linearity, more noise

["Les Slaback, Health Physics" <SLABACK@micf.nist.gov>]

Some random thoughts regarding models, and their implications:

1. Does anyone believe that a single model fits every
carcinogenic disease?  If not, then how can one 'prove' or even
reasonably assure that a single model adequately represents the
collection of these diseases, especially when the members of the
collection typically are not even specified and characterized?

2. The 'linear' model is only a portion of the dose-effect
phenomenological model.  Projecting the effect also involves the
selection of an additive or multiplicative (absolute or relative
risk) relationship.  Why should these be separable variables?
Simplicity in modeling is certainly a pragmatic justification,
and it provides a basis to pool risk data that could not
otherwise be compared.  But if the goal is to identify the
'right' model, shouldn't this separability be reviewed?

3. Before considering other models, presumably most are willing
to accept that linearity fits the data for certain diseases over
a defined dose range for a specified dose rate and radiation
quality. So the question becomes how far you are willing to
project that dose range downwards in dose, in dose rate, and to
other radiation types for the linear model, e.g., ...
   far enough to determine that 5 rem (or 2 rem) is an acceptably
safe annual dose limit?
   below the annual dose limit sufficient to support the ALARA
concept and the public dose limit?
   below the public dose limit far enough to support allocating
portions of the public dose limit to various pathways or sources,
e.g., 40CFR61
   far enough to support a BRC level

In the absence of direct data so that one has to depend on
theoretical modeling, how much lower in dose are you willing to
accept a linear projection?  On what basis are you willing to
accept one level of projection but not another, particularly for
diseases for which there is no direct data?  Once you determine
that the linear model does not apply at some low dose you need to
specify the cross-over.

4. Scientific curiosity is a driving force for many theories, but
clearly this folderol is driven by the social implications of the
theory, e.g., economic impacts.  Given that there is no
scientific data or tool to definitively identify the 'right'
model in the dose range of social interest, one is forced to
accept the most socially acceptably model among several that fit
the data.  If that is the case, then it would seem to be a
decision for the politicians (pending definitive data from the
scientific community).  Given the tendency for political
decisions to ignore cost issues and to concentrate on minimizing
potential negative outcomes (accept no risk at any cost), a
supralinear model would seem a more likely outcome.  Are we sure
we want to turn this into a political process (any more than it
is already)?

5. Having some faith (blind trust?) in progress in the biological
sciences I believe that in my lifetime the definitive experiment
will be formulated that will identify the fundamental processes
that create and promote each cancer type, including the influence
of outside physical agents like radiation.  If one accepts that
hypothesis and the possibility that the linear hypothesis might
be excessively conservative, then decisions now should be of a
reversible type and in the context of 30-50 yr impact (I am an
optimist, i.e., a golfer), not in terms of 1000 year or 10,000 yr
impacts where social cost is a major issue.


It is certainly legitimate to debate the linear-no threshold
hypothesis, but it would seem to me that ultimately it has to be
in the context of the overall dose-effect model, with full
involvement of the current effects database, and with explicit
summaries of the social impact associated with conservatism or
lack of conservatism of the various staistically acceptable model
choices.  While such summaries imply more knowledge or certainty
than we have, they probably have as much confidence as the choice
between different models in the very low dose range (<1 rem).
And such summaries would add a touch of reality to what can be a
somewhat abstract discussion.

SLABACK@MICF.NIST.GOV
   ...a little risk, like a bit of spice, adds flavor to life