[ RadSafe ] AW: Low level radiation and cancer:

Christian Hofmeyr chris.hofmeyr at webmail.co.za
Sun Aug 7 14:24:06 CDT 2005


Rainer Facius and others,
I would have liked to follow the discussion of LNT in more
detail, but unfortunately I have been suffering from a bad
internet connection and furthermore I do not have ready
access to a technical library to look up references.  So
please bear with me if my offerings seem slightly off the
mark.
I believe that solving the LNT conundrum is very important.
 A vindication would be useful in terms of making rational
decisions regarding risk.  The first prize would be an
indication of a threshold or even hormetic benefit, as it
would allay public fears about numerous beneficial nuclear
and radiation applications.  It seems to be a tough nut to
crack, but some of the problems may be of our own making. 
The first point I would like to raise concerns the
statistics of background subtraction. I maintain that one
should try and avoid this procedure if at all possible and
only use it as an absolute last resort. This is very much
at variance with general practice, which scientists seem to
imbibe with mother’s milk, but the reason is at least
threefold:
1.	Variances add, whether one adds or subtracts variables,
causing a much greater relative uncertainty in the
subtracted result.  So much the worse when one is dealing
with large uncertainties.  Extracting radiogenic cancers
from the total incidence seems a case in point.
2.	One should also not be fooled by ‘accurate’ knowledge of
the average background – the sampling of the foreground
determines the variance on the background relevant to the
sample also, irrespective of how well the average
background is known.
3.	In many cases one is dealing with Poisson distributions
in which the variance is equal to the average and the
standard deviation is therefore its square root.  Now the
clincher:  if one adds two Poisson distributed variables,
the result is Poisson distributed, but if one subtracts,
the resultant distribution loses its important Poisson
quality. 

Now the definition of radiogenic cancer as an excess
already implies that one can and should subtract the
‘other’ cancers (of the same type).  From a general
perspective this need not be correct.  By using the concept
of excess cancer, one is already forcing a specific
additive model of radiogenesis.  A more general approach
would also allow e.g. a multiplicative model or a
combination.

To get back to LNT:   It stands to reason that the damage
caused by radiation is proportional to the dose, at least
in the low to medium dose (and doserate) range.  It is the
reaction of biological systems to the rate and amount that
might be non-linear.  LNT predicts an outcome which is
linear with dose, i.e. either there is no biological
response to radiation damage in the organism which counters
the outcome, or, otherwise, the effectiveness of the
response is constant with dose (i.e. proportional to dose
from zero up).  This implies that every bit of dose counts
and the number of resultant radiogenic cancers is directly
proportional to the collective dose of any suitably defined
group.  This forecast is the most risky one for LNT and IMO
provides the most sensitive test.  (For this purpose a
carefully conceived ecological study could have clear
advantages over an epidemiological one.)

In view of my first thoughts,  the logical thing to
validate LNT would therefore be to plot total cancer
incidence (by type) against collective dose, obviously
controlling for a number of variables like age, gender,
group size, etc. The basic proviso is that the average dose
to the individual must vary. [If group size would not be
constrained, one could perform a gedanken experiment which
proves LNT: consider a homogeneous or otherwise
well-controlled group and vary the collective dose by the
size of the group considered! QED.]  If one does not obtain
a linear relationship with positive gradient and a positive
y intercept, then LNT fails.  This should be infinitely
more sensitive than trying to follow excess cancers as a
function of individual doses.  It is not necessary or
productive to try and subtract a non-radiogenic background
(although this compromises the log-log presentation, I
guess).  Natural radiation background must be included in
the collective dose calculations.  
One can think of interesting possibilities in constituting
different cohorts.  Comparison of cohorts living under
different ecological radiation conditions could be a very
useful approach.  Call me naïve, but I therefore am not
completely pessimistic about validating or otherwise
disproving LNT within certain confidence limits, as with
all physical theories.  The use of collective dose would
probably only yield a YES or NO for LNT.  More
sophisticated studies would be required to delve into
mechanisms if the LNT answer is NO.

The very weakness of radiation as a carcinogen makes it
very difficult to control observations adequately for
sensitive confounding variables.  Apart from the
statistics, important problems are
1. recognition and adequate control of confounding
variables,
2. reliability of cancer incidence and mortality
statistics,
3. dose and collective dose evaluation and its consistency,
4. choice of time frames,
5. etc.
  
The above proposal is over-simplified.  I have not even
touched on the choice of dose definition.  There are
complications with e.g. the study of cancer, primarily the
very sharp increase of incidence with age, so that
controlling for age must be ultra-diligent.  Cohorts could
be varied by e.g. excluding and including certain age
groups.  It can be assumed that the reliability of cancer
incidence and mortality statistics is rather variable
between regions and countries. Reliable statistics is a
clear prerequisite for a valid LNT study and this
reliability would have to be quantified for all relevant
variables and the sensitivities established.

Another problem is the effect of confounding changes in
lifestyle with time. Particularly lung cancer is very
problematic in this regard. The SEER statistics for the USA
show a very large gender differential in mortality of
almost a factor 5 during the 1970s, which narrowed
dramatically subsequently due to a levelling out and slight
decline of male lung cancer mortality, coupled with a
simultaneous steep rise in female LC mortality since the
1970s.  These dramatic changes are most probably due to
changing smoking behaviour.  This should give an indication
of the problems confronting domestic radon studies.
Chris.Hofmeyr at webmail.co.za


_________________________________________________________________
Need software for your hardware? Click here http://www.asg.co.za



More information about the RadSafe mailing list