Radiation Protection Science - NRPB flannels about epidemiology

NRPB's head of epidemiology lets the "averaging" cat out of the bag
NRPB's Director has sent us a "Note"
in response to Busby and Scott Cato's paper on infant leukaemia after Chernobyl,
which shows ICRP's risk factors are out
by a factor of at least 100.

Written by NRPB's head of epidemiology, Colin Muirhead, it has not been published, as far as we know (it isn't on their web site),
but correspondence with NRPB reveals that it is being used as their official line on the Busby and Scott Cato paper.
One is entitled to ask whether it has been peer reviewed. Busby and Scott Cato were.

The most notable thing about it is that it admits the invalidity of the averaging inherent in the Linear No Threshold model. This is a fundamental shift in NRPB's logic, though Muirhead himself doesn't appear to realise the implications.

Dr. Muirhead's Note is on the left of the following table. LLRC's comments are on the right.

At the bottom of the page is the Note as it was sent to us.

Dr. Muirhead's Note:
Leukaemia after Chernobyl

LLRC's comments

1. Correlation studies, based on aggregated data, can be subject to bias and confounding of a type not seen in cohort or case-control studies, in which information is collected for specific individuals.
 
He is saying, we must suppose, that there is a confounding factor - [another cause of the disease]. But what? Population mixing? Too much washing? These babies were foetuses or eggs when the reactor blew, so what gave them leukaemia, if it wasn't the radiation?
If exposures vary within geographical areas, as was the case within Wales and Scotland following the Chernobyl accident, it is not possible to tell using correlation studies whether the persons who developed the disease in question received higher exposures than the population as a whole.
 
 
 
 
 
 
 
This is crucial, and Muirhead has put himself at odds with the very basis of the NRPB model.
The LNT assumption is that if you know the collective dose to the population you can calculate the leukaemia yield -- variations in the distribution of dose should not make any difference, because people getting more than the average dose are compensated for by others getting less.
In complaining that you cannot predict the numbers of leukemias because you can't tell whether people with leukaemia had higher than average exposures Muirhead has accepted that the Linear No Threshold model is not valid.
Confused? see this link for help.
Correlation studies can sometimes be useful in looking for trends. However, epidemiologists generally place more reliance on quantitative risk estimates derived from cohort or case-control studies, rather than from correlation studies.
 
 
 
 
  irrelevant
2. The results of individual correlation studies can be variable, not only because of possible bias or confounding, but also because of small numbers in some instances ...
 
  Not small numbers in this instance - the numbers in the at risk population are huge.
... and the possibility of chance findings when sub-groups of the data are analysed....
 
 
 
 
 
 
 
 
 
 
Again, not chance findings in this case. The p value expresses the probability of the observations being due to chance. For Wales and Scotland p = .0001 - that is a probability of 1 in 10,000 that is was a chance occurrence.
Multiplying the p-values in each of the separate countries and studies gives an overall p-value less than 0.0000000001 - 10 billion to one against.
Muirhead must know there is no "small number" problem because the statistical method used is designed to analyse rare events in a very large population.
... For example, there did not appear to be a prior reason for expecting that any raised risk of childhood leukaemia following the Chernobyl accident would be concentrated in infancy - as suggested by Petridou et al (1996) from their study in Greece - rather than being spread across a wider age range (Darby and Roman, 1996).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Well, we're not saying risk would be concentrated in infancy. We have found evidence of increased risk in older children as well. See this link for the figures showing that after the accident children up the age of 5 in Wales suffered a 41% increase in lymphoid leukaemia incidence, and a 27.5% increase in all malignancies. In Scotland the same age group had a sudden jump in the same diseases after 1987 and rates remained higher than before - 50% up in the case of lymphoid leukaemia; 20% in the case of all malignancies.
NRPB was told about this in 1996. Why don't they listen?
The Darby and Roman article which Muirhead cites discussed Petridou et al's 1996 paper on infant leukaemia in Greece. Darby and Roman said Corroborating evidence is needed, for example from other groups of children exposed in utero to different levels of environmental radiation ... The causes of childhood leukaemia are likely to prove many and complicated. The more data and hypotheses we have ... the better.
and Muirhead himself has written (ref.): Greatest weight should be placed on results which are consistent across several studies ...
The Busby and Scott Cato cites studies from five countries. Isn't this corroboration enough?
Studies on infant leukaemia have also been conducted in Germany and Belarus.
In Germany, there was an indication of a raised risk that, if anything, was greatest in the areas with the lowest contamination

