Linear Model

["Will Atkinson (B364 X4370)" <will.atkinson@aea.orgn.uk>]

Just got back from vacation to find Jim Muckerheide's report of
Professor Pollycove's critique of the IARC Lancet paper waiting for me.
 Ploughed through all the rest of the email confident that someone
would have corrected all the mistakes.  Since noone has, here goes:

>... Wade Patterson writes:
>
>> 4. The paper by the IARC study group on cancer risk among nuclear industry
>> workers, Lancet, 344:1039;1994 fits the data to a linear model and therefor
>> cannot be used as an argument for the validity of the linear model. (not
>> unless you believe one can pull himself up by tugging on his bootstraps.) ...

The IARC study fits data to a linear model for a low dose population
and compares the excess relative risk (ERR) per Sv with that obtained
from a high dose population (the Bomb survivors).  If risk was non
linear over the whole dose range and the studies had sufficient power
we would expect the values to be significantly different.  This is a
perfectly valid approach.  In fact the IARC ERRs are not significantly
different from the Bomb survivors, but the study team don't make much
of that because of the low power of the study.

> ...  First, in this combined occupational exposure group it chooses
>to ignore the most accurate data, the Nuclear Shipyard Worker
>Study compared to the early weapons facility workers with the
>questionable dosimetry and confounding factors in their exposure
>data... 

The IARC study is not a review paper.  It combined data donated to it
by the Nuclear Industries of three countries most of which was not from
early weapons facility workers.  Considerable efforts were made to
standardise the dosimetry over time and between employers.

>..., one-sided tests are
>presented throughout." This states that they explicitly ignore
>all contrary data, even within the context of such statitical 
>small-numbers results...

I would question the use of one-sided tests myself, but it certainly
doesn't mean that they ignored all contrary data in the sense implied
by the quote below:-

>... Since only
>positive data are allowed to be considered, ONLY THE DATA FROM THESE THREE
>GREATER-THAN-EXPECTED DOSE GROUPS ARE USED, even though these dose groups are
>not even contiguous. 

This is just plain wrong.  I assume it is a bizarre misinterpretation
of the effect of one-sided testing.

>...Since the selected data are not significant, the IARC
>performs a Monte Carlo calculation on 5 000 trials (effectively multiplying
>the data by roughly a factor of 100) to "find" that the results show
>a "significant" linear dose-response "trend"...

This is equally bizarre!  The Monte Carlo simulation was done to
calculate the probability of obtaining a trend  as large as observed by
chance. Compared to the alternative of using the normal approximation
it REDUCES the chance of seeing a significant result.

>   These are the sole bases for the "conclusion" that the IARC
>study finds that the data is consistent with the linear model. 
>IARC is blatantly abusing science to achieve the results
>desired by government bureaucracies.

This just sounds like right wing politics rather than science.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
These are my views, not my employer's

Will Atkinson                                Internet: will.atkinson@aeat.co.uk
Health Effects                              
AEA Technology, 364 Harwell, Didcot          Phone:    +44 1235 434370
Oxfordshire, OX11 0RA, U.K.                  FAX:      +44 1235 432134
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -