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

Letter to HERA Editor



Hi Ted and All,



Here is the radon letter as submitted to HERA. It appeared as:



Seiler, F.A., and J.L. Alvarez, "Confounding Factors and the Use of

Epidemiological Data in Risk Assessment."  Hum. Ecol. Risk Assess. 8,

627-630, 2002.



The text is:



        CONFOUNDING  FACTORS  AND  THE  USE  OF

        EPIDEMIOLOGICAL  DATA  IN  RISK  ASSESSMENT



To the Editor  -  Recently, there were quite a number of discussions in

the literature and on mailing lists concerning the proper treatment and

use of ecological and epidemiological data.  A lot of these arguments

seem to ignore the logical connection between the data generated by

epidemiology and the data needed for risk assessment.  Above all, they

do not state what kind of scientific question is being debated, and that

omission is responsible for a good part of the confusion in these

discussions.  Therefore, we think that it may be useful to discuss these

aspects in some detail.

        Historically, the discipline called epidemiology arose from

attempts to

find the causes of epidemics such as the black death (most probably

bubonic plague) in the 14th century.  Later, in the 16th century, there

were also attempts by Paracelsus and others to understand the etiology

of occupational diseases such as those occurring in miners (Jacobi

1988).  Today, the main purpose of epidemiology is still to isolate

causes of health risks through the study of special population groups,

and to use these data to help evaluate the health risks caused by

exposures to the agent thus identified.  Therefore, there are two kinds

of legitimate issues here: One, there are the academic problems of

epidemiology itself in identifying the causes of health risks, and Two,

there are the problems with the scientific questions inherent in the

application of those data to risk assessment.  In this letter, we intend

to focus on this second kind of issues.

In a health risk assessment we often ask what are the consequences of

exposing a population to a particular agent.  Once we have the data, we

then speak of a risk to individuals, but the risk value we calculate is

really based on the average population response.  We can define the

population and its risk more concisely if we have criteria by which we

can stratify it, i.e., we can separate it into groups such as males or

females, children or adults, smokers or non-smokers.  Some risk

coefficients may be determined for such strata in order to simplify the

response as far as possible and to minimize the errors.  However, if we

have just an average population response without stratification, we can

only stratify by inference.  For example, a study that mixes smokers and

non-smokers may not have sufficient numbers of either group for

statistical significance.  For a larger study of this kind, the response

of non-smokers can be inferred by a subtraction of the assumed smoking

effect, but the result will likely be highly uncertain, in both the

random and systematic error components.

When data are available for one stratum only, the response of the whole

population or of some other stratum must also be inferred.  Such an

inference may introduce an even more important systematic uncertainty

that should be quantified as well as possible.  This is the opposite

problem from the one treated before, as it requires that a specific,

stratified response be converted to either a response for the general

population or for another specific stratum.  

In addition, there are several problems in transferring risk

coefficients determined from one population to another one (NRC 1988,

1999; ICRP 1991).  This operation requires either invoking the

assumption of equivalent populations (Seiler and Alvarez, 2000) or then

developing numerical corrections for the differences anticipated.  Both

methods involve additional random and systematic uncertainties that need

to be quantified.

Yet another important problem with a risk estimate can be caused by an

interaction between the effect of the prime agent and that of another

agent or condition (Seiler 1987, 2002).  Such an interaction can add to

the agent effect, reduce it, or even intensify it  -  effects usually

called confounding factors.  As an example, the mathematical risk model

for lung cancer due to an exposure to radon contains a sizeable

interaction between the terms describing the effects of radon exposure

and the terms for the effects of smoking.  This interaction term

contributes directly to the risk of lung cancer (Whittemore and McMillan

1983; Seiler 1987, 2002).  In epidemiological terms, this interaction

term corresponds to the large confounding factor caused by smoking

(Cohen 1995).

>From a purely academic point of view, a correction for smoking should be

made in order to obtain data for an exposure to radon alone, the

so-called marginal radon risk.  That might be the way to evaluate the

risk of radon exposure to non-smokers.  There is, however, an unsolved

aspect in this argument because  -  as so often in risk matters  -  no

appropriate uncertainty evaluation has been performed.  In this

situation, the question arises whether the small marginal excess radon

risk, remaining after the large interaction term has been subtracted,

yields a set of marginal radon risk values that are nonzero and larger

than their errors.  Unfortunately, this uncertainty information is not

available.  Thus, it is quite likely that a new careful study involving

a smaller group of non-smokers would yield better results with smaller

errors for the risks at the individual radon exposures.  

In the context of a risk assessment, it is thus important to state

clearly in every case the nature of the scientific question involved. 

If the question involves the marginal risk for a group of non-smokers at

different radon exposures, we doubt that we have at present sufficiently

good data for that marginal risk, even if the assumption of equivalent

populations can be made with some level of confidence.  However,

performing the experiment just discussed can make up for this

deficiency.

The marginal risk for smokers, on the other hand, cannot be measured

directly because lung cancer risks for smokers are always measured in

the presence of finite radon levels.  Therefore, the risks for smokers

is best determined by an analysis of all combined exposure-risk data by

a set of combined risk models, and an interpolation of the best model

for the actual radon and smoking exposures (Seiler 1987, 2002).  