 
 
 

 
 
 Muirhead has forgotten what he wrote in his first paragraph and has clicked back into the Linear No Threshold assumption.
Data from many studies support an alternative model where the dose response curve is bi-phasic. The observation in Germany is consistent with this.
In Belarus, the relative risk of infant leukaemia was not statistically significantly different from 1, and was lower than the corresponding values in Greece and Germany, in spite of the higher contamination
 
 
And again! The bi-phasic dose response curve (which, incidentally is derived from observations, not guesswork) implies that as contamination levels increase the organisms tend to die rather than accruing mutations. This means the more highly contaminated foetuses didn't survive to develop leukaemia - they died in the womb.
More generally, childhood leukaemia rates within Belarus do not appear to be raised
 
 
 
 
So what? In lots of places the rates for infants were raised. Muirhead cannot explain away an observation by reference to the Linear No-Threshold dose/response model which, for the low dose region we are discussing here, is a highly dubious hypothesis.
3. The problems with small numbers and variability in findings from individual studies ...
alleged problems. The numbers are not small.
... can be addressed by performing a combined analysis of different datasets, using the same methodology. The European Childhood Leukaemia-Lymphoma Study (ECLIS) has monitored trends in rates of these diseases since the Chernobyl accident, based on data from 36 European cancer registries, including data from Scotland and Wales (Parkin et al, 1996). To allow for differences in baseline rates between the various regions studied, information on temporal variation in rates within each region was pooled across regions, rather than pooling the rates directly.
 
 
 
 
 
 
 
 
This combining technique is not appropriate - it smudges out variations in dose and genetic factors in individual populations. See this link for what the Busby and Scott Cato paper says about it.
It is clear that individual exposures from the Chernobyl accident varied within regions. However, whereas Busby and Scott Cato (1998) reported results for the whole of Scotland and the whole of Wales, ECLIS analysed data from south-west Scotland and north Wales separately from the remainder of Scotland and Wales, to reflect differences in average exposures between these areas. .. it should be Busby and Scott Cato (2000) - [i.e. it is published in the peer reviewed literature]. Apart from that petty slight, it's not clear what point Muirhead is making. If he cares to explain, we'll put an appropriate comment in here.
ECLIS showed a slight increase in the overall rate of childhood leukaemia during the period 1987-91, but this trend was similar to that in the period before the accident. Furthermore, the temporal variation in rates was not related to the geographical pattern of exposure, using either a one or a two-year latent period for leukaemia.
 
 
  Again, if this is an argument, it is one that depends on the validity of LNT.
With regard to the comment of Busby and Scott Cato (1998)! on possible effects of latency, epidemiological data do not indicate that the latent period for leukaemia varies by dose following exposure in early life (UNSCEAR, 1994, 2000).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
There's plenty of evidence of latency in cancers of all kinds, including BEIR V cited in the Busby and Scott Cato paper - see for example Wings of Death pp, 220, 247, and 259. [and have a look at this note on latency]
The post Chernobyl leukaemia thesis does not hang on this point, and the latency period is, in any case, only one of the fudges inherent in Parkin's paper.
It's interesting that Muirhead does not take Darby and Roman's line, in criticising Petridou (see above) for
taking no account of the fact that exposures from Chernobyl lasted several years, so that children born after the period defined as "exposed" would also have received some Chernobyl exposure.
Nobody really knows whether the problem is lag in expression or dose protraction, but the fact remains that there is a definite effect definitely associated in time with Chernobyl, at very high statistical significance; it's at least 100 times greater than the effect predicted by ICRP's risk factors and demonstrates a dose response curve confirmed by other studies.
This is the grim silence of fact that remains after all the wriggling and whingeing.
4. As well as correlation studies of leukaemia following Chernobyl, the results of cohort and case-control studies have been reviewed recently by UNSCEAR (2000). These latter studies are based on workers who took part in the clean-up of Chernobyl following the accident.
 
