidence Was Seen :

          Whether one considers only the A-Bomb Study, or additional studies too, the message from the direct human data was that risk reduction-factors were a mistake which would produce underestimates of cancer-risk at low acute doses. The direct human data were overwhelmingly suggesting linearity or supra-linearity by 1980. And I was not alone in seeing it (Go81). The record above shows that the NCRP and BEIR-3 radiation committees were seeing it too.

1980 -- Statement by NCRP :

          Nonetheless, NCRP's 1980 report based DREFS on animal experiments rather than the available human epidemiology. NCRP explicitly admitted that its method could provide "no rigorously-defensible approach to deriving satisfactory DREFS for the human being" (Ncrp80, p.2):

          "Because of the complexity and wide spectrum of the tumorigenic responses to radiation in the experimental animal, however, there appears to be no rigorously-defensible approach to deriving satisfactory DREFS for the human being, for either single tumor types or for all tumors collectively. Thus, the NCRP is reluctant at this time to go beyond providing a range of factors within which a single factor for the total yield of tumors in man after exposure of the whole body would probably lie. The DREF range is 2 to 10, when the actual absorbed dose is 20 rads or less, or the dose rate is 5 rads per year or less."

          The tone of that statement suggests, to me, that NCRP was not eager to defend its DREFS. And if a lack of enthusiasm was present in NCRP's opening statement, it would have been appropriate, in my opinion. For in 1980 (Go81), I was examining the very same 1950-1974 evidence from the A-Bomb Study which NCRP had examined, and the evidence was indeed badly at variance with NCRP's underlying premise for DREFS -- namely, the premise of a concave-upward dose-response in the human.

1980 -- Statement by BEIR :

          As for BEIR-3, the Committee split in bitter dispute over the shape of dose versus cancer-response in the human, with Harald Rossi arguing against any linear term at all, and for dominance of a quadratic term (Beir80, pp.254-260), and with the chairman, Edward Radford, arguing for a linear model (Beir80, pp.227-253).

          In the end, the 1979 Draft Report was replaced by a compromise (Beir80, p.190), in which the Committee designated the linear-quadratic model (with a positive Q-term) as its preferred basis for making risk-estimates. (Details are in Part 3 of this chapter.) In other words, the human evidence was disregarded. According to Edward Webster, BEIR-3 member:

          "A linear-quadratic dose/effect relationship, defensible in the light of current radiobiologic findings, has been adopted by most of the Committee members as a reasonable basis for prediction of risks of radiation-induced cancer" (Beir80, p.261).

1981 and 1983 -- Reduction-Factors Challenged :

Why Do Experts Disagree ?

          By 1981, I reported that examination of the Hiroshima-Nagasaki evidence seriously pointed to a supra-linear relationship, between cancer-risk and dose of low-LET radiation, throughout the dose-range (Go81). This was the shape which the BEIR-3 Committee had ruled out, by actively constraining its equation (see above).

          The finding in Go81 of supra-linearity, in agreement with Baum's earlier finding, had the additional weight of much more follow-up data since Baum's report. Furthermore, there was no basis for ascribing the supra-linearity to the presumed neutron-exposure at Hiroshima (see Chapter 8, Part 5).

          People in and outside this field often ask, "Why do you experts disagree?" But it is not at all clear that we actually disagree about what the evidence is saying. If the BEIR-3 Committee had not artificially constrained its regression analysis (Table V-9), it probably would have found exactly what I reported from the same data in 1981 (Go81): A supra-linear, concave-downward dose-response for radiation carcinogenesis in the human.

          Of course, both supra-linearity and linearity are incompatible with the use of risk reduction-factors for acute-low and slow-low doses. So the independent analysis in Go81 was a clear challenge to DREFS.

RERF Analysts Challenge DREFS :

          In 1983, Wakabayashi and co-workers at RERF attempted to address the dose-response relationship in A-bomb survivors, by using cancer-incidence data for Nagasaki. This choice was made because neutrons had never been considered prominent in the Nagasaki dose. These workers tried to compare a full quadratic model (Q), a linear-quadratic model (LQ), and a pure linear model (L) for the study of all cancers combined (omitting leukemia) in the Nagasaki cancer-incidence data. We quote the findings made by these workers directly (Waka83, pp.128-129). When they refer to "all cancers except leukemia," they do not mean that leukemia is different; they mean that leukemia is not part of the analysis:

          "The Q model does not fit the incidence data on all cancers except leukemia, whereas the L and L-Q models fit equally well. The linear term is significant in the L-Q model, whereas the quadratic term is not. Thus the linear model appears to be the better for all cancers except leukemia. A similar tendency was observed for several specific sites of cancer, i.e. cancers of the lung, breast, thyroid, and stomach; the Q model either does not fit (for breast cancer) or fits more poorly than the L or L-Q model, and the quadratic term in the L-Q model does not differ from zero (the calculated value is negative). These findings, when compared with the analysis of the fit of these models to cancer-mortality in 1950-1978, where the neutron component was also considered, are seen to to be very similar.

          "Thus it seems reasonable to use the linear model in risk estimation in the present analysis, though we cannot statistically distinguish one model from another among these three alternatives except for cancers other than leukemia and for breast cancer [emphasis added]. In the dissenting section in the BEIR III report, Rossi stated that the dose-response for mortality from all cancers in Nagasaki (1950-1974) fits a quadratic model best. The present analysis does not support this. Rather, the data suggest a linear model (see Radford's comments in the same dissenting section) or at least a linear-quadratic model, which the BEIR III Committee used as the basis of risk estimation." (Parentheses are in the original.)

          Another Clear Warning : So in 1983 -- only three years after the NCRP and BEIR-3 reports of 1980 -- Wakabayashi and co-workers were clearly alerting the radiation community (again) that for all cancers combined, and for breast-cancer specifically, the linear model fit best -- which meant that reduction-factors rested on a fantasy, with respect to the human evidence.

          And in pointing out that the quadratic term in the L-Q model was negative (though not provably significantly so), they were alerting the radiation community (again) that supra-linearity might be the case.

1985 -- Refusal To Abandon DREFS :

          In 1985, the radiation community produced two new reports on radiation risks: Nrc85 and Nih85. Both reports endorsed the use of DREFS in extrapolating from high acute to low acute doses, even though the real-world human evidence was at variance with the presumption on which the DREFS were based.

          Exemptions from DREF-treatment have been made, however, one cancer-site at a time. If human evidence is conclusively against a concave-upward dose-response for a particular kind of cancer, then DREFS are no longer used by the radiation community for that one site. Cancers of the breast and thyroid are examples of such exceptions.

          For instance, the report of the NIH Working Group conceded that, for these two cancers, the linear model fit the data best, but its authors still were clinging to the concave-upward or linear-quadratic fit for all other cancers, with associated DREFS (Nih85, p.iv, p.55). In the quotation below, PC refers to Probability of Causation.

          "In general, the Working Group has sought to use the dose-effect model for each cancer which is most consistent with both the human epidemiological data and the radiobiological data. For leukemia, the data are consistent with a so-called linear-quadratic model; hence this model is the basis for the PC tables calculated for leukemia. This model uses two constants and, in general predicts that small doses of radiation have a lesser effect per rad than do higher doses. There are radiobiological reasons for assuming that a linear-quadratic model is generally applicable to other cancers, which are discussed both in the BEIR III report and in Chapter III of the present report. Accordingly, we have used this approach for all cancers except those of the thyroid and breast. For carcinoma of the breast and thyroid, the data appear to be best described by a simple linear relationship in which the carcinogenic effect of radiation is directly proportional to dose; again the tables are based on this interpretation" (Nih85, p.iv).

