[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Gofman's models tested
Hi Mike,
This a great scientific criticism of Gofman's model, but it a more wonderful
read. This is from the '73 ANL report. The lit ref is a proceedings article
from an IAEA Conference in 1976 on effects of natural background radiation
(its in BEIR III - arbitrarily trashed). Note that even though the AEC
undertook this in response to the court case on Clavert Cliffs, when they got
it they suppressed it even though there was no substantive scientific
criticism of the analysis (beyond the general conclusion that the data was
'course'), but stopped work on producing a more detailed (eg, county level)
analysis. It was up to Bernie Cohen to do on his own, work at the county level
for the more specific issue of radon and lung cancer, that the US government
has refused to perform for 25 years. Note the rigorous treatment of the
meaning of statistics in addressing the physical world represented by this
kind of rigorous analysis. Let me know if you need refs, or the report - it's
not large.
Regards, Jim
Dr. Norman Frigerio, Dr. Keith Eckerman, and Ralph Stowe reported (1973) that:
"In late 1971 the Environmental Statement Project was formed at Argonne
National Laboratory to aid the U. S. Atomic Energy Commission in the
preparation of environmental statements for the nuclear facilities being
licensed by the Commission. Since methodology for assessing the impact of such
facilities was in its infancy, it was necessary for us to develop methods and
programs adequate to the task. ... This report, ARIP I, concerns itself with
evaluation of the carcinogenic hazard that might be associated with the
radiation and radioactivity from nuclear facilities."
"(T)he hypothesis has been advanced that a significant fraction of human
cancer mortality may be due to the human radiation background. (Pauling 1958,
Gofman 1971, Tamplin 1971, ICRP 1969, ANL 1970, BEIR 1972, Hutchison 1972). For
a normalized irradiation of 170 millirem/yr, these authors have estimated U. S.
cancer mortality excesses of about 3,000 to 100,000 per year, i.e., about 1% to
30% of current experience. Since the identification of so important an etilogic
factor would be an event of major significance in the field of cancer
epidemiology, we addressed ourselves to the examination (Frigerio, Eckerman and
Stowe in print) of the degree to which these hypotheses could be justified from
current vital statistics and from the known variations in the radiation
background.
"This examination occupied a fair span of time, during which the mists of
our comprehension cleared only slowly... we present is less in logical than in
chronological order. We hope thus to indicate how it was possible for us to
begin with the presumption that background radiation must be carcinogenic only
to be forced, after something very much like the classic Drunkard's Walk,
(Schreider 1966) to conclude that it is not."
In consideration of the additive linear model, r = ro + kD, Frigerio states:
"There must be roughly equal distribution of radiogenic malignant mortalities
among the 56 Mn types. At k = 3.2 and D = 5.1, this would lead to an excess
national mortality of about 33,600/yr, i.e., about 10% of current experience.
Spread over 56 Mn types, this corresponds to r ~ 0.3 per Mn type, i,e,, an
expectation of about 54 radiogenic deaths per million population per Mn type for
the 18-year period of observation. In order to run the gamut of proposed
expectations, we first identified populations presenting at rates below 0.3 and
0.03, discriminated by sex, race, Mn type, and state of residence. The hope was
that we would thus bracket some rate which the linear additive model would be
easily tenable."
"Although in cancer epidemiology, one would not usually consider
expectations less than 5 or so, (Brownlee 1960) much less express them as
decimals, this has been the practice in radiation carcinogenesis studies.(ICRP
1969; BEIR 72) Thus, we have allowed this practice at r = 0.3 or 0.03. At r =
0.003 we have used only a plus sign (+) to indicate that the value for t is
mathematically real but less than 1.
"With so many Mn sites violating'the requirements of the model, even as
judged simply by the normal "t" test, we had to admit that it was
extraordinarily improbable, at least at these levels. So, we continued dropping
our search value for r until, at 0.006, all of the observations went to zero
except for those three stalwarts, ICD 151, 153 and 171. Since we thought we
might have something here, we did our estimations on the basis of r = 0.003, the
mid-value of the interval, rather than 0.006, its upper bound.
"In this range level, the normal "t" test becomes awkward.(Brownlee, 1960;
Kendall, 1958) Thus, we resorted to the much more powerful, albeit lengthy and
expensive, Monte Carlo method. (Shreider, 1966) Briefly the population of the
U. S. was subjected to a random "rain" of radiocarcinogenci deaths at r = 0.003
for 100 18-year periods, and the results of each period analyzed as above.
