Thank you to all people who answered my query. Here are the answers I got.
Roberto Ferrara
------------------------------------------------------
You need to have a rough idea what the sensitivity and
specificity might be. You then decide how wide you want
the confidence interval to be. Tis is a function of the
sensitivity and the sample size, hence you have the sample
size.
When you analyse the actual data you may need the exaact
confidence interval for the proportion, because these
proportions tend to be close to 1.0. A free program on my
web site will do this if your own software won't.
Martin
------------------------------------------------------
Martin Bland
Prof of Medical Statistics
Dept of Public Health Sciences
St. George's Hospital Medical School
London SW17 0RE
England
Tel: (+44) (0)20 8725 5492
Fax: (+44) (0)20 8725 3584
Email: [log in to unmask]
Web pages: www.sghms.ac.uk/depts/phs/staff/jmb/jmb.htm
------------------------------------------------------
Try Altman 1991 Practical Statistics for Medical Research
page 458 and following
Dr ZJ Nugent
DHSRU, 9th Floor
Dundee Dental Hospital & School
Park Place
DUNDEE DD1 4HR
Scotland
Telephone 01382 425757
Web http://www.dundee.ac.uk/dhsru/
The obvious thing to do is to use tables to ensure the
precision of a single proportion, and based on estimates of
what you think sensitivity and specificity will turn out to
be, ensure that your sample is large enough to give narrow
(specificed by you) 95% confidence intervals around
estimates of these rates.
Ruth Pickering
----------------------
Dr Ruth M Pickering
Lecturer in Medical Statistics
University Of Southampton
[log in to unmask] tel +44(0)1703 796565
fax +44(0)1703 794460
Dear Roberto,
My suggestion would be to size your study to obtain a given precision on
the confidence intervals.
You can use nquery for proportions.
Best regards,
Jean Lecot
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|