An Eight-Point Commentary :

          I consider the statements above to be faulted on several grounds:

(1) Human Evidence Disregarded :

          Notwithstanding its claim of considering human epidemiological evidence as well as radiobiological data, the NIH report appears simply to disregard the findings which show the linear fit to be best for all human cancers combined, in both Beir80 (p.186, Table V-9) and in Waka83. The report also disregards my own 1981 analysis, which was highly suggestive of a supra-linear fit for all human cancers combined, in the A-bomb survivor experience (Go81).

(2) Site-Specific Approach :

          It is scientifically far better to use the findings based on all cancers combined, than to subdivide the observations by single sites of cancer. The NIH report invites error by examining each cancer-site separately. This approach creates the small-numbers problem, except for a very few common cancers. For all the other cancer-sites, when analysts attempt to analyse the sites one at a time, and the numbers pertaining to each are inadequate for reliable analysis, then spurious results are easily obtained.

(3) Leukemia Dose-Response :

          The NIH Report rejected the data in the much larger Hiroshima and Nagasaki Leukemia Registries. As noted earlier in Part 2, those data do not fit the linear-quadratic, concave-upward model best; they fit the linear, or even the supra-linear model best. (See "1980 -- What the Record Shows" above; also Go81). It is interesting to note that, in the A-Bomb Study, the radiation community mounted a massive effort to revise the dose-estimates (DS86), but the community has made no meaningful effort to resolve the relatively minor problems which inhibit full use of the data-rich Leukemia Registries for both cities.

(4) Reasonable Presumption Rejected :

          When all cancers combined (leukemia omitted) show a dose-response which is not concave-upward (Beir80, Go81, Waka83), and when two common types of cancer (breast and thyroid), analyzed separately, show a dose-response which is not concave-upward (Nih85), then the reasonable presumption is that the other cancers would not show a concave-upward dose-response either, if there were enough evidence to be reliable. However, in opting for the site-by-site approach, the NIH Working Group was rejecting the reasonable presumption.

(5) A Question of Consistency :

          Substituting for the reasonable presumption, the NIH Working Group accepted "...radiobiological reasons for assuming that the linear-quadratic model is generally applicable to other cancers..." (Nih85, p.iv). Facing a choice between generalizing from strong, real-world evidence directly from the human, versus generalizing from other species and from their preferred radiobiological hypothesis, the NIH Working Group chose the latter -- and used risk reduction-factors.

          The NIH Working Group appeared to require no suitable human epidemiological evidence in order to embrace a concept (the concave-upward dose-response) which reduced risk-estimates, but appeared to require a mountain of human epidemiological evidence -- extending to each cancer-site separately -- before embracing a concept (the linear dose-response) which would mean higher risk-estimates.

(6) "Not in Peoria!" :

          Refusal by the NIH Working Group, to apply the findings from all cancer-sites combined to the individual cancer-sites, or from cancers of the breast and thyroid to other sites, amounted to the "Not in Peoria" response to evidence -- a response which we explained and criticized earlier (Chapter 21, Part 3). The Peoria approach, with special DREF-exemptions for cancers of the breast and thyroid, is found in additional reports from the radiation community (see Part 3).

          No rational explanation has been offered in such reports for assuming that the shape of dose-response in one cancer would differ from the shape in another cancer -- an assumption which seems particularly irrational when, within the existing evidence, the shape is the same for thyroid-cancer, breast-cancer, and for all cancer-sites combined.

(7) No Demand from Radiobiology :

          As we will show in Chapter 23, there was no reason for the NIH Working Group to have assumed that "radiobiological reasons" demand or even suggest the necessity of a concave-upward dose-response in the human.

          For decades, it had been understood that a linear-quadratic equation can be modified away from a concave-upward shape by an exponential modifier, and that this should be done, if the modifier provides a better fit to actual observations than the equation without such a modifier. Indeed, the 1980 NCRP report itself is on record as recognizing this fact (Ncrp80, p.19, Figure 3.5).

          However, it appears that the various radiation committees did not feel obliged to use the actual human observations. In the next chapter, we shall demonstrate how the linear-quadratic model, even with positive coefficients, can produce a curve which is concave-downward (supra-linear) -- in accord with the the actual observations.

(8) Rejection of Human Evidence :

          The NIH Working Group appears to have rejected the strong human evidence which was at variance with risk reduction-factors.

          I do not object because the NIH Working Group appears to have paid no attention to my own work -- I object because it appears to have paid little attention to anyone's work if that work was in conflict with DREFS.

3.   The Exact Statements
      Which Convert Myth into Consensus

          We call the supposed propriety of using DREFs "myth" because the practice is in conflict with good human evidence. How a myth can become a general consensus may be illuminated by the chronological assembly of exact statements here about the expected concave-upward ("linear-quadratic") dose-response in humans and the consequent use of reduction-factors in risk-estimation. Our phrase above, "general consensus," is adapted from entry #9 below.

          As far as we know, this assembly of exact statements has not been available before.

          Readers will see that some of these reports attempt to justify the use of risk reduction-factors, by prediction from non-human data and radiobiology, and that the rest simply quote the others as justification.

          Although DREFS have spread throughout the literature and will be found at every turn, their basis is the same presumption stated by UNSCEAR and ICRP in 1977 -- a presumption which was invalidated in the same year by the reality-check of direct human data from the 1950-1974 A-Bomb Study (TR-1-77; Bee77).

1.   The UNSCEAR Report of 1977 :

2.   The ICRP Report of 1977 :

3.   The BEIR-3 Report of 1980 :

4.   The NCRP Report of 1980 :
5.   The NIH Report of 1985 :

6.   The NRC Report of 1985, and 1989 Up-Date :

7.   Evans and Co-Workers, 1986,
      New England Journal of Medicine :

8.   UNSCEAR Report of 1986 :

9.   The DOE Report of June 1987 and December 1988 :

10.   Pierce, at OECD Meeting in Paris, October 1987 :

11.   Preston and Pierce in RERF Report TR-9-87 :

12.   UNSCEAR Report of 1988 :

4.   Unwarranted Conclusions
      from the Canadian Fluoroscopy Study

          As we explained in Chapter 21, Part 1, the Canadian Fluoroscopy Study consists of two distinct series: The Nova Scotia women (number 1 in Chapter 21) versus the other-Canadian women (number 4 in Chapter 21). The Nova Scotia series is distinct in at least two ways. First, the total breast-dose accumulated was much higher, and second, the per-rad risk appears higher than in the other-Canadian series.

          UNSCEAR-88, as noted in our Part 3 above, is suggesting that the lower per-rad risk in the other-Canadian series (Howe84) is supportive evidence for a dose-rate effect.

          In at least three separate places, Un88 cites the conclusion by Howe 1984 that the dose-response in the Canadian Fluoroscopy Study is either linear-quadratic or quadratic -- in other words, concave-upward (Un88, p.439, para.241; p.455, para.361; p.456, para.367).

