False positives and you -- important!

• From: starbug_56
• Date: Thu, 20 Dec 2001 14:10:30 -0800
• Message-ID: <000U24OEM.0019FM33I.12170551-quizbowl_at_...>

This is a bit off-topic, but I think it's an
important thing for folks to know. If anything, it points
out some of the problems in situations where people
are mass-screened, sometimes by folks with unclear
scientific or clinical credentials, for conditions that may
carry a social stigma.

Edmund wrote:

>
Isn't there some statistics theorem that says there's
always way more false positives than
 false
negatives?

To which Dave Thorsley responded:

>
Suppose there's a condition that affects 1% of the
population, and also that there's a
 diagnostic test
that's 99% effective and assigns false positives and
false negatives with equal

probability.

(details deleted for brevity -- see previous
message)

A VITALLY important take-home message, reflected
very nicely in Dave's parameters: It's important to
realize the contribution of three factors here.

*
The sensitivity of the test (roughly speaking, its
ability to detect true positives).

* The
specificity of the test (roughly speaking, its ability to
detect true negatives)

* The prevalence in the
tested population of the condition for which one is
testing.

How does this affect me, you say?

Well, for
one thing, in a mass screening for any condition
that's not expected to be highly prevalent in the test
population, one can expect a lot of false positives --
sometimes even many times those of true positives. One
example might be random drug testing in a large
population where only a small percentage of the test
subjects are actually drug users. (In a population where
drug abuse actually *is* highly prevalent, such as a
group of addicts who have had repeated relapses after
multiple trips to detox, the true-to-false-positive ratio
would be much higher for the same test
method.)

If anyone wants a more detailed but not overly
mathematical treatment of this subject, here's a resource from
the CDC's web
site:

<a href=http://www.cdc.gov/hiv/pubs/rt/sensitivity.htm target=new>http://www.cdc.gov/hiv/pubs/rt/sensitivity.htm</a> 

This is a model of HIV screening, using a
test with 99.6% specificity, in populations with
different HIV prevalences. 

A somewhat more detailed
site from the Medical University of South Carolina
points out that even a test with 99.9% specificity and
99.9% sensitivity must be interpreted with prevalence
in
mind:

<a href=http://www.musc.edu/dc/icrebm/sensitivity.html target=new>http://www.musc.edu/dc/icrebm/sensitivity.html</a>


Julie