Dear Tony,
The Issue:
Some of the published approximations to a given % confidence interval for
number needed to treat (NNT) by reciprocal of the relevant confidence
interval for absolute risk reduction (AAR) are inadequate for general use.
The Stats:
Robert Newcombe recently reviewed the coverage probabilities of different
methods for constructing confidence intervals for the difference between two
independent proportions (such as ARR). NNT confidence interval is the
reciprocal of these. The reference is - Newcombe R. Interval estimation for
the difference between independent proportions. Statistics in Medicine
1998;17:873-890. The there is no generally acceptable closed form (i.e.
easy to calculate) approximation for the relevant interval. The acceptable
methods require computer software to perform some iterative maths.
The Solution:
Go to http://www.camcode.com and download Arcus. There you will see a
calculator for NNT in the miscellaneous section of the analysis menu. The
help file has all references that you might want to quote on the statistical
aspects of the subject.
I hope this helps.
Regards
Iain
Dr Iain E. Buchan
Centre for Clinical Informatics
Institute of Public Health
University of Cambridge
CB2 2SR
Tel/Fax 01223 312229
email [log in to unmask]
-----Original Message-----
From: Anthony Redmond <[log in to unmask]>
To: [log in to unmask]
<[log in to unmask]>
Date: 18 September 1998 07:07
Subject: 95% CI's for NNTs
Andrew, David et al
Many thanks for your thought provoking contributions this week. However
reading these papers has left me with a problem that my inadequate
mathematics will not allow me to conquer alone.
Please be kind if I am asking the obvious but..
Why, when I calculate the CI's using the 'confidence interval derivation
for
proportions' formula in
> http://www.jr2.ox.ac.uk:80/Bandolier/band18/b18-9.html
do the CI results bear no resemblance to the results in for instance,
http://www.acponline.org/journals/annals/01may97/numeric.htm
despite the NNT's themselves being OK. I have repeated this failure with
the results from several papers.
I am quite comfortable with CIs derived from SDs but am struggling with
CI's for the assocation equations.RR /ARR /NNT
I suspect I am being somewhat slow, and that the answer will be either
painfully easy or painfully difiicult, but equally obvious in either case.
Any and all help would be appreciated
Tony
University of Western Sydney, Macarthur
P.O. Box 555
Campbelltown
NSW 2560
Tel ++2 4620 3790
Mobile 0412 509 408
Fax ++2 4620 3792
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|