Re: LNT models -Constancy of Radon levels over time??

[Michael McNaughton <mcnaught@lanl.gov>]

At 01:41 PM 3/8/00 -0600, you wrote:
>How would you design a study to examine the risk posed by residential radon?
>How would you test the validity of the LNT?

These are excellent questions. Challenging questions require an unusual
approach, as follows.

1. Because we are looking for small effects, it is essential to study a
large population, e.g., 100 million people or more, but we have to avoid
the ecological fallacy.

2. Because we don't have the resources to collect complete data about such
a large population, we compile as much data as possible. At a minimum, we
compile the means AND the distributions about the mean for all the relevant
variables such as smoking, radon, and lung cancer. Whatever data we have on
correlations will be valuable but not essential.

3. We generate a very large data base (several gigabytes) to represent 100
million hypothetical people, and allow the computer to adjust the details,
subject to the constraints of the data we have compiled. For example,
because we do not know how the variables correlate for each individual
person, we allow the computer to vary the correlations, subject to the
constraints of the data.

4. We program the computer with a hypothetical model (such as the BEIR
model) to predict the incidence of lung cancer as a function of the
variables. The computer program adjusts the model AND the correlations to
produce the best fit to the data. If the model were perfect, the
uncertainty would be limited only by counting statistics. In practice, the
closer the model comes to this ideal, the better it is.

mike

Mike McNaughton
email: mcnaught@LANL.gov or mcnaughton@LANL.gov
phone: (505)667-6130
************************************************************************
The RADSAFE Frequently Asked Questions list, archives and subscription
information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html