Feasibility of Low-Dose Epi: Minimum Detectable Amount

[dj_strom@ccmail.pnl.gov]

Reply to Frank R. Borger <Frank@rover.uchicago.edu >

You're right: from a purely science point of view, we are wasting 
a lot of money on low-dose epidemiology studies.  But there are 
other valid reasons for doing science that doesn't stand a chance 
of sholwing something, primarily the quest for peace of mind or 
the quest for re-election.  I believe many epidemiology studies 
of low-dose effects are done for some political purpose or other, 
be it to show that "a little nukie never hurt anyone" or for some 
"ain't it awful" cause such as to show how dangerous radiation is 
and that there are conspiracies everywhere to cover it up.  

Since the late 1970s, good epidemiologists have performed "power 
calculations" for studies based on some hypothesis (e.g., "this 
population will be like the Japanese bomb survivors").  See, 
e.g., Dreyer, N.A.; Kohn, H.I.; Clapp, R.W.; Covino Jr., S.J.; 
Fahey, F.H.; Friedlander, E.R.; Loughlin, J.E.  The Feasibility 
of Epidemiologic Investigations of the Health Effects of 
Low-Level Ionizing Radiation.  NUREG/CR-1728; Springfield, 
Virginia: National Technical Information Service; 1980.  

A power calculation is conceptually identical to the so-called 
"minimum detectable amount" (Currie's "detection level," L{sub 
D}):  the MDA is that amount, which, if actually present in a 
sample, will most likely be distinguishable from background (say, 
with a 95% chance of giving a result greater than our decision 
level (Currie's "critical level," L{sub C}).

Thus, a power calculation will tell you if you can reject the 
null hypothesis of no effect with your study size and design if 
the postulated effect is thus-and-such.  Power calculations 
apply, of course, to the designs of case-control and cohort 
studies, probably not to ecologic study designs since there are 
so many difficulties with those.  Dreyer et al. (1980) found 
that, if other populations respond to radiation the way the bomb 
survivor did up to that date, only occupational studies such as 
nuclear power plant workers or DOE workers, and perhaps air 
crews, had any merit.  Of course, low promise of likely outcome 
never deterred a researcher who had managed to get funding 
anyway, or someone with an ax to grind.

With regard to rare diseases, power calculations can be done for 
those, too.   For common diseases, such as lung cancer, power 
calcuations will show that you're hosed without a huge effect 
(remember, cigarette smoking has a 10-fold effect; even Cohen 
shows only a predicted 2-fold effect from radon, and without 
controlling for smoking on an individual case-by-case basis, 
you're not going to see anything).  There are ways to increase 
power using special study designs, such as the nested 
case-control study or the cohort study with a non-exposed 
comparison group drawn from the same workforce.  

Nonetheless, as has been said before, epidemiology is an 
extremely blunt tool.

- Dan Strom <dj_strom@pnl.gov>