 
 
 
  High doses to adults, not low doses to infants. What's this got to do with it?
There are methodological problems with some of the studies that have reported findings to date, reflecting small numbers ...
 
  The numbers in Busby and Scott Cato are huge
... the lack of an adequate comparison group ...
 
 
 
 
The comparison group in Busby and Scott Cato is highly adequate, consisting of infants in the same age groups in the same countries, who differed from the study group by not being in the womb or about to be conceived at the time of Chernobyl.
... and/or uncertainties in dose estimates. The UK dose estimates were worked out by NRPB
Nevertheless, the more reliable of these studies ...
i.e. those that don't embarrass NRPB
... have not demonstrated an association between leukaemia risk and radiation exposure at Chernobyl (UNSCEAR, 2000). Attempts are ongoing to obtain more definitive information on risks among these workers.
 
 
 
Workers whose main dose was external, not babies with mainly internal doses.
5. To conclude, whilst research on this topic is continuing, overall the current evidence for increased leukaemia risks associated with environmental or occupational exposures from the Chernobyl accident is not convincing.
 
So is he saying that these babies didn't have leukaemia?
If he accepts that they did have it, what does he say caused it? ...
and has he read Sir Richard Doll's recent comments on the Nordic leukaemia study?


The argument about NRPB's "averaging" assumptions is hard to grasp.
You may find
this link helpful.

Leukaemia after Chernobyl

This note addresses issues relating to studies of leukaemia after the Chernobyl accident.

1. Correlation studies, based on aggregated data, can be subject to bias and confounding of a type not seen in cohort or case-control studies, in which information is collected for specific individuals. If exposures vary within geographical areas, as was the case within Wales and Scotland following the Chernobyl accident, it is not possible to tell using correlation studies whether the persons who developed the disease in question received higher exposures than the population as a whole. Correlation studies can sometimes be useful in looking for trends. However, epidemiologists generally place more reliance on quantitative risk estimates derived from cohort or case-control studies, rather than from correlation studies.

2. The results of individual correlation studies can be variable, not only because of possible bias or confounding, but also because of small numbers in some instances and the possibility of chance findings when sub-groups of the data are analysed. For example, there did not appear to be a prior reason for expecting that any raised risk of childhood leukaemia following the Chernobyl accident would be concentrated in infancy - as suggested by Petridou et al (1996) from their study in Greece - rather than being spread across a wider age range (Darby and Roman, 1996). Studies on infant leukaemia have also been conducted in Germany and Belarus. In Germany, there was an indication of a raised risk that, if anything, was greatest in the areas with the lowest contamination (Steiner et al, 1998). In Belarus, the relative risk of infant leukaemia was not statistically significantly different from 1, and was lower than the corresponding values in Greece and Germany, in spite of the higher contamination (Ivanov et al, 1998). More generally, childhood leukaemia rates within Belarus do not appear to be raised (Ivanov et al, 1996).

3. The problems with small numbers and variability in findings from individual studies can be addressed by performing a combined analysis of different datasets, using the same methodology. The European Childhood Leukaemia-Lymphoma Study (ECLIS) has monitored trends in rates of these diseases since the Chernobyl accident, based on data from 36 European cancer registries, including data from Scotland and Wales (Parkin et al, 1996). To allow for differences in baseline rates between the various regions studied, information on temporal variation in rates within each region was pooled across regions, rather than pooling the rates directly. It is clear that individual exposures from the Chernobyl accident varied within regions. However, whereas Busby and Scott Cato (1998) reported results for the whole of Scotland and the whole of Wales, ECLIS analysed data from south-west Scotland and north Wales separately from the remainder of Scotland and Wales, to reflect differences in average exposures between these areas.
ECLIS showed a slight increase in the overall rate of childhood leukaemia during the period 1987-91, but this trend was similar to that in the period before the accident. Furthermore, the temporal variation in rates was not related to the geographical pattern of exposure, using either a one or a two-year latent period for leukaemia. With regard to the comment of Busby and Scott Cato (1998) on possible effects of latency, epidemiological data do not indicate that the latent period for leukaemia varies by dose following exposure in early life (UNSCEAR, 1994, 2000).