          The next year, November 1989, an up-date of the Canadian Fluoroscopy Study was published (Miller 1989) on which Howe was a co-author. In the up-date, the authors now disagree with the statement which is so important to UNSCEAR-88. In Mi89, the authors state (Mi89, p.1287):

          "... the evidence from Table 2 indicates that the most appropriate form of dose-response relation is a simple linear one, with different slopes for Nova Scotia and the other provinces ... For these models there was no evidence of any upward curvature in the dose-response relation (i.e., the addition of a quadratic term did not significantly improve the fit ...).

          We are not in any position to make an independent evaluation of the dose-response relationship in the Canadian Study. We would need raw data before their reduction, and these data are not published.

          What needs emphasis is that UNSCEAR's statement is now in conflict with the more recent statement by the study's own authors.

          UNSCEAR-88, in the search for human evidence to support its recommendation of risk reduction-factors for slowly delivered doses, suggests that the lower per-rad risk in the other-Canadian series compared with the Nova Scotia series is due to a 20-fold lower dose-rate per exposure for the other-Canadian series:

          "In Nova Scotia, the patients were examined in the anterior-posterior position (facing the x-ray tube) whereas in the other provinces the patients were mainly examined in the reverse position, resulting in doses per fraction about 20 times smaller" (Un88, p.456, para.367).

          Miller, Howe and co-workers make a similar comment in their discussion-section (Mi89, p.1288): "The only substantial difference in the dose-estimation procedures for Nova Scotia and the other provinces was in the proportion of women who faced the x-ray source. This difference is well established, and even varying the proportions substantially does not eliminate the difference in the slopes. One possible biologic reason for this difference is a dose-rate effect. Although the mean numbers of fluoroscopic exposures were similar, the rate per unit dose was more than an order of magnitude greater in Nova Scotia than in the other provinces."

Unwarranted Conclusions :

          Both Un88 and Mi89 are mistaken in their conclusions that a dose-rate difference up to 20-fold exists between the Nova Scotia series and the other-Canadian series. In reality, the biologically relevant rate at which total doses were accumulated in the two series was not even two-fold apart.

          Readers are referred back to Chapter 21, Part 1, Study 3 (Massachusetts Fluoroscopy). There, we showed that the average delivery-rate of the rads in that study -- which is the same as the other-Canadian series, Study 4 -- is about 4.6 rads at one time. The average delivery-rate of the rads in the Nova Scotia series is about 7.5 rads at one time. The ratio of delivery-rates is (7.5 / 4.6), or 1.63 -- far below a factor of 20. This finding is due to the fact that only a very small fraction (about 13 %) of the total rads received in the other-Canadian series was received at the low dose-rate of 0.261 rad per exam.

          Table 21-A provides a convenient way to compare all three fluoroscopy studies, not only in delivery-rate of the rads at one time, but in the tracks-per-nucleus at one time. In the Nova Scotia series, the tracks-per-nucleus are 10.0335 compared with 6.1539 in the other-Canadian series.

          In other words, there is no meaningful difference in dose-rate between the studies.

          Moreover, at such low doses and track-rates, the quadratic term (for inter-track carcinogenesis) is just negligible -- as is generally acknowledged -- and as is illustrated in our Chapter 23, Part 7. Thus, there is not even a basis in principle for invoking a dose-rate effect to explain the different slopes or per-rad risks in the Nova Scotia versus other-Canadian series.

          In any case, we have shown that no appreciable difference in dose-rate exists between the two series. Thus, the Canadian Fluoroscopy Study provides no human evidence supportive of a dose-rate effect.

5.   Unwarranted Conclusions
      from the Holm Radio-Iodine Study

          We are going to give some close attention here to a particular study of patients who received diagnostic radio-iodine, because the study has been recently featured by the 1988 UNSCEAR Committee as important human evidence supportive of a dose-rate effect (Un88, p.34, para.208, and p.459, para.389, and p.491, para.602, 604).

          The study is "Thyroid Cancer after Diagnostic Doses of Iodine-131: A Retrospective Cohort Study," by Holm and eleven co-workers, published in the (U.S.) Journal of the National Cancer Institute, September 21, 1988 (Holm88). A preliminary report on a small fraction of the study-sample was published in 1980 (Holm80a, Holm80b).

          The 1988 report states in its abstract, "Overall, these data provide little proof that I-131 is carcinogenic in humans and support the notion that the carcinogenic potential of internal I-131 beta particles might be as low as four times less than external x rays or gamma rays" (Holm88, p.1132). The same report states in its closing discussion, "... I-131 did not increase thyroid cancer risk in this cohort ..." (Holm88, p.1137).

          This is the message which is used by others as human evidence supporting a safe-dose or at least a greatly reduced risk if exposure is gradual rather than acute. (A single dose of Iodine-131 decays gradually, and does not deliver its total dose to the thyroid all at one instant.)

          As noted above, the 1988 UNSCEAR Committee features the Holm Study as important human evidence supporting the Committee's decision to recommend large risk-reduction factors, for radiation doses which are slowly delivered. Individual authors of the 1988 UNSCEAR Report are acknowledged in our Chapter 37. Lars-Erik Holm is among them.

          Edward Webster, also a member of the 1988 UNSCEAR Committee (and a key member of the BEIR-3 Committee), features the 1980 Holm Study in the course of claiming that the cancer-consequences from Chernobyl will probably be small (see our Chapter 24, Part 9). Webster says (Webs87, p.424): "The effect of protraction [slow delivery of dose] may be the reason why iodine-131 has been judged to be three times less effective as a carcinogen per unit dose than x-rays delivered at high dose rates (Ncrp85, Table 11.3). This conservative judgment was largely based on the investigation by Holm et al (Holm80a) which found no excess thyroid cancer in 10,000 patients who had received gland doses between 58 rem (adults) and 159 rems (persons aged less than 20) after an 18-year follow-up."

          Rosalyn Yalow, a co-author of the 1985 NIH Report on radiation risk (Nih85), also features Holm80a and Holm88 in a 1989 discussion of "radiation phobia" (Ya89, p.160): "Let us consider first what we know about the importance of dose-rate effects in radiation-induced malignancy for any given cumulative dose ... The relevant human evidence depends in part on the use of iodine-131 for diagnosis of thyroid disease and for the treatment of hyperthyroidism. Although only a small fraction of the more than one million patients who had I-131 uptake studies 20 or more years ago and received 50-100 rem thyroidal doses have [sic] been studied, no increase in thyroid cancer has been observed in this group (Holm80a; Holm88). Only 5 % of the more than 35,000 patients evaluated were less than 20 at the time of examination. These authors concluded (Holm80a) that the carcinogenic potential of I-131 would be fourfold less than would result from equivalent externally administered x- or gamma-ray exposure."

Reality -- An Epidemic of Thyroid Cancer :

          In great contrast to the above statements -- which claim that no excess thyroid-cancer occurred in the Holm Study and therefore the slow delivery of dose from iodine-131 must account for this unexpected result -- it turns out (1) that a huge excess of thyroid-cancer occurred in the Holm Study, and (2) that the results of the study have not been clearly revealed.

          Because the huge excess is revealed only indirectly in the 1988 Holm Study (half of one sentence, on page 1134 of Holm88), we had to go through a series of calculations to evaluate it. Before going through the steps with the reader, we must first describe the nature of the study.

