Six-fold error in risk factors

NRPB epidemiologist screws up

Here we re-analyse radiation workers' cancer statistics
to reveal the true healthy worker effect -
and a six-fold error in risk factors at the 30 milliSievert level

The second analysis of the National Registry of Radiation Workers by NRPB's Colin Muirhead was a major new analysis of the risk of working with radiation.

The NRPB press release of 15 March 1999 focused mainly on the confirmation of the earlier 1992 finding of a strong ‘healthy worker effect’ (HWE). Apart from this finding, the main conclusion of the report was that the increases in cancer found in radiation workers

are consistent with those risks predicted by other bodies
also consistent with rationale for lower occupational dose limits.
However, new data is provided on the radiation workers in this report that we can use to rework the figures given and show that the real effects are much greater than those calculated.

The Healthy Worker effect

Deaths from all causes and also from cancer are related to factors such as wealth and poverty, social class, childhood environment, and so forth. It is not therefore surprising that workers in the nuclear industry, who are generally of a higher educational and skill level, are certainly given better wages and healthcare, and are subjected to ‘rigorous medical entry standards’, should have lower rates of death than national average rates, which include the unemployed and low waged, disabled, uneducated, disadvantaged and others predisposed to poor health.

This is why studies of nuclear workers are difficult to analyse for health effects of low level radiation. Clearly the national rates used in the report are not suitable as controls against which radiation effects can be tested. It is like exposing a very healthy man to radiation which causes him to die of cancer at the age of 65 but then saying that the radiation did not have an effect because he lived longer than the average lifespan of AIDs sufferers in Africa. The question is, how can we factor out the healthiness of the nuclear worker and examine the effect, if any, of radiation exposure?

In this report, NRPB use a method that assumes that there is a simple straight-line relation between exposure and cancer incidence down to the smallest dose. All radiation studies show that this is not actually the case. Despite this, the results show significantly high Standardised Mortality Ratios (SMR) for cancer of pleura (193), uterus (129) and thyroid (180). These SMRs are ratios of rates in radiation workers to rates in the general population multiplied by 100. But there is a way that we can discover what the baseline SMR for the healthy nuclear worker is.

Healthy radiation worker control

In the first year of radiation work, the worker has received little exposure. As time continues, exposure accumulates and its genetic damage effect adds up, causing increases in illness, particularly cancer. Thus a graph of the increase in SMR with years of exposure should show the effect of this cumulative genetic damage. In the report, data is given that enables this graph to be constructed.
(12KB) Variation of Standardised Mortality Ratio (SMR) with period of time exposed to radiation as a consequence of working in the nuclear industry. Note the shape of the curve. This result is for men, who make up most of the sample, but the curve for women is almost exactly the same.
It is immediately clear that the SMR for the new employee can be obtained from this graph by extrapolating the curve to zero time. We have used a figure of 57 on the basis of a polynomial regression fit to the data. This is the SMR of the healthy worker and this value can then be used as a control to establish the effect of radiation exposure. The results obtained show an enormous and highly significant excess of most cancer types and also increased general mortality. We have calculated some SMRs and P values based on Chi-squared statistics below. The new SMRHWs are obtained using the tabulated observed deaths in the report but multiplying the Expected number by 57/100 to give a new Expected number based on a Healthy Worker Control whose SMR is 57.

Site SMR
Chi2 P<
All cancer 144 475 0.000
Oesophagus 137 12 0.001
Stomach 149 52 0.000
Lung 120 57 0.000
Pleura 338 70 0.000
Bone 131 0.75 NS
Melanoma 198 21 0.001
F. breast 1.33 61 0.000
Uterus 175 4.8 0.05
Prostate 167 60 0.000
Testis 209 13 0.001
Thyroid 267 12.5 0.001
NHL 175 33 0.001
All leukaemia 164 22.7 0.001

Error in ICRP risk factors

Use of the Healthy Worker as control enables us to calculate the number of fatal cancers predicted by the ICRP risk factor of 0.05 per Sievert. The average dose to the whole study population of 124,752 workers was given as 30.5mSv. The predicted number of radiation induced fatal cancers is thus 0.05 x 0.0305 x 124,752 = 190.25

Using the SMRHW of 57 we obtained by polynomial regression we can adjust the Expected Numbers of fatal cancers calculated in the report to allow for the healthy worker effect. We find that 4396 x 0.57 = 2505 deaths from cancer would be expected in the radiation worker population if they had not received any exposure. The number of observed deaths is given as 3598, a difference of 1093. This is just under six times that predicted from ICRP risk factors, and this therefore represents an error in these models of that magnitude.

External and internal

Of course, the 30.5mSv is an external film-badge dose. It is clear from The Inside Story (in Radioactive Times Vol. 3 No.1) that irradiation from internal isotopes may carry a far greater risk. Also, for individual cancer sites, the error factor calculated above goes up by an order of magnitude or more.

There is no justification in Muirhead et al. using the data presented in this report in the way they have. It is quite clear that the magnitude of the healthy worker effect is the key to establishing the true effects of the doses. Why did they not use their data as we have?

Wiggly dose response

Instead, they use a method that assumes a linear relation between risk and dose. They then confirm or deny their hypothesis based on whether a statistical measure of trend shows that there is such a linear relation. This is poor epidemiology and poor science, since they assume their result within their hypothesis. What they find is that as dose increases, there is an initial sharp increase in cancer which then falls and then rises again. This kind of wiggly or biphasic dose response has been found elsewhere, including within the Hiroshima cohort results and was discussed in RaT 2(1) p.11 and also by Chris Busby at the STOA workshop of the European Parliament. It is a result of subpopulations of high sensitivity cells. It may also be caused by sub-populations of high sensitivity individuals.

The referees of the Journal of Radiological Protection should have pointed this out.

Constable Weird comments on HWE (5KB) Constable Weird comments some more on HWE (5KB)

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]
Use the Health Effects of low level radiation button to see what else we have to say on this topic.

Send email to: with questions or comments about this web site.

This page was last updated May 2001