...Ergo, not only is the null hypothesis (r = 0.003) improbable, but the Monte
Carlo results suggest that a level of roughly 0.003/20 would be needed to reach
even a 63% confidence level. This corresponds to about 16 deaths/yr per
200,000,000 population, or about 0.005% of current U. S. mortality.
"In any case the model certainly seemed untenable at any level much
greater than
r =0.003/20=1.5 x 10^-4, at least as its authors originally presented it. In
theory, we thought it might be saved ...confining higher expectations to 10 of
the 56 types, even though these 10 corresponded poorly to those for which human
radio-carcinogenesis had been shown. (BEIR 72) In practice, though, even this
turned out to be improbable when we examined this unholy decade..."
"As an aside, we might note that this sort of epidemiological approach
possesses some peculiar advantages over less direct studies of radiation
effects. For one thing it addresses itself directly to the population of
interest, in this case that of the U. S. rather than to small. select
populations of the war-torn (e.g., Hiroshima and Nagasaki), of the ill (e.g. ,
those irradiated for spondylitis, tuberculosis, thyroiditis, malignancies,
thymic disorders, etc.), of the young (e.g., irradiation for tinea), or of the
occupationally stressed (e.g., uranium miners). It may be noted, for example
that even the smallest population groups in Table 1 are as large and usually
much larger than these select irradiated groups.
"Then, too, the time span of observation is large. Although we have deal
only with an 18-year span here, the data can quite properly be regarded as an
18-year sampling of a continuing procession of cohorts which span the full
biblical 'three score years and ten'. And, of course, the radiation is being
delivered over this entire time span, at the very rates of interest, an
compounded, so to speak, for effects in utero, on the young, on the general
population, and on the aged.
"While each malignancy type at r = 0.003 is within its respective
statistical expectation, the consistent string of zero observed deaths, over
populations that range from tens of thousands to nearly ten million, is a bit
unsettling."
"Despite this heterogeneity, some socioeconomic bias exists in the U. S.
data. Many of the 10 apocalyptic horsemen.., who refused to show significant
groups below r = 0.3, appear less formidable when viewed in the light of
worldwide malignancy mortalities. .. (T)he fact that cancer is a reportable
disease in the Scandanavian countries (Ringertz 1971) with their remarkably
complete registries, estiminates 163 entirely and produces some very low rates
for 174 and for some of the ICD types included in 65...(T)he very low rates
observed help to dispel the enticing, if mildly parochial, notion that the
remaining 7 types somehow constitute "common" malignancies, while the 46
eliminatedble are 'rare'. While these 7 do account for perhaps a third of
total U. S. malignant mortality, (Burbank 1971) they hardly constitute
important malignancies in other lands. Indeed, the only one of these to
coincide with the BEIR list ( BEIR 72) of important radiogenic malignancies,
ICD 151 [stomach], is dropping so linearly and rapidly in the U. S. that it
bids fair to reach zero within the coming two decades. (Burbank 1971; Cowdry
1968)
"If these are indeed valid, a goodly fraction of the total radiogenic
insult must have been received by age 10 and a significant number of radiogenic
mortalities should have appeared by age 30. However, this does not seem to be
the case, ...Here we have isolated those national rates for which we had
age-specific data, and for which R = O up to age 30 or beyond. Again, if these
ten horsemen were truly riding to the beat of a radiogenic drum, they were
certainly riding more slowly than predicted by the linear additive models so far
proposed.
"All in all, then, it appears that even the abandonment of
polycarcinogenesis would do little for the additive model, especially in the
long run, and this model will probably have to be abandoned in toto."
In presenting the multiplicative model, r = ro + ro(D/DD):
"Admittedly, radiation at low dose rates does seem to be remarkably
ineffective as a complete pancarcinogen, or even as a complete carcinogen of any
sort. But it could well be a pan-co-carcinogen, precisely as envisioned by the
multiplicative model.