Nature of the Holm Study :

          The study-population consists of 38,653 patients (79 % females) who "were examined with diagnostic doses of iodine-131" between 1951 and 1969. These patients were "recruited from seven oncologic centers in Sweden ..." Twenty-nine percent were examined in the period 1951-1959, and 71 % in the period 1960-1969.

          Age at the time of first I-131 examination ranged from one to 74 years, with a mean age of 44 years for the females and 46 years for the males. Only five percent of the total cohort was below age 20 at the time of examination.

          Reason for performing the exam was obtained from the patients' medical records:
          31 % -- suspicion of thyroid tumor.
          42 % -- suspicion of hyperthyroidism.
          16 % -- suspicion of hypothyroidism.
          11 % -- other or unknown reasons.

          One can certainly not assume that this is a study-population which will be just like the general population in risk of thyroid-cancer, except for its radiation-dose from I-131.

          Far from it. People with histories of thyroid abnormalities such as enlarged (hyperplastic) thyroid, goiter, or history of thyroid nodules, go on to show a rate of thyroid-cancer enormously higher than patients without such conditions (Pre87, Table 2; McTier84, p.581; Ron87, p.4). By contrast, the evidence on hyperthyroidism as a risk-factor is inconclusive. McTier84 and Ron87 report finding no basis for calling it a risk-factor. With regard to hypothyroidism, McTier84 shows suggestive evidence in a case-control study that people with hypothyroidism may have a lower risk of thyroid-cancer than people without hypothyroidism (Relative Risk = 0.40 in Table 6, McTier84), but she concludes: "In this study, a history of hypothyroidism was not associated with an altered risk of developing thyroid cancer" (McTier84, p.580).

          In the Holm Study, mean radiation dose to the thyroid from the I-131 was estimated to be about 0.5 Gy or 50 rads (Holm88, p.1136). We independently checked this estimate of mean dose, by starting with microcuries of administered iodine-131, the mean weight of the gland, and the 24-hour uptake. We arrived at an estimate of 52 rads, in very good agreement with the estimate in Holm88. Holm and co-workers note that the distribution of doses was not random among the patients: "... patients who were examined for a suspected thyroid tumor received higher I-131 activities per examination than did others" (Holm88, p.1135). We shall return to this later.

          "The follow-up period lasted from the time of the first I-131 examination until the date of diagnosis of thyroid cancer, the date of emigration or death, or December 31, 1984" (Holm88, p.1134). The mean follow-up was 20 years (p.1134).

          "The cohort was matched against the nationwide Swedish Cancer Register (SCR) to identify malignant thyroid tumors occurring between 1958 and 1984. The SCR was started in 1958 and receives notifications on diagnosed cancers from pathologists/cytologists and clinicians ... The completeness of registration of thyroid cancers is higher than 97 % ..." (Holm88, p.1134).

          Holm and co-workers assume that about 33 % of all the patients "had some sort of thyroid treatment at some time after the I-131 examination" (Holm88, p.1137). They specifically include thyroid surgery and thyroid hormone medication among the likely treatments.

Unabridged Results :

          "Within 5 years of follow-up, each of 136 patients had a thyroid cancer diagnosed, and an additional 3,443 patients died" (Holm88, p.1134). This is the only mention in the entire report of what happened during the first five years after the exposure.

          Beyond 5 years, 50 (total) additional thyroid-cancers were observed by December 31, 1984 (Holm88, p.1134, and Tables 4, 5, 7). Of these 50 additional cases, 34 occurred in the patients whose initial exam was due to suspicion of thyroid cancer. The Holm Study provides no way of knowing what fraction of the 136 early cancers came from this initially suspect group.

          The figures above mean that at least 186 thyroid-cancers (136 + 50) were found in this study-population. We say "at least" for a reason. The number was assuredly greater than 186.

          We must add some cases for the following reason. Holm and co-workers state (Holm88, p.1134) that "Thyroid cancers occurring between 1951 and 1957 could not be identified because of the lack of nationwide incidence data." And from the previous page, we know that 29 % of the 38,653 patients were examined in the 1951-1959 period. This means that all thyroid-cancers occurring in this segment (11,209 patients) before 1958 were missed. It also means that there were fewer than 38,653 patients in the base-population which gave rise to the 136 cases which were not missed during the first five years of the follow-up. We surely will not overestimate total cases if we add only 20 cases to the 136 observed within the first five years of follow-up.

          Thus, a very reasonable approximation is that the number of post-irradiation thyroid cancers observed in 38,653 patients was: 136 + 20 + 50 = 206 cases, during a mean follow-up time (starting with the initial iodine-131 exam) of 20 years.

          Was this an excess?

Observation of a Huge Excess :

          Holm and co-workers say, "The expected numbers of malignant thyroid tumors were calculated by direct standardization; adjustment was made for age- (in 5-yr groups), sex-, and calendar year-specific cancer incidence rates for the whole country obtained from the SCR [Swedish Cancer Register] between 1958 and 1984" (Holm88, p.1134).

          On this basis, they provide 39.4 as the expected number of thyroid cancers during the follow-up beyond the first five years. This expectation applies to the 35,074 persons still in the study as the sixth year begins (of the 38,653 initial patients, 3,443 have died, and 136 with identified thyroid-cancers have been dropped from follow-up). However, Holm and co-workers do not tell what the expected number was during the first five years of follow-up. Therefore, on this crucial issue, we will have to make an estimate on our own. It is surprising that the peer-reviewers of this article did not insist on it.

          As an approximation, we will assume that the expected incidence rate (39.4 cases per 35,074 persons) in the sixth through 20th year of follow-up will also apply to the first through fifth follow-up years. In the U.S., the incidence of thyroid cancer in women is flat from forty years of age onward through 80 years of age (Beir80, p.167), and we will use this for Sweden too.

          A rate of (39.4 cancers per 35,074 persons during 15 years) is an average of (2.63 cases per 35,074 persons each year). Adjusting the cancers for the larger population during the first five years, we have (2.63 cancers) x (38,653 / 35,074), or (2.9 cancers per 38,653 persons each year). Finally, multiplying by 5 years, we have an expectation during the first 5 years of about 14.5 cancers. We can call it 14.

          But the observation during the first 5 years was about 156 cases. The ratio of observed over expected is (156 / 14), which means a rate some 11-fold above normal.

          When we compare the O / E ratio (observed over expected) for the entire 20 years, we have:

          In other words, there is a huge excess of thyroid cancer in the patients who received the diagnostic radio-iodine. Readers of the Holm Study learn nothing about this.

What Became of the Excess ?

          The Holm Study was undertaken by oncology centers to find out if their diagnostic use of radio-iodine is causing excess thyroid-cancer, and if it is, to estimate the magnitude of the elevated risk.

          However, no effort was made to establish the expected rate from a control group having comparable thyroid conditions except for the exposure to Iodine-131. Instead, a predictably inappropriate control group -- the general population -- was used. By comparison with this inappropriate control group, a huge excess of thyroid-cancer was observed in the radio-iodine patients (an excess which the Holm Study does not evaluate or discuss at all).

          This finding is handled in the Holm Study not by throwing out the 31 % of the sample suspected at the outset of thyroid tumor, and not by trying to establish a true expected rate for the remaining 69 % of the study population. The excess is not even mentioned. Instead:

          "In the calculations of person-years at risk, the first 5 years after the initial I-131 administration were excluded for each patient. This was done to reduce the possibility of cancer being present but not diagnosed at the time of the examination and not detected clinically until some years later. All thyroid cancers occurring during the first 5 years after the initial iodine-131 examination were also excluded from the analyses for the same reason" (Holm88, p.1134).