4. As well as correlation studies of leukaemia following Chernobyl, the results of cohort and case-control studies have been reviewed recently by UNSCEAR (2000). These latter studies are based on workers who took part in the clean-up of Chernobyl following the accident. There are methodological problems with some of the studies that have reported findings to date, reflecting small numbers, the lack of an adequate comparison group and/or uncertainties in dose estimates. Nevertheless, the more reliable of these studies have not demonstrated an association between leukaemia risk and radiation exposure at Chernobyl (UNSCEAR, 2000). Attempts are ongoing to obtain more definitive information on risks among these workers.

5. To conclude, whilst research on this topic is continuing, overall the current evidence for increased leukaemia risks associated with environmental or occupational exposures from the Chernobyl accident is not convincing.

Colin Muirhead
9 January 2001

References

Busby, C and Scott Cato, M. Quantifying the error in risk estimates. Increases in leukaemia in infants in Wales and Scotland following Chernobyl: evidence for errors in statutory risk estimates. Aberystwyth, Wales: Green Audit, Occasional Papers No. 98/2, June 1998.[The correct citation is Energy and Environment Vol. 11 2000, No. 2 127-139 ]

Darby, S C and Roman, E. Links in childhood leukaemia. Nature, 382, 303-304 (1996).

Ivanov E P, Tolochko G V, Shuvaeva L P, Becker S, Nekolla E and Kellerer A M. Childhood leukemia in Belarus before and after the Chernobyl accident. Radiat Environ Biophys, 35,75-80 (1996).

Ivanov, E, Tolochko, G V, Shuvaeva, L P, et al. Infant leukemia in Belarus after the Chernobyl accident. Radiat Environ Biophys, 37, 53-55 (1998).

Parkin, D M, Clayton, D, Black, R J, et al. Childhood leukaemia in Europe after Chernobyl: five year follow-up. Br J Cancer, 73, 1006-1012 (1996).

Petridou, E, Trichopoulos, D, Dessypris, N, et al. Infant leukaemia after in utero exposure to radiation from Chernobyl. Nature, 382, 352-353 (1996).

Steiner, M, Burkart, W, Grosche, B, Kaletsch, U and Michaelis, J. Trends in infant leukaemia in West Germany in relation to in utero exposure due to the Chernobyl accident. Radiat Environ Biophys, 37, 87-93 (1998).

UNSCEAR (United Nations Scientific and Committee on the Effects of Atomic Radiation). Sources and Effects of Ionizing Radiation. 1994 Report to the General Assembly, with Scientific Annexes. New York, United Nations (1994).

UNSCEAR (United Nations Scientific and Committee on the Effects of Atomic Radiation). Sources and Effects of Ionizing Radiation. 2000 Report to the General Assembly, with Scientific Annexes. New York, United Nations (2000).


Muirhead's observations on Latency:
It is reasonable to ask how much epidemiological evidence exists that is really informative on this. But whatever it may reveal, there are theoretical grounds for expecting a lag in expression, proportional to dose. The larger the number of radioactive atoms / particles in the environment the greater is the number of cells bearing precancerous mutations. A large population of such cells increases the probability of a second mutation leading to the expression of the disease, simply by increasing the size of the target. This increased probability will be expressed as earlier onset at higher doses. Armitage and Doll spotted this decades ago. Note: it is the onset that will be proportional to dose, not the yield of clinical diagnoses.

If you are seeing this page full screen (i.e. without a navigation bar on the left) you can't see how the rest of the site is organised.
This Home page link takes you to the index page, which has links to all the topics we discuss on the site [only use it if this page is full screen]


Send email to:
SiteManager@llrc.org with questions or comments about this web site.

This page was last updated May 2001