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

Re: Cases and Controls



John & Group,

Thanks for the tip of Rothman/Greenland's tome on Modern Epidemiology. I

can't say it's interesting reading, but it is much more current than the

epidemiology text I have on my shelf.



The multivariate regression studies "averages within levels of multiple

regressors". I take this to mean that smoking is not corrected on the

individual level, but at the group level by some pre-established average.



The section on population controls says that "we would employ density

sampling, in which an individual's control selection probability is

proprotional to the individual's person-time risk." This (at least to me)

supports the premise that smoking needs to be more adequately controlled,

especially since smoking and radon, at least in part, have the same

underlying risk, but the controls and cases have very different

person-times. Let me explain,



The purported risk to individuals from radon is due to alpha radiation,

particularly from radon progeny. One of the purported risks to smokers (and

those exposed to second hand smoke) is alpha radiation from Po-210 (also a

radon progeny). Smoking tends to expose smokers to far greater doses of

alpha radiation that radon. In case-control studies where smoking

prevalence is greatly different between the case and control subjects, the

two groups cannot be adequately compared because the most significant

contribution of alpha radiation (i.e. smoking) is not controlled. 



If multivariate regressions adequately corrected for smoking, then radon

studies including smokers with appropriate corrections should agree with

studies that exclude smokers. Generally, they do not. Studies with smokers

tend to have a small positive correlation for radon, whereas studies

excluding smokers have been negative. That makes me concerned that

multivariate regressions being applied to data sets do not adequately

correct for smoking.



To remedy the situation, smoking habits between cases and controls should

be the same. Controls represent the general population so that the

percentage of smokers should be consistent with the group. Reducing the

percentage of smokers in the case group to match the controls would lead to

selection bias. The best options is to exclude smoking altogether in both

case and control groups.



Advantages of excluding smokers in radon case-control studies:

(1) The cases and controls can be more closely matched.

(2) No statistical correction needs to be applied for smoking.

(3) Confounding due to chemical carcinogens present in cigarettes is

eliminated.

(4) Since smoking reduces radon levels, radon levels will be higher and

more compatible with most of the general population, considering that most

people do not smoke.

(5) Investigators have reported that Retrospective (track-etch on glass)

Radon Detectors are more variable in residences of smokers than

non-smokers. These results would be more consistent.

(6) Investigators can include in their research other low order potential

carcinogins, such as prior lung diseases, second hand smoke, occupations

exposures, etc. The focus of the studies can be "What causes lung cancer in

the 10% of men and 15% of women who do not smoke?" 



Disdvantages of excluding smokers in radon case-control studies:

(1) The number of lung cancer cases from which to study falls off

precipitously when smokers are excluded. Regardless, the numbers should be

similar to other less prevalent cancers and sufficient information should

be attainable to obtain meaningful conclusions.

(2) The percentage of smokers in the control group would not be the same as

the general population. This is acceptable since about 75% of the general

population does not smoke.



Sorry for the long diatribe.  

Tom



John Williams wrote:

> 

> Tom,

> 

> Adjustment for confounding is a fairly straightforward procedure in

> Epidemiology.  See Rothman's and Greenland's Modern Epidemiology text

> for more details.  There is no algorithm, but rather part of the

> multivariate analysis.  I just looked over the Iowa AJE paper and

> they did check for residual confounding by smoking and did not detect

> any.

> 

> http://www.amazon.com/exec/obidos/ASIN/0316757802/ref=pd_sim_books/103

> -0905965-4263056

> 

> Sent by Law  Mail



-- 

Thomas Mohaupt, M.S., CHP

University Radiation Safety Officer



104 Health Sciences Bldg

Wright State University

Dayton, Ohio 45435

tom.mohaupt@wright.edu

(937) 775-2169

(937) 775-3761 (fax)



"An investment in knowledge gains the best interest." Ben Franklin

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

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/