Re: Cancer clusters

[Kai Kaletsch <info@eic.nu>]

From: "Dov Brickner" <brickner@zahav.net.il>



>  It is true that  most epi works consider the alpha to be 0.05. It doesn't

> mean that 5% of these studies are false positive - on the contrary. Since

> you look for cancer in many sites , every study examine more then 20

> parameters and therfore carry agood chnace to find at least one site to be

> (falsely) positive (i.e. significant).



I was involved in a study once where a cohort was supposed to be followed

for some 40 years to see if there was an increase in lung cancer. Apparently

this was a little boring for the lead epidemiologist and she suggested that

we look at other outcomes as well (ones where there was no a priory

suspicion that they were connected with the substance under investigation).

She further suggested that we didn't have to wait for 40 years, we could

check the data before and we could publish results at any time that we found

something.



If you check for 20 outcomes at 20 time intervals (400 combinations) and

used an alpha of 0.05 you would have 20 false positives from just the one

study. (Yes, I know the 400 combinations are not independent, but its close

enough.) A few of us who had a more fundamental understanding of statistics

objected to this, but the lead epidemiologist insisted that this was

standard practice and we didn't know anything about epidemiology.



Kai



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

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/