[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Cohen's Ecologic Studies



Gary,



I am sorry you don't understand some of my points.  

Perhaps I have not done a very good job, on this list 

and limited format, explaining my position on this 

issue.  However, if you sincerely want to try to 

understand my perspective on this issue, please read the 

papers on this subject we published in the Journal, 

Health Physics.  Please email me directly if you do not 

have the references.  



As for your question,  an excerpt from one of our papers 

answers your question.  Please email me directly if you 

do not understand the response below so that we can 

limit further gnawing on this topic on the list.  We 

have previously answered his question and even gave his 

question a name.



The quotation below is from: Field RW, Smith BJ, Lynch 

CF. Cohen's Paradox. Health Physics. 77(3): 328-9, Sep 

1999 



"Cohen (1999a)continues to challenge scientists to 

suggest a plausible explanation to explain the inverse 

relationship he notes between mean county residential 

radon measurements and mean county lung cancer mortality 

rates. We will call this inverse relationship "Cohen's 

Paradox." Cohen (1999a) states that his challenge is for 

someone to suggest a "not implausible model" as a 

possible explanation and that the burden of proof will 

be on him to show that "the explanation is highly 

implausible." We maintain that even if additional 

plausible models are offered, Cohen will likely not be 

able to explain his own paradox. Cohen has not accepted 

the fact that it may be impossible to explain Cohen's 

Paradox in definitive analytical terms with his existing 

data because it is not always possible to identify 

empirical sources of ecologic bias from aggregate 

(ecologic) data alone (Field et al. 

1998a).



Cohen (1999a) states that he has not been able to 

explain the inverse relationship (Cohen's Paradox) for 

his studies even with years of effort. We are not 

surprised. Cohen (1999a) continues to miss the point 

made previously (Greenland and Robins 1994; Smith et al. 

1998; Field et al. 1998a) that characterizing biases is 

often extremely difficult in ecologic studies of 

geographic regions because of the high probability of 

interacting covariates that may differ across these 

regions. Greenland and Morgenstern (1989) point out that 

ecological control of a covariate contributing to 

ecologic bias will usually be inadequate to remove the 

bias produced by the covariate even in the absence of 

measurement error. Researchers (Greenland and Robins 

1994; Lubin 1998; Smith et al. 1998; Archer 1998; 

Goldsmith 1999) have already presented very plausible 

theoretical examples of how Cohen's data can produce 

incorrect and even contradictory risk estimates. Cohen 

has rejected all of these examples.



Lagarde and Pershagen (1999) recently performed 

concurrent analyses on individual and aggregated data 

from a nationwide case-control study of residential 

radon and lung cancer in Sweden. The authors reported 

that the results confirm that ecologic studies may be 

misleading in studies of weak associations. So, are 

Cohen's negative point estimates a true effect or are 

they attributable to bias? To move the explanation 

beyond the theoretical level, analyses would require 

individual level data beyond the quality of Cohen's 

aggregate data.



Cohen continues to maintain that his ecologic studies 

avoided the ecologic fallacy, because he was testing the 

BEIR-IV LNTT model (Cohen 1997,1998,1999a). Cohen also 

continues to deny our assertion (Smith et al. 1998; 

Field et al. 1998a) that he was not testing the BEIR-IV 

LNTT model. As we stated (Smith et al 1998; Field et al. 

1998a), Cohen's risk model is not the BEIR-IV risk 

model. Cohen attempted to equate his derived LNTT model 

to the BEIR-IV model by applying unsupported primary and 

secondary rigid assumptions. The assumptions all have 

both an error associated with them and a non-linear 

component, which as previously pointed out, cannot be 

quantitatively described."



If Dr. Cohen would use an LNTT model that eliminates his 

unsupported assumptions that assume linear relationships 

within counties, then it would be much easier to explain 

his findings.  His inverse findings in my opinion are a 

direct result of the erroneous LNTT model he uses, and 

the findings are due to residual confounding of non-

linear variables within county such as smoking and other 

covariates.  