          This approach reduced 206 cancers to 50. Since Holm and co-workers used an expectation of 39.4, they report (50 / 39.4), or 1.27 as the Standardized Incidence Ratio (SIR), with a 95 % confidence interval of 0.94 to 1.67. In other words, with the 5-year exclusion, the excess is not provably different from zero.

          With the 5-year exclusion, the following details are reported:

          In the patients younger than 20 years old during the exam, 2 thyroid cancers were observed. The SIR was 2.02, with a confidence interval of 0.24 to 7.22. In the patients who received radio-iodine because of suspicion of a thyroid tumor, the SIR was 2.77, with a 95 % confidence interval of 1.92 to 3.87. In the initially non-suspect patients, the SIR was 0.62 (0.35 to 1.00).

          The Holm Study also explores the effect of throwing away the first ten years of results. This reduces the 206 cancers to 27, and the overall SIR to 0.93. Of the 27 remaining cases, 19 are in the group examined because of suspicion of thyroid tumor, and their SIR is 2.17.

A Fatally Flawed Study :

The Control Group :

          The Holm Study relies on a control-group which may supply utterly inappropriate expected values, both for the initially suspect group and for the initially non-suspect group of thyroid patients. If the expected values are inappropriate, this would make all the risk ratios (Standard Incidence Ratios) and all the conclusions therefrom misleading, at best.

          It seems self-evident that no one can possibly know, from the Holm Study, how much of the observed excess cancer in the initially suspect group is due to the radio-iodine and how much is due to the patients' pre-radio-iodine condition. To say that none of the excess is due to the radiation would require some evidence. To say that all of the excess is due to radiation would certainly be foolish, too, in view of the higher risk of thyroid-cancer among patients suspected of thyroid-cancer.

          Nor does the Holm Study provide a basis for confidence that the general population is an appropriate control group for the initially non-suspect group. On the contrary. The finding by Holm88 (p.1135) that the risk-ratio is only 0.62, after the 5-year exclusion, strongly suggests that the "natural" risk (meaning, without radio-iodine experience) of thyroid-cancer in this special population of people with thyroid disorders, may be a lot lower than the natural risk in the general population.

Treatments Post-Radio-Iodine :

          Moreover, no one can evaluate the impact, on the study's outcome, of post-radio-iodine treatments (including thyroid removal and thyroid hormones) received by an estimated one-third of the 35,000 patients in the study-population. UNSCEAR 1986, referring to the smaller Holm Study of 1980, rejects such post-radio-iodine treatments as a likely contributor to the study's presumed deficit of cancers (Un86, p.229, para.397). In 1988, Holm and co-workers dismiss this problem with a single sentence: "The absence of any increased thyroid cancer risk was considered not to be ascribable to the thyroid treatment" (Holm88, p.1137). Their allusion to the absence of excess risk is, of course, to an absence after the first five years of results have been thrown away. Then the risk ratio becomes 1.27, which is not provably different from 1.00 under the circumstances.

Diseases in the Studied Organ :

          It is interesting that the 1986 UNSCEAR Report, in listing several reasons for the claimed shortage of excess cancers in the overall preliminary (1980) results, points out that "... the subjects are a selected unhealthy population, with a high percentage of thyroid involvement, to whom specific rates of thyroid cancer induction, valid in the general population, may not apply" (Un86, p.229, para.397).

          Such insights about the fatal flaws of the Holm Study seem to be discarded, however, in the 1988 UNSCEAR Report (Un88). Un88 relies heavily on the Holm Study on the key issue of risk reduction-factors.

Unknown Latency Period :

          Another confounding variable in the Holm Study, not mentioned by UNSCEAR, is the real possibility that thyroid diseases themselves alter the latency period for radiation-induced cancer. For instance, one or another condition might induce promotional agents not present in healthy thyroid cells, and the peak incidence of radiation-induced cancer might occur 3, 5, or 8 years post-irradiation. Such early peaking is well-observed for radiation-induced leukemia, about 7.5 years after exposure.

Pre-Judgments versus Inquiry :

          Throwing out the observed excess, in 5-year or 10-year stages, is no solution whatsoever to these very serious confounding variables. Throwing away any part of such a follow-up reflects an unwarranted pre-judgment, not a scientific inquiry, in our opinion.

          We remind readers that the Holm Study is examining a study-population which -- during the first five years of follow-up -- showed an 11-fold excess of the exact variable (thyroid-cancer) which the investigators were hoping to study (see "Unabridged Results," above).

          As an independent analyst, I cannot just pretend to myself that this is a normal population showing normal behavior during a latency period, and that if I throw away these startling results, I can tell myself that I have a normal population entering the sixth year of a radiation follow-up study. It would really require some supernatural omniscience on my part to decide that truth would be best served by not mentioning the 11-fold excess and by throwing away the first five years of results.

          Nonetheless, I am unaware that any other analysts, peer-reviewers, or radiation committees have (1) asked for an explanation of the 3.9-fold higher rate of thyroid-cancer in the exposed group (unabridged results), or (2) challenged use of the general population as a control group for this very abnormal study-population, or (3) challenged the failure even to divulge just how very abnormal the study-population is -- e.g., an 11-fold excess rate of thyroid-cancer during the first five years of follow-up, or (4) challenged the throwing out of the first five years of the results. What I think I see, so far, is an uncritical rush to embrace this fatally flawed study with its welcome (welcome to me, also) but unwarranted conclusions.

Consistency in Standards ?

          Radio-iodine studies have also tested consistency regarding reliance on human versus non-human evidence. This chapter has shown, with regard to risk reduction-factors (DREFS), how much of the radiation community greatly prefers to generalize from the non-human evidence than to generalize from the human evidence.

          However, in 1982, Lee and co-workers published a rat-study in which they found no lesser carcinogenicity of slow doses from iodine-131 compared with acute doses from 250 kVp X-rays, and the study extended down to thyroid doses of 80 rads (Lee82). Clearly, this finding is not as welcome as lesser carcinogenicity from iodine-131 would be.

          The Lee Study was challenged as follows by Holm88 (p.1135):

          "Iodine-131 has frequently been used to induce tumors in experimental animals, although its effectiveness relative to external photon exposures has been studied only to a limited extent. Earlier studies with high doses to the thyroid gland suggested that iodine-131 was one-tenth to one-fourth as effective as x rays in producing thyroid tumors. Lee et al. observed that with lower doses the difference in effectiveness between the two types of radiation was less pronounced and perhaps even the same at doses less than 3-4 Gy. [We are splitting the Holm88 paragraph here.]

          "Like many other experiments on animals, their results are limited by the fact that iodine-131 is an efficient cancer inducer in certain animal species and strains only, such as the CBA mice and the Long-Evans rats. Lee et al. used female Long-Evans rats in their study, and the results may well have differed had they used male rats or a mixture of the two sexes. Regardless, there is a great deal of uncertainty in extrapolation from animal data to human populations. Epidemiologic data are therefore the preferred source of information for obtaining risk estimates in humans" (Holm88, p.1135).

          We agree. But we would say "appropriate epidemiologic data."

