Replies to post #16981 on Biotech Values

[rfj1862]

Thank you!

That is correct. Just so you know, I had four molecular biologists in my lab--plus one physicist--working on the answer, and none of us got it right.

You have indirectly contributed to the advancement of science, because now I can get back to my experiment.

John

[DewDiligence]

It’s been discussed on this board many times just how poor a 95% specificity is for testing a low-risk population.

This is an ever-present pitfall for investors in diagnostic companies: PR’s that tout a 90%+ specificity are designed to impress, but they ought not to.

(The actual situation is made even worse because the sensitivity of these tests is not perfect; i.e. they will miss some patients who actually have the underlying medical condition.)

In short, standalone diagnostics intended for a low-risk population must have close to a 100% specificity (almost a zero rate of false positives) to be useful in practice.

[Biowatch]

io_io - "False positives"

>>"Imagine that you are a doctor and one of your patients asks to take an HIV test. You assure her that the test is unnecesssary as only one woman out of a thousand with her age and sexual history is infected. She insists, and sadly the test result indicates viral infection. If the HIV test is 95% accurate, what is the change that your patient is actually sick?"<<

io_io, that's the answer I came up with at first, since 50 of 1000 tests would come up with a false positive. However, if 1 in a 1000 will give a true positive (given that is the assumed frequency of true positives), then that means that testing 1000 people will give 50 + 1 = 51 positive results, of which only one will be a true positive.

So, I'd say the correct answer is that there is only a 1 in 51 chance that the patient is actually sick.

[/end picky mode]