Friday, 15 October 2010

Meta-analysis

Often, perhaps usually, a treatment will make some people better, do nothing for others, and make some people worse.

When a treatment makes about as many people worse as it makes better, then meta-analysts say it doesn’t work.

But often it *does* work - for *some* people. Indeed, it might work very well indeed for them.

The other people - those for whom it makes no difference or makes worse - just need to stop using it - then everyone benefits.

It is just a matter of using patient response as a feedback loop.

Is that really so difficult to understand? - yet I can never get this point across to bio-statisticians/ epidemiologists, meta-analysts and the like…

3 comments:

Anonymous said...

"Is that really so difficult to understand? - yet I can never get this point across to bio-statisticians/ epidemiologists, meta-analysts and the like…"

But this is not difficult to understand. Indeed, it is just common sense.

Which reminds me; I had a teacher who used to say, "The only trouble with common sense is that it's not the common".

Mike Kenny said...

Apropos your post:

http://www.ribbonfarm.com/2010/09/28/learning-from-one-data-point/

xlbrl said...

In a bureaucratic state, a treatment that will prolong the life and well-being of ten terminal patients is not sufficient if that treatment will kill one terminal patient. And no one will be permitted to choose for himself.

A bureaucrat is not paid to understand. Our original mistake was in permitting him any access to decisions.