          Earlier, in 1985, NCRP commented on Lee82 in a different manner (Ncrp85, p.33): "For the production of thyroid carcinomas, the two radiation types appeared to be of equal effectiveness at all three doses although the results did not preclude a relative effectiveness of iodine-131 of as little as one-third compared to external radiation."

          If one is going to discuss the confidence-limits on a best estimate, it is certainly not an appropriate practice to mention only the lower limit. Yet NCRP does not mention that the Lee82 findings are also consistent with a higher risk from the slow exposure than from the acute exposure. There appears to be asymmetry in the NCRP approach to the Lee82 Study.

          UNSCEAR-86 shows the two dose-response curves from Lee82 practically superimposed on each other (and both looking supra-linear, not concave-upward), and comments (Un86, p.208, para.254): "Thus, there was no difference in the effectiveness of the two radiations over the observed range of doses, but a lower effectiveness of iodine-131 per unit dose (up to a factor of about 3) could not be excluded on statistical grounds (Ncrp85)." Thus, UNSCEAR-86 passes along the NCRP comment without any criticism of its asymmetry.

          To help restore symmetry, we repeat: The best estimate from Lee82 does not support DREFS and is also consistent with higher risk at slow-low doses.

Looking at the Initially Non-Suspect Group :

          An obvious question with respect to the Holm Study is: What would this study have shown if the "initially suspect" 31 % of the study-population had never been included?

A Substantial Excess of Thyroid-Cancer :

          Because Holm88 does not report what fraction of the early, discarded cancers occurred in the initially suspect group, and what fraction occurred in the intially non-suspect group, the question cannot be answered with certainty. Indeed, the data do not exist at all for 1951-1957.

          We can explore an answer by making an approximation. Of the total 50 cancers observed after the 5-year exclusion, 16 cancers occurred in the initially non-suspect group. The fraction was (16 / 50), or 0.32. And with the 10-year exclusion, no meaningful change occurred: The fraction was 8 cancers out of 27 total cancers, or 0.30. We shall use the approximation that the fraction which occurred during the 5-year exclusion was the same as the fraction which occurred afterwards: 0.32.

          Since we estimated (see "Unabridged Results") that at least 156 thyroid-cancers occurred in the total study-sample during the first five years of follow-up, we would approximate that (0.32) x (156 cancers), or 50 cancers came from the initially non-suspect group. Beyond five years, another 16 thyroid-cancers occurred in this group, so the estimated total of observed thyroid-cancers would be 66.

          What is the expectation, without radio-iodine, if the rate in this diseased group is comparable with the rate in the general population? We are using the same big "if" used (without discussion) by the Holm Study.

          We showed above ("Unabridged Results") that the expectation in the full cohort during the first five years of follow-up was 14 cases. Since the initially non-suspect group represents 69 % of the total study-population, its expection is (0.69) x (14 cancers) = 9.66 cancers during the first five years of follow-up. For the follow-up beyond five years, its expection is (0.69) x (39.4 cancers) = 27.19 cancers during the rest of the follow-up. Total expectation, if radio-iodine had no effect, would be (9.66 + 27.19) = 36.85 cancers. And if the "natural" rate of thyroid-cancer in this special group is lower than in the general population, the expectation would also be lower than 36.85 cancers.

          So this approach suggests that the relative risk, of observed cancers over expected cancers, might be (66 / 36.85) = 1.79 if the first five years of follow-up were included. This is about three-fold higher than the value of 0.62, reported in the Holm Study with the five-year exclusion for the initially non-suspect group. And if the appropriate expected value is even lower than 36.85, then the risk ratio would be higher than 1.79.

          We cannot know. We are only pointing out a good basis for thinking that the missing data on the first five years of follow-up might transform a "no provable excess" report into a highly significant excess. And this excess might even be wholly due to the radio-iodine administered. Only a pre-judgment would allow a claim that it was not.

Presence of a Dose-Response Trend :

          One issue which the Holm Study may be capable of addressing is the issue of dose-response. On this issue, it does not matter whether the risk ratios are correct or incorrect (correct meaning that they compare rates in two groups which are alike in risk, except for their radiation dose). What matters is how the risk ratios change (if they do) with rising dose.

          Within the results which Holm and co-workers do report, there is a basis for thinking that there is a strong dose-response trend in the initially non-suspect group.

          Holm and co-workers, in their Table 5, divided the entire sample of 35,000 patients into three dose-levels as follows:

                                     Risk Ratio:  
Microcuries         Observed            SIR       
of I-131             Cancers      (Obs / Expected)
   <30                   14            0.96       
 30-74                   19            1.15       
   >74                   17            2.04       

   All                   50            1.27       

          Their Table 6 shows the comparable entries for just the patients who were examined for a suspected tumor:

                                     Risk Ratio:  
Microcuries         Observed            SIR       
of I-131             Cancers      (Obs / Expected)
   <30                   12            3.69       
 30-74                   11            2.06       
   >74                   11            2.96       

   All                   34            2.77       

          In Table 5, the total group (35,000) shows evidence of a dose-response trend, toward an increasing incidence of thyroid-cancer with increasing dose of radio-iodine. Holm and co-workers say (p.1135), "The thyroid cancer risk increased with increasing administered I-131 activity (Table 5)." In Table 6, by contrast, the initially suspect group by itself shows no evidence of a trend: "There was no relation between SIR and administered activity of I-131 (Table 6)."

          This means that the study's inclusion of the initially suspect group is tending to dilute and to conceal a positive dose-response trend in the initially non-suspect patients. Their dose-response trend must be even stronger than indicated in Table 5 -- where it is clear despite dilution by the initially suspect patients. Rising incidence with rising dose is powerful supportive evidence for causality, of course. It is regrettable that Holm and co-workers chose not to evaluate the dose-response trend for the initially non-suspect patients by themselves, in the same way that these authors evaluated the initially suspect patients by themselves.

No Evidence of a Dose-Rate Effect :

          We shall continue our exploration of the Holm Study as if patients who were initially suspected of a thyroid tumor had never been included, and as if only the 69 % who had thyroid disorders (but were initially not suspected of a tumor) were in it.

          Now we shall ask if there is any indication of a dose-rate effect (for instance, reduced risk from slow exposure compared with acute exposure) in this group, when we make no pre-judgments -- which means that we look at the entire follow-up.

          Readers who proceed step-by-step, through the two analyses which follow, will see for themselves that there is no evidence at all for a dose-rate effect from slow versus acute exposure.

Search for a Dose-Rate Effect :

          In our work above ("A Substantial Excess"), we estimated 66 observed thyroid-cancers versus 36.9 expected. The difference, or radiogenic excess, is 29.1 thyroid-cancers.

          Is this a smaller excess than we would expect, if we use expectations based on observations from acute exposure?

          We are going to answer this first in the way which we think is the scientifically best way, and then in the way used by Holm88.

Avoiding the Site-Specific Pitfall :

          One of the most questionable practices in this field is the excessive subdivision of data, including undue reliance on site-specific risk-coefficients. (See our Index entries "Scientifically questionable practices" and "Site-specific analysis.")

          Even a casual inspection of Studies 1 through 9, in Chapter 21, Part 1, demonstrates the delusion of thinking that reliable risk coefficients (K-values) can be directly determined for specific sites of cancer.