Lubin has just presented an example of this in the June 

publication I previously referenced.   



Gary, I will gladly stop responding to emails on this 

subject.  If you recall, all I did was post a message 

concerning Dr. Lubin's paper, nothing more.  I have 

merely been answering emails, such as yours, in response 

to that first post.



I have one question for you.  How do you explain a 

similar inverse association between other cancers 

associated with smoking with Dr. Cohen's radon data?  



Which is more feasible?



1.) Alpha exposure to the lung decreases smoking related 

cancers at other sites other than the lung,



or 



2.) The inverse effect between radon and lung cancer is 

really an artifact of residual confounding not accounted 

for by Dr. Cohen's LNTT model or analyses?  The majority 

of researchers polled elsewhere, favored number 2 (see: 

http://www.rivm.nl/rca/opinion_cohen.html 



Regards, Bill



> 

> Your postings usually seek to discredit the Cohen study, but they

> usually fall short of being understandable.  If you have a clear and

> valid point to make, please do so.  Its no good saying there is a

> problem without showing clearly what the problem is.

> 

> So how about it?  Since you are determined to discredit the study on

> Radsafe, make your case clear.  Respond to the question:

> ------------------------------------------------------

>         --Why can't you make up a specific numerical example and show

> how

> it can affect my results?

> ------------------------------------------------------

> If you can't do that, then I agree with you, "there is no point to

> continue this dialog" and you should stop gnawing this bone.  

> 

>    _______________________________________________

> 

> 	Gary Isenhower

> 	713-798-8353

> 	garyi@bcm.tmc.edu

> 

> epirad@mchsi.com wrote:

> > 

> > Dr. Cohen,

> > 

> > It is clear you do not understand the points I am trying

> > to make. If you really believe that non linearity of

> > confounding has nothing to do with your findings than

> > there is no point to continue this dialogue.  After all,

> > it is not the first time we agreed to diagree.  I think

> > we can both agree on that.

> > 

> > Lubin has recently given you an example of the problem,

> > there is no need for me to repeat it.

> > 

> > J. Radiol. Prot. 22 (June 2002) 141-148

> > 

> > The potential for bias in Cohen's ecological analysis of

> > lung cancer and residential radon

> > 

> > Jay H Lubin

> > 

> > Biostatistics Branch, Division of Cancer Epidemiology

> > and Genetics, National Cancer Institute, EPS/8042, 6120

> > Executive Blvd, Rockville, MD 20892-7244, USA

> > 

> > Abstract. Cohen's ecological analysis of US lung cancer

> > mortality rates and mean county radon concentration

> > shows decreasing mortality rates with increasing radon

> > concentration (Cohen 1995 Health Phys. 68 157-74). The

> > results prompted his rejection of the linear-no-

> > threshold (LNT) model for radon and lung cancer.

> > Although several authors have demonstrated that risk

> > patterns in ecological analyses provide no inferential

> > value for assessment of risk to individuals, Cohen

> > advances two arguments in a recent response to Darby and

> > Doll (2000 J. Radiol. Prot. 20 221-2) who suggest

> > Cohen's results are and will always be burdened by the

> > ecological fallacy. Cohen asserts that the ecological

> > fallacy does not apply when testing the LNT model, for

> > which average exposure determines average risk, and that

> > the influence of confounding factors is obviated by the

> > use of large numbers of stratification variables. These

> > assertions are erroneous. Average dose determines

> > average risk only for models which are linear in all

> > covariates, in which case ecological analyses are valid.

> > However, lung cancer risk and radon exposure, while

> > linear in the relative risk, are not linearly related to

> > the scale of absolute risk, and thus Cohen's rejection

> > of the LNT model is based on a false premise of

> > linearity. In addition, it is demonstrated that the

> > deleterious association for radon and lung cancer

> > observed in residential and miner studies is consistent

> > with negative trends from ecological studies, of the

> > type described by Cohen.