In a recent paper, we have demonstrated that the confounding factors in

the radon data of Cohen (1995) do not invalidate these data for risk

assessments, contrary to the claims of some epidemiologists (Seiler and

Alvarez 2000).  Actually, making the 'usual assumptions' for data use,

we showed that the uncorrected Cohen data were the only data appropriate

for this risk assessment.  In the same paper, these 'usual assumptions'

for data use were discussed in some detail and globally termed the

Assumption of Equivalent Populations.  This assumption is generally made

in the use of every model, not only the health risk models discussed

here (Seiler and Alvarez 2000, 2002).  For health risk assessments and

in a direct answer to the scientific question asked, we shall,

therefore, have to treat the so-called ecological data sets in the same

way as any other epidemiological data set, each with its applicability

and its uncertainties, both random and systematic. 

As a consequence, there is only one more cohort for which risk

assessments can be made easily: the average American population group,

with the typical smoking prevalence of 25%.  Here, Cohen's uncorrected

data can be used directly as long as the assumption of equivalent

populations can be made, i.e., as long as the exposed population group

is a representative sample of the U.S. population for which the Cohen

data were determined (Cohen 1995).  

Here, an additional problem needs to be discussed: the strong

anti-correlation between smoking and radon exposure found in Cohen's

data.  Contrary to many opinions voiced in papers and on mailing lists,

we contend that this correlation cannot lead to doubtful risk data; in

fact, as it occurs in Cohen's study of almost the entire U.S. population

and whatever its cause, it must occur in any equivalent sub-population. 

A simple argument can elucidate this problem: Defining sub-populations

by selecting randomly every 1,000th, 10,000th, or 100,000th person, for

instance, must show the same correlation as long as the assumption of

equivalent populations can still be made.  If it cannot be made any

more, then the population has either become too small to show the

correlation, or there is a critical population size below which the

mechanism responsible for the correlation may no longer occur.

Yet another way to calculate the same risk would be to use the

theoretical model calculations of Bogen (1997, 1998).  His fitted model

parameters were based on independent data and were then used to predict

the shape and size of Cohen's data as well as  -  and from the point of

view of the Scientific Method most convincingly  -  the inverse dose

rate dependence of the carcinogenic effects of high-LET radiation.  This

model is, therefore, another solid basis for risk calculations with

reasonable intrinsic uncertainties.  The total systematic and random

errors of the risk estimates can then be determined by the inclusion of

the uncertainties of the exposure data, after propagation through

Bogen's model.

In conclusion, we would like to point out that the main step in our

reasoning here is one that should quite generally be taken in the

initial phases of every risk assessment.  This step is best taken by

asking the question: What is the exact scientific problem that we are

trying to solve with this risk assessment?



Fritz A. Seiler

Sigma Five Consulting, P.O. Box 1709, Los Lunas, NM, 87031-1709, USA

Tel: 505-866-5193; Fax: 505-866-5197; e-mail: faseiler@nmia.com



Joseph L. Alvarez

Auxier & Associates, 9821 Cogdill Rd., Suite 1, Knoxville, TN, 37932,

USA

Tel: 865-675-3669; Fax.: 506-866-5197; e-mail: jalvarez@auxier.com





  REFERENCES



Bogen KT. 1997. Do U.S. county data disprove linear no-threshold

predictions of lung cancer risk for residential radon  -  A preliminary

assessment of biological plausibility. Hum Ecol Risk Assess 3: 157-186



Bogen KT. 1998. Mechanistic model predicts a u-shaped relation of radon

exposure to lung cancer risk reflected in combined occupational and U.S.

residential data. Hum Exper Toxicol 17: 691-696



Cohen BL. 1995.  Test of the linear-no threshold theory of radiation

carcinogenesis for inhaled radon decay products. Health Phys 68: 157-174



ICRP, International Committee on Radiological Protection. 1991.

Recommendations of the international commission on radiological

protection. ICRP Publication 60. Pergamon Press, Oxford, UK



Jacobi J. (Ed.). 1988. Paracelsus: Selected Writings. Princeton

University Press, Princeton, NJ, USA 



NRC, National Research Council. 1988. The Health Risks of Radon and

Other Internally Deposited Alpha-Emitters: BEIR IV. National Academy

Press, Washington, D.C., USA 



NRC, National Research Council. 1999.  Health Effects of Exposure to

Radon, BEIR VI. National Academy Press, Washington, D.C. USA. In print.

Available at the National Academy Press website with URL:

http://www.nap.edu/readingroom/ books/beir6/



Seiler FA. 1987. Analysis of health or systemic effects caused by two

toxicants. Environ Internat 13: 459-467



Seiler FA and Alvarez JL. 2000. Is the 'ecological fallacy' a fallacy?

Hum Ecol Risk Assess 6: 921-941



Seiler FA and Alvarez JL. 2002. A scientific definition of physical

risk.  Submitted to Risk Anal; preprints available from FAS



Seiler FA. 2002. The risk of toxic agent mixtures: A phenomenological

semi-mechanistic approach. Submitted to Hum Ecol Risk Assess; preprint

available from FAS



Whittemore AS, McMillan, A. 1983. Lung cancer mortality among U.S.

uranium miners: A reappraisal. J Natl Cancer Inst 71: 489-49





Well, I hope this works out!



Best regards



Fritz  





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



Fritz A. Seiler, Ph. D.

President

Sigma Five Consulting

P.O. Box 1709

Los Lunas, NM 87031, USA

Tel.   505-866-5193

Fax.   505-866-5197

e-mail: faseiler@nmia.com



***************************s 





Ted Rockwell wrote:

--

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

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/