          Let us consider a K-value of 0.02, which is equivalent to a 2 % increase in spontaneous risk per rad. If K = 0.02, a dose of 50 rads causes a 100 % increment above the spontaneous expectation. The dose which adds as much cancer as the spontaneous rate is commonly called the doubling dose, so when K = 0.02, the doubling dose is 50 rads. In short, a dose which doubles the spontaneous rate is one doubling dose, and a dose which triples the spontaneous rate represents two doubling doses.

          Now, we can inspect the breast-cancer doubling doses in Chapter 21, Part 1. The doubling dose in Study 3 (Massachusetts Fluoroscopy) was 150 rads, and the doubling dose in Study 7 (the British Luminizers) was about 80 rads -- even though the medical X-rays have a higher Relative Biological Effectiveness than the gamma rays from radium-226. So there is perhaps a 4-fold difference. (We cannot agree with any analyst who casually says that breast-cancer risk looks similar from one site-specific analysis to the next.)

          A large range for the doubling dose occurs also in the in-utero studies of Chapter 21, Part 1. Even if we narrow the range by saying that in Study 5 (the Stewart Studies), a half rad causes a 50 % increment instead of a 94 % increment in childhood cancer, this choice would make the doubling dose 1 rad. By contrast, in Study 6 (the MacMahon Study), 0.9 rad provoked a 40 % increment, so (0.9 rad x 2.5) or 2.25 rads would provoke a 100 % increment -- a doubling. Thus there is more than a 2-fold difference in the magnitude of doubling dose derived from these studies of a single cancer-class (childhood cancer).

          It is perfectly valid to use such studies to test the hypothesis of flawless repair. As long as a significant excess of radiogenic cancer occurs, the excess is evidence that repair was not flawless. The exact risk coefficient or doubling dose is irrelevant for such a test.

          But it is a very different matter indeed when analysts attempt to use studies of specific kinds of cancers (such as childhood cancers, breast-cancers, thyroid-cancers) to test for an effect of slow versus acute exposure upon the magnitude of risk, when the magnitude of the acute effect is so poorly known for single sites and classes of cancer. Such attempts just invite large errors, in my opinion.

          I do not think site-specific studies are suitable for a dose-rate analysis, but if such analysis is done nonetheless, then I think there will be less likelihood of large errors, if analysts use the very reasonable approximation that all types of cancer have about the same fractional increase in their spontaneous rate per rad, if all other variables are held constant. Until and unless appropriate evidence develops which shows otherwise, I would regard myself as skating on scientifically "thin ice" not to make this approximation. (Go81, Chap.10; Go85, pp.19-20.)

All-Cancer K-Value from Our Table 15-L :

          If we use the approximation that the thyroid gland is no more and no less radio-sensitive than other organs, we must use the all-cancer K-value from our Table 15-L in order to calculate the radiogenic expectation from 50 thyroid-rads in the Holm Study (initially non-suspect group).

          Since 79 % of the Holm-Study patients were women, with a mean age of 44 years, and since the dose-rate per day was far below 50 rads, and since the DS86 dosimetry is only supplemental in the A-Bomb Study, we shall use the low-dose K-value of 0.00615 from the T65DR dosimetry.

          The average energy of beta particles from iodine-131 is about 189 KeV (Strom58). This may mean a somewhat higher RBE than A-bomb radiation, but we shall not raise the K-value directly. Instead, we shall use just the female K-value (which is higher than the male K-value), and we shall use 50 rads as the thyroid dose, even though Holm88 states that the initially non-suspect group received a lower average dose. (Holm88 does not say how much lower.)

Results by Our Method :

          The estimate is easy to make:

Expected number of cancers without I-131 = 36.9
Thyroid-dose = 50 rads.
K = 0.00615 per rad from acute exposure.
Estimated radiation-induced cancers = (36.9) x (50) x (0.00615) = 11.35 cases.
Radiation-induced plus spontaneous cancers = 11.35 + 36.9 = 48.25 cases.

          So do these calculations from the Holm Study suggest that the radiation-risk from radio-iodine would be 3-fold lower than the risk from acute thyroid exposure?

          Not at all. The estimated observed cases in this sample are 66 cancers, based on reasonable approximations (see "A Substantial Excess" above). This is a higher number than 48.25 cases expected on the basis of acute exposure.

          The number 66 is consistent even with a 3-fold higher K-value from the radio-iodine than from the A-bomb radiation. With a 3-fold higher K-value, the radiation-induced cases would grow to (3 x 11.35) or 34.05 cases. So 34.05 radiation-induced cases plus 36.9 spontaneous cases would mean an observation of 70.95 cases -- still in good agreement with an estimated observation of 66 cases.

          Readers who have followed this, step-by-step, can judge for themselves whether our approximations are reasonable or not.

          We are certainly not claiming that this comparison, using the all-cancer K-value from the A-Bomb Study, means that slow dose-rate from iodine-131 is three-fold more carcinogenic than acute dose-rate. We have tried to make it clear that we think the Holm Study is completely inappropriate for addressing the issue at all.

          But, because of the weight given to the Holm Study by UNSCEAR 1988 and others, we have been obliged to point out that the Holm Study is consistent with exactly the opposite conclusions from the ones ascribed to it.

Analysis by the Holm Model :

          We are not quite finished, because we promised that we would search for a dose-rate effect by using the Holm model too, although we think it is not a good model for such a purpose.

          By "Holm model," we mean the use of a site-specific K-value, rather than an all-cancer K-value, to compute the radiogenic expectation of thyroid-cancer in this study. Holm and co-workers say (p.1136) that they used the site-specific K-value for thyroid from the 1985 NIH Report. The NIH Report (Nih85) in turn used thyroid incidence data from the A-Bomb Survivors, Hiroshima plus Nagasaki, 1958-1979 (Nih85, p.255). This would have meant no correction for the very large overestimate of neutron-dose at Hiroshima.

          Perhaps because of the neutron-error for Hiroshima, the 1988 UNSCEAR Report (Un88, p.434, para.209) explicitly recommends the thyroid-cancer incidence data from Nagasaki alone as "the best," for which Un88 cites Wakabayashi and co-workers (Waka83). Nagasaki data never had a neutron problem. On the other hand, subdivision of the cities reduces the cases and thus increases uncertainty in the estimates. In any case, we should find out if, and how much, the site-specific K-value differs in Nih85 versus Waka83.

Checking the Site-Specific K-Value :

K-Value Based on Wakabayashi :

          The Wakabayashi et al analysis divides the Nagasaki A-bomb survivors into two classes: Unexposed survivors and survivors receiving 100 kerma rads and more. On this basis, these analysts report on the relative risk (100+ rads versus zero rads) as follows in their Appendix Table 2:

RR = 1.70 for all cancers (leukemia omitted).
          Excess RR = 0.70 .
RR = 3.23 for thyroid cancers.
          Excess RR = 2.23 .

          The ratio of excess relative risk (2.23 / 0.70 = 3.186) is for the same kerma dose, but not for the same absorbed dose in the organs from which the cancers arose. Site-specific analysis requires site-specific body transmission-factors (Chapter 8, Part 2). The body transmission-factor for thyroid is estimated at 0.7 in TR-12-87 (Shi87, p.43), which is higher than the factor for colon (see our Table 9-A).

          We can proceed by establishing the kerma dose to which the excess relative risks apply. Using our Table 9-B, we calculated the weighted mean dose received by the Waka83 exposed class (Dose-Groups 5,6,7,8) as 243.7997 kerma rads.

