Re: Radon and Lung Cancer: What the studies really say.

[epirad@mchsi.com]

I do not think you understand either the significance of R2 or cross level 

bais, but we have been down that road too many times.  I have just read your 

passages you cited and they are far from persuasive.  Your smoking data should 

be able to eplain the variabilty of lung cancer mortality especially in such a 

large data set.  A simple plot of the data shows the poor agreement.  The 

predictability of your lung cancer mortality estimates do even a poorer job of 

predicting the variation in mortality incidence for the time period of 

interest.   



At this point, let's agree to see if the NCRP provides a view on the adequacy 

of your smoking data since this dialogue is going nowhere.   



Bill Field  

> 

> On Fri, 20 Jun 2003 epirad@mchsi.com wrote:

> 

> >    We repeated the

> > regression of lung cancer mortality rates on these adjusted smoking

> > percentages.  The resulting R2 values indicate that S explains only 23.7% of

> > the variation in lung cancer mortality rates among females and 34.5% among

> > males.

> 

> 	--The problem here is that you do not understand the meaning of

> R-squared. It's affected by all the small random fluctuations in smoking

> and lung cancer rates; this was explained in my paper in Health Physics

> 72:489-490;1997. A much more meaningful test of the ability of our smoking

> prevalences, S, to explain lung cancer rates is to do the same regression

> you do, fitting the data to

> 		m = P + Q S

> where P and Q are fitting parameters, and observe the standard deviations

> in the two terms. From Table 2 in Item #15 on my web site (the factor 100

> in Column 3 should be deleted)

> 	P = -9.7 +/-2.1     and   Q S(average) = 60 +/- 4



> Thus the second (smoking)term is completely dominant, with negligible

> uncertainty. (The minus sign for the first term would suggest that with no

> smoking, there would be negative lung cancers)

> 

> 	--One of the three methods I use for determining S-values is to

> derive them from the lung cancer data for counties of similar radon

> levels. Surely you can't claim that these do not explain the lung cancer

> vs smoking relationship, as they are derived from it. Using those S-values

> gives only a slightly higher R-squared, 41% vs 35%, and

> performs well in the above test

> 	P = -10 +/-2        and   Q S(average) = 66 +/- 2

> It gives very similar results to those obtained from my principal set of

> S-values, that is essentially the same discrepancy with LNT

> 

> > 	The poor predictive power of S is due, in part, to a failure to allow

> > for the effects of smoking intensity and duration.

> >  Cohen’s failure to

> > incorporate intensity and duration in his analysis naturally leads to a 

> smoking

> > variable which accounts for fewer lung cancer deaths.

> 

> 	My not using smoking intensity in my original work was due to the

> fact that I was using the BEIR-IV theory which does not include intensity

> of smoking. Two of my three methods for deriving S-values do incorporate

> implicit weighting for intensity of smoking -- deriving S-values from lung

> cancer rates, and from cigarette sales tax.

> 	However I do give an elaborate treatment of intensity of smoking

> in Health Physics 78:522-527;2000, summarized in Sec. 4.4 of Item #7 on my

> web site. It shows that this can do little to resolve the discrepancy with

> LNT.

> 

> > Puskin's finding further supports that you smoking rates do a poor job of

> > predicting lung cancers in counties since your inverse assocaition was found

> > for other smoking related cancers which are not related to radon exposure.

> 

> 	--This is thoroughly refuted in Item #15 on my web site. Can you

> say how you find that it is not conclusive.

> 



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

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/