Several years ago, we tackled this issue in the setting of P. Vera
diagnosis. You may be interested in the following paper: Djulbegovic B, Hozo
I. Computer program for the diagnosis and treatment of polycythemia rubra
vera.MD Comput 1999 Jan-Feb;16(1):83-9
Erratum in:
MD Comput 1999 Sep-Oct;16(5):10
let me know if this was useful
best
bd
---
-----Original Message-----
From: Andrew Jull
To: [log in to unmask]
Sent: 9/23/2002 7:07 PM
Subject: Re: Confidence intervals and Diagnostic tests
[Andrew Jull] It's an interesting issue, and I agree with the approach
being sold as a sensitivity analysis.
However, what is tested would depend on where the greatest level of
uncertainty lay. For instance, if the greatest level of uncertainity lay
with the pretest probability, then generating worst and best case
scenarios using the lower and upper bounds for the pretest probability
may make sense.
But where you have good quality data regarding the pretest probability,
then the greater uncertainty may reside in the precision of the
likelihood ratios, and you may be better off generating worst and best
case scenarios using the 95% CI for the likelihood ratios.
Of course, you could do both, but then that gets quite complex, and may
just add to the clinician uncertainty, rather than quantifying the
uncertainty of the test.
Andrew Jull
HRC Foxley Fellow
Clinical Trials Research Unit Tel: +64 9 373-7599 Ext 4744
Dept of Medicine Mobile: +64 21 173 7755
Private Bag 92019 Fax: +64 9 373-1710
University of Auckland Email:
[log in to unmask] < mailto:[log in to unmask]
<mailto:[log in to unmask]> >
Note - the University of Auckland is changing its extension numbers.
Beginning 29 November, my number will be 84744.
2. Most health care professionals are resistant to the complex
calculations that you are performing. They want to keep it simple, and
it may make more sense to sell your approach as a sensitivity analysis.
Do this in three steps:
a. Calculate the post test probability as Sackett et al would.
b. Recalculate using a lower bound on the pre test probability and
a lower confidence limit on the likelihood ratio.
c. Recalculate using upper bounds.
You now have an estimate of the post test probability and two
alternatives under worst case and best case scenarios. If you read some
of the stuff published on likelihood ratios, they are sort of implying
something like this, but it would help if they were more explicit and
gave a few good examples.
That's my thoughts, anyway. I am not an expert in Bayesian statistics,
so you might want to consult someone who knows this stuff better than I
do.
Steve Simon, [log in to unmask], Standard Disclaimer.
The STATS web page has moved to
http://www.childrens-mercy.org/stats
<http://www.childrens-mercy.org/stats> .
######################################################################
This transmission may be confidential or protected from disclosure and
is only for review and use by the intended recipient. Access by
anyone else is unauthorized. Any unauthorized reader is hereby
notified that any review, use, dissemination, disclosure or copying of
this information, or any act or omission taken in reliance on it, is
prohibited and may be unlawful. If you received this transmission in
error, please notify the sender immediately. Thank you.
######################################################################
|