"If this were the case, one would predict a fair increase of malignant
mortality with increasing background, and this prediction has been made quite
explicit by the model's authors, (Gofman 1971; Tamplin 1971) e.g., from 1% to
30% increase at 170 mrem/yr, depending on various assumptions of latency,
plateau, and doubling dose.(Gofman 1971; Tamplin 1971; BEIR 1972)
"With this in mind it was intriguing to note, ...the resolute insistence
on dwelling in regions of high background that seemed to characterize the low
mortality groups. At r = 0.03 and 0.003 only six groups were at the 170 mrem/yr
national average, none were below the average, and at least 40 were above 180
mrem/yr. At first we thought this might only be a secondary association with the
well-known urban trend of U. S. cancer mortality. (MacDonald 1967; Grahn 1971)
Tests failed to substantiate this, however. A white female resident of Dallas,
for example (140 mrem/yr), simply seems to be about twice as likely to contract
leukemia as her counterpart in Denver (290 mrem/yr). Since we doubted that
anyone was prepared to ascribe oncolytic properties to the radiation background,
we felt obliged to search for some other association. Surely there must be some
sort of mortality increase with increasing background. (Gofman 1971; Tamplin
1971; BEIR 72)
"However, plots of U. S. rates for white, malignant mortality (Burbank
1971) against natural background for the 50 states showed, if anything, the
reverse e.g., Figure 1. Now, were it not for the insistence of the hypothesis
(Gofman 1971; Tamplin 1971) that there must be a correlation between malignant
mortality and background, we would be inclined to dismiss Figure 1 as an example
of simple noncorrelation. (Neyman 1972). However, of the 14 states above 140
mrem/yr, 12 were very significantly (P < 0.01) below the U. S. average, one
insignificantly lower, and only one slightly, but significantly, higher. The
probability of this occurring by pure chance proved to be <0.001. Similar
results were obtained with an independent estimate of natural backgrounds.
(Oakley 1972).
"Several features of Figure 1 might be worth noting. First of all, some
states at common background had rates identical to the third significant figure,
so that some of the single points actually represent pairs.
"Secondly, no error bars are shown because the standard errors are less
than the size of the points. The data base is, literally, enormous . Each point
represents an average of about 10^5 deaths, and a coefficient of variability, V,
of about 0.3%.... it is evident that the vertical dispersion displayed is not
"scatter", at least not in the usual sense. Rather, it reflects the operation of
the genetic, cultural, socioeconomic and other environmental factors so well
known in the epidemiology of malignancy. [multiple references]"
"Finally, in addition to the seeming negative correlation of rate with
background, the ten lowest states in the U. S. all lay at backgrounds >135
mrem/yr. Thus, there seemed to be some real, if hidden, association between
high backgrounds and low malignant mortalities. Although a similar and even more
dramatic effect was noted in the non-white population, we confined ourselves to
the white population because of its greater homogeneity, better statistics, the
better availability of socioeconomic data, etc. ( Vital Statistics 1950-1968;
Statistical Abstracts 1950-1972)
"For purposes of further comparison, we discriminated three groups: A, the
seven states of natural background above 165 mrem/yr; B, the fourteen states of
natural background above 140 mrem/yr; C, the fourteen states with the lowest
backgrounds. These were compared with all 50 U. S. states, (Vital statistics
1950-1968; Statistical Abstracts 1950-1972).
"We first analyzed the 50 states for each of the 56 (Mn) types to see if
the low mortalities of groups A and B could be due to particularly low rates for
a few types. These two groups, however, proved to be lower in all categories
than the U. S. average, and this premise had to be discarded. A summary is
presented in lines 3-7 of Table 4. The rates for all categories, in fact, tended
to decrease with increasing background.