> > 

> > URL: stacks.iop.org/0952-4746/22/141

> > 

> > Regards, Bill Field

> > >

> > > On Tue, 4 Jun 2002 epirad@mchsi.com wrote:

> > >

> > > > Dr. Cohen,

> > > >

> > > > To ignore non-linearity is the root cause of your

> > > > findings.

> > >

> > >       --Linearity of confounding factors has no relevance to my study

> > >

> > >  In an ecologic analysis you are limited to a

> > > > summary statistics to adjust for confounding.  Since you

> > > > do not have information on covariates at the county

> > > > level, accurate ratio functions cannot be calculated.

> > >

> > >       --Why can't you make up a specific numerical example?

> > >

> > >

> > > > This becomes very problematic if the data structure is

> > > > non linear (e.g. not everyone in the county smokes

> > > > cigarettes for the same duration and intensity; not

> > > > everyone spends the same amount of time in their home,

> > > > not everyone is exposed to the same radon concentration,

> > > > etc...), and non additive, which is the case at hand.

> > >

> > >       --Why can't you make up a specific numerical example and show how

> > > it can affect my results?

> > >

> > > >

> > > > I believe the onus is on you to show that multiple non-

> > > > linear covariates are not the cause of your problem.

> > > > The only way I know you can attempt to do this is use

> > > > the methods of Sheppard and colleagues.

> > > >

> > >

> > >       --I never assume anything is linear, except lung cancer vs radon.

> > > I need a numerical example to understand what you are talking about.

> > >

> > > > Lubin has demonstrated the problem in a recent paper

> > > > just using smoking.

> > >

> > >       --My papers give examples of how errors in smoking can explain my

> > > results, but then I show that the required correlations are completely

> > > implausible. Lubin never addresses the issue of plausibility.

> > >

> > >  Your inverse associations are found

> > > > for other smoking related cancers that should not be

> > > > related to radon.  This further strengthens my argument

> > > > that your inability to adjust adequately for smoking is

> > > > driving your findings.  Or do you believe the reason the

> > > > other smoking related cancers also have an inverse

> > > > association with your radon concentrations is because of

> > > > a hormetic response due to alpha radiation exposure to

> > > > the lung?

> > >

> > >       --I have addressed this in previous messages

> > >

> > >  I find it far more credible that the

> > > > explanation is lack of control of confounding by smoking

> > > > and other factors as Lubin has just demonstrated.

> > > >

> > >       --In BEIR-IV, smoking is not a confounder. Smokers and non-smokers

> > > are treated as entirely different species.

> > >

> > >

> > > > You 1997a) claim that simple linear least squares

> > > > regression of m on S indicates that nearly all lung

> > > > cancer is due to smoking.  However, the results of this

> > > > analysis do not support such a claim.  We repeated the

> > > > regression of lung cancer mortality rates on your

> > > > adjusted smoking percentages.  The resulting R2 values

> > > > indicated that S explains only 23.7% of the variation in

> > > > lung cancer mortality rates among females and 34.5%

> > > > among males.  Puntoni et al. (1995) compared six

> > > > mathematical models relating cigarette smoking to lung

> > > > cancer risk using data from nine large cohort studies.

> > > > They found that 67% of the variation in relative risks

> > > > could be explained by a two-stage model of

> > > > carcinogenesis.  In comparison, very few of the lung

> > > > cancer deaths are explained by your smoking variable.

> > > > Therefore, the smoking variable is inadequate to adjust

> > > > for the effects of smoking.

> > > >

> > >       --I have addressed the R-squared issue previously and shown that

> > > it is irrelevant

> > >       --I have also shown that any choices of the smoking prevalences in

> > > the various counties that are not completely implausible would not affect my

> > > results.

> > >

> > > ************************************************************************

> > > 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/

> > >

> > ************************************************************************

> > 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/

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

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/