Mechanisms are Needed to Explain Cohen's Data

[Kai Kaletsch <info@eic.nu>]

To explain Dr. Cohen's large and very statistically significant inverse

correlation between radon and lung cancer requires a mechanism. Simply "Bad

Data" will usually bias results toward the null. That means that after

adjusting for errors in the data, the inverse correlation would be even

stronger.



For example, if you hypothesize that smoking has not been correctly

accounted for in the analysis because people might drive across state lines

to buy cheaper cigarettes, then you must also come up with a theory where

counties with high radon have higher or lower cigarette prices. If there is

no relationship with radon, then the fact that people drive across state

lines to buy cigarettes, simply decreases the signal to noise ratio, making

it more difficult to see a relationship between radon and lung cancer.



Such a relationship between cigarette prices and county radon levels would

easily be testable. If such a relationship was found, then the next step is

to look for a mechanism by which radon and cigarette prices are related.

Perhaps, politicians in high radon areas have been made aware of the

presumed association between radon and lung cancer in smokers and have

raised cigarette taxes in order to head off a lung cancer epidemic. The

point is that the mystery is not solved until some mechanism or causality is

found.



Exactly the same holds for the argument whether one should use cancer

incidence or mortality data. Unless the misclassification problem is somehow

associated with the radon level at the county level, the results will be

biased toward null. A possible mechanism for such a correlation would be if

radon would make doctors smart or stupid or if radon caused or prevented

other diseases, which would then interfere with the cause of death

determination in people with lung cancer. It could also be that people

diagnosed with lung cancer would be more likely to commit suicide in high

radon areas, because of radon hysteria. All of these mechanisms are

testable, although I don't consider them very likely.



I think the following mechanisms are more likely. They deal with

correlations between smoking and radon or radon progeny or lung dose. First,

we have to decide if we are interested in radon concentrations or radon

progeny exposures or lung dose. If we claim to test the BEIR IV model, then

we should use the quantity relevant to the BEIR IV model. (I think this was

WLM, but I don't have the report in front of me.)



Of course, it is also interesting to test a better model than BEIR IV.

Therefore, we could also test a model that correlates lung cancer to lung

dose. This is a separate issue.



1.    The first mechanism that I want to propose is the possibility of an

incorrect smoking to radon relationship on the county level. Homes with

smokers have on average 0.9 times the radon concentration of homes where no

one smoked. One mechanism that would explain the correlation is that houses

of smokers need to have a window open once in a while to let the smoke (and

the radon) out.



How does this translate into a correlation at the county level? How much

less radon do we expect in a county with 1% smokers vs 0% smokers? If each

smoker were to only affect his or her own air, one would expect the 1%

smoking county to have (0.01 * 0.9 + .99) = 0.999 of the radon level in the

0% smoking county.

Smoking will affect the radon level for everyone in the house. If 3 non

smokers share a house with each smoker, then the 1% smoking county is

expected to have (0.04 * 0.9 + .96) = 0.996 of the radon level in the 0%

smoking county. Smokers could also affect the radon level in other people's

homes by smoking in them and forcing the occupants to open windows. In areas

with a high proportion of smokers, it is probably more socially acceptable

to smoke in another person's home than in areas with a low proportion of

smokers.



It is not a trivial matter to take the known relationship that homes with

smokers have on average 0.9 times the radon concentration of homes where no

one smoked and move this to the county level. How was this done in the

analysis of Cohen's data?



2.    The second mechanism deals with the influence of smoking on the

Equilibrium factor F. There are competing factors that affect the

Equilibrium factor in smokers' houses (increased ventilation reduces F,

increased aerosol concentrations increase F, air cleaners reduce F .). I

have no idea which one would be dominant, but for the sake of argument, let'

s choose one that moves the data in the same direction as the first

mechanism. I.e. something that increases the negative correlation between

smoking and the suspected secondary carcinogen (WLM in the BEIR model).



One can imagine that the use of air purifiers (HEPA filters,

precipitators .) is related to smoking. The use of these units will

dramatically reduce the equilibrium factor in the living areas. That means

that there is much less radon progeny for each Bq of radon, increasing the

negative correlation between smoking and radon progeny exposure.



3.    The third mechanism deals with the influence of smoking on the number

of unattached radon progeny in the air. This becomes important if we want to

test the data against an "improved" BEIR model, one that uses lung dose as

the suspected carcinogen, rather than WLM. Unattached radon progeny are said

to deliver larger lung doses than attached radon progeny.



Smoking would reduce the number of unattached radon progeny. Smoking

increases the aerosol concentration, giving radon progeny something to

attach to, reducing the number of unattached radon progeny. Smoking causes

people to open windows, some unattached radon progeny will escape out the

window. Air purifiers will catch some of the unattached radon progeny.



In all cases the dose due to the unattached radon progeny is reduced due to

smoking, further increasing the negative correlation between smoking and the

suspected secondary carcinogen, lung dose.



Conclusion: I don't really think that the above mechanisms will explain

Cohen's strong negative slope. [However, if they do, I'll take the $5 000.00

:) ] The point is that the mechanism is all-important. Every observation

must have an explanation, whether the observation is on an individual level

or an aggregate level. If you are not looking for a mechanism, you are not

doing science!



Kai Kaletsch



http://www.eic.nu



************************************************************************

You are currently subscribed to the Radsafe mailing list. To unsubscribe,

send an e-mail to Majordomo@list.vanderbilt.edu  Put the text "unsubscribe

radsafe" (no quote marks) in the body of the e-mail, with no subject line. You can view the Radsafe archives at http://www.vanderbilt.edu/radsafe/