          So, for all cancers, excess RR = (0.7 per 243.7997 kerma rads) = 0.002871 excess per kerma rad (or a K-value of 0.002871 per kerma rad).

          But for equal kerma rads, thyroid is 3.186 times more sensitive than all organs combined (if you take site-specific analysis seriously). So the K-value for thyroid is (3.186 x 0.002871), or 0.009147 per kerma rad.

          But the thyroid's absorbed dose was lower than its kerma dose, so the K-value will be higher than 0.009147. It needs adjustment for the site-specific transmission-factor of 0.7 . So we divide (0.009147 / 0.7), and we obtain a site-specific K-value for the thyroid of 0.013067 per absorbed rad. This is the same as an excess relative risk of 0.013067 per thyroid-rad.

          The value of 0.013067 arises from a population with an average age at the time of bombing of about 27. The value might be adjusted downward to apply to the older study-population in Holm88, but we will simply compare it, as it is, to the site-specific K-value from Nih85.

K-Value from Nih85 :

          In the 1985 NIH Report, Table X-12 (p.261) provides the following values for "Relative Excess by Exposure Age and Sex" per thyroid-rad:

          The Holm Study (Table 1) has a female to male ratio of 3.8. If we say m = the male fraction, then 3.8m is the female fraction, and 4.8m = 1. Therefore m = 0.2083. And (1-m) or 0.7917 is the female fraction. So the weighted K-value for the overall Holm Study would be (0.0176 x 0.7917) + (0.00935 x 0.2083) = 0.01588 .

Results by the Holm Model :

          We shall use both site-specific K-values to compute the radiogenic expectation in the initially non-suspect group. The radiogenic expectation is (the spontaneous expectation of 36.9 cancers) x (site-specific K-value per rad) x (50 rads -- which is an exaggeration for this group).

          With the K-value of 0.013067, based on Waka83, the radiogenic expectation = (36.9) x (0.013067) x (50) = 24.1 radiation-induced cancers. The spontaneous expectation (36.9 cancers) plus the radiogenic cancers (24.1) = 61.01 cases. And the estimated observed number was 66 cancers. So there is no indication of any risk-reduction from the slow delivery from iodine compared with acute delivery from A-bomb radiation here.

          With the K-value of 0.01588, based on Nih85, the radiogenic expectation = (36.9) x (0.01588) x (50) = 29.3 radiation-induced cancers. The spontaneous expectation (36.9 cancers) plus the radiogenic cancers (29.3) = 66.2 cases. And the estimated observed number was 66 cancers. So there is no indication of any risk-reduction from the slow delivery from iodine compared with acute delivery from A-bomb radiation here.

          For those who would say, "We want to look only at the period beyond 5 years," we say the following:

          One must avoid distorting the outcome by pre-judgments which are totally unwarranted. Just what does anyone know about when radiation-induced cancers will occur following radio-iodine in a group of manifestly abnormal people with diseased thyroids? If you see a large excess of thyroid cancers in the initially non-suspect group during the early follow-up, it needs explaining. There would be no basis whatsoever for simply claiming that an early excess (if any occurred here) could not have been caused by the radiation.

Summary on Unwarranted
Conclusions from the Holm Study :

          We wish to emphasize a point. Our exploration of what the Holm Study might have shown, if the 31 % of initially suspect patients had never been included, is not a statement by us that we think the initially non-suspect group is an appropriate group to compare with the general population. Far from it, as we already indicated above (see "A Fatally Flawed Study"). The general population appears to be an unsuitable control-group for both the initially suspect and initially non-suspect study-samples.

          We have shown our reasons for saying that (A) the Holm Study in its present state is consistent with opposite conclusions about dose-rate, and (B) no one should regard the Holm Study in its present state as meaningful about anything concerned with DREFS.

          In other words, we disagree with its acceptance by the 1988 UNSCEAR Committee as a piece of notable human evidence in support of a dose-rate effect and risk-reduction factors.

          Perhaps the Holm Study illustrates the fact that the peer-review system can perform unevenly. For instance, reviewers can be ultra-careful about the choice of control-groups for the in-utero studies (see Chapter 21, Part 1), and yet overlook glaring problems with the control group in a study like the Holm Study.

6.   The Bottom Line

          1.   For over a decade, the radiation community has been using risk reduction-factors to make its estimates of cancer-risk at acute-low doses and at slow-low doses. These reduction-factors are based on the premise that dose versus cancer-response is concave-upward -- in other words, on the premise that the risk per rad (cGy) is smaller when dose is either acute-low or slow-low than when dose is high. This premise was explicitly stated in 1977 by both UNSCEAR and ICRP (see Part 3, above), and has been echoed again and again by other radiation committees. Of course, if dose-response is either linear or supra-linear, it would be a mistake to use risk reduction-factors, because they would produce underestimates of risk at both acute-low doses and at slow-low doses. The inappropriate use of reduction-factors with respect to Chernobyl-induced cancers is illustrated in Chapter 24, Part 7.

          2.   Since 1977 (TR-1-77, or Bee77), human epidemiological evidence has repeatedly shown that the premise of risk reduction-factors (the premise of a concave-upward dose-response in humans for radiation carcinogenesis) is fundamentally flawed. And the record shows that the radiation committees knew it by 1980 (see Part 2, above).

          3.   Nonetheless, from 1977 through mid-1989, almost all of the radiation community has subordinated the human evidence against using risk reduction-factors, in favor of using such factors on the basis of non-human evidence and cell studies -- "radiobiology." I do not disparage radiobiological evidence, and we should learn all that we can from such work. But in science, when predictions from radiobiology are invalidated by the reality-check of direct human evidence, the direct evidence must prevail. This chapter shows that, for years, it has not.

          Perhaps it will. In 1988, Warren Sinclair, president of the NCRP, conceded that in the A-Bomb Study 1950-1982, "... it appears that the dose-effect response is fitted about as well by a linear as by a linear-quadratic equation, and this may also influence risk estimates ..." (Sin88, p.154). And in 1988, Albrecht Kellerer (see Chapter 37) offered his opinion -- after studying the A-bomb survivors through 1985 -- that "Today, the use of a reduction factor in extrapolation from high doses to low doses which are relevant for radiation protection purposes, is less easily defensible ... Although even the extreme hypothesis remains unfalsifiable, that at the lowest doses there is no excess cancer incidence, a prudent extrapolation can nevertheless make use of a linear extrapolation and can drop the assumption of a reduction factor" (Kelle88, p.51; translated from the German by Dr. Rudi Nussbaum).

          Such statements are hedged. Moreover, they are competing with vigorous pressure in the opposite direction from some other members of the radiation community, who are pressing for the ultimate reduction-factor -- namely, for treating low doses as safe, and excluding them completely from risk-estimates (see Chapters 24 and 25).

          4.   The use of risk reduction-factors has meant that, for years, most radiation reports have been presenting linear estimates as the "upper limit" on risk, despite human evidence showing that linear estimates represent either the best values or a lower-limit of risk.

          5.   There is no longer any need to extrapolate from acute high-doses above 100 rads (100 cSv), in order to make risk-estimates at acute-low or at slow-low doses. The A-Bomb Study has already provided direct evidence at low doses for all cancers combined (see Chapter 13), and it will continue to do so, provided its legitimacy as a credible, prospective study is maintained (as proposed in Chapter 6).

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