Table 4. U. S. tow and High Background White Populations, 1950-1967
No. Characteristic A B U. S. C
1 Natural background, mrem/yr 210 170 130 118
2 White population, thousands 5735 16,897 158,051
59,683
3 r, Mn 140-159 42.9 45.6 52.4 50.3
4 r, Mn 160-164 15.8 16.9 22.3
23.4
5 r, Mn 170-181 36.8 38.2 41.5 40.1
6 r, Mn 190-205 30.8 31.5 33.3
33.0
7 r, All malignancies 126.3 132.2 149.5
146.8
8 Residence altitude, ft 4510 2650 900 730
9 Urbanization, % 63 57 69 74
10 Per capita personal income,$ 2021 1922 2215 2255
11 Median family income, $ 5600 5400 5660 5650
12 Physicians/1000 population 1.27 1.25 1.49
1.49
13 Hospital beds/1000 population 8.24 8.82 9.49 8.70
14 Median years of school completed 11.8 11.7 10.9
10.8
15 Poor diet households,% 16.5 21.2 19.1
19.1
16 Population on Federal Food Assist, % 2.6 3.2 3.2
2.5
17 Unemployment,% 4.3 3.9 3.9
3.3
18 Accepted, Military Selective Service 65 63 56 53
19 Life expectancy, male 67.7 67.7 67.6 67.5
20 Life expectancy, female 74.5 74.7 74.2 74.3
21 Urban air, particulates, ugm/m3 129 119 115 116
22 Urban air, benzene soluble, ugm/mJ 10.1 9.3 9.5
9.6
23 Urban air, radioactivity, pCi/m3 8.5 7.7 6.8 6.3
24 Urban air, beta, pCi/m3 5.5 5.2 4.4
4.2
25 r, Mn 140-205, age 0-9 8.11 8.31 8.54 8.31
26 r, Mn 140-205, age 10-19 6.80 6.61 6.82
6.72
27 r, Mn 140-205, age 20-29 10.46 10.73 11.09
11.19
28 r, Mn 140-205, age 30-39 27.61 28.39 31.45
32.27
29 Mortality rate, all causes 892.0 893.2 928.5
903.9
30 U. S-group, all causes 36.5 35.2 -
24.6
31 U. S-group, malignancy 23.2 17.3 - 2.7
32 r, Stomach, 151 11.7 11.6 11.8
11.0
33 r, All G. Z., 150-159 40.7 43.0 49.0
46.7
34 r, Lung, 163-164 14.5 15.5 20.4
21.5
35 r, Breast, female, 170 21.5 22.6 25.3
24.4
36 r, Thyroid, 194 0.055 0.054 0.057 0.054
37 r, Bone, 196 0.92 1.03 1.12
1.07
38 r, Leukemia, 204 7.03 7.23 7.13
6.91
"However , plots of U. S. rates for white, malignant mortality (Burbank
1971) against natural background for the 50 states showed, if anything, the
reverse e.g., Figure 1. Now, were it not for the insistence of the
hypothesis
(Gofman 1971; Tamplin 1971) that there must be a correlation between malignant
mortality and background, we would be inclined to dismiss Figure 1 as an example
of simple noncorrelation. (Neyman 1972). However, of the 14 states above 140
mrem/yr, 12 were very significantly (P < 0.01) below the U. S. average, one
insignificantly lower, and only one slightly, but significantly, higher. The
probability of this occurring by pure chance proved to be <0.001. Similar
results were obtained with an independent estimate of natural backgrounds.
(Oakley 1972).
[ FIGURE 1 ]
"Several features of Figure 1 might be worth noting. First of all, some
states at common background had rates identical to the third significant figure,
so that some of the single points actually represent pairs.
"Secondly, no error bars are shown because the standard errors are less
than the size of the points. The data base is, literally, enormous . Each point
represents an average of about 105 deaths, and a coefficient of variability, V,
of about 0.3%."
"Certain possibilities, for example, were ruled out by the nature of the
observed mortality pattern. Thus, if the decedent populations of groups A or B
above were to contain significantly large numbers of immigrants from other parts
of the U.S. , (i.e. , the decedents had not been exposed to the high backgrounds
until late in life), one would have expected the rates in groups A and B to be
higher than those of the remaining states . This because the Mn rates of the
remaining states are much higher than those of A or B, e.g., 150.4 for the
U.S.-minus-A, and 151.6 for the U.S.-minus-B. Instead, the reverse was true.
Accordingly, if short-term residents are a factor, the true rates for the
long-term residents must be even lower than those given in Table 4." [Analysis
of "competing risks", longevity and age, radiographic exposures, radiation rate,
are considered here, with no effect on the conclusions.]
"Regardless of which of the suggested values (Gofman 1971; Tamplin 1971;
BEIR 1972) we used for D or DD, V invariably increased, i.e., the results were
always the opposite of what would have been expected if the model represented a
real factor in U. S. malignant mortality. Furthermore, this increase in V was
found to hold for essentially all U.S. malignancies, even for leukemia, the
classic of radiogenic malignancies Thus we seemed to be left without
statistical support for a multiplicative model, either for all malignancies
(pancarcinogenesis), or even for specific ones."
Note: Dr. Frigerio, with Eckerman and Stowe, then goes on to consider "other
models" and "future models" relative addressing the implications of the data
and analysis.