Dear all
I have a problem gettting to grips with setting cut offs for
screening tests from survey data. Can somebody advise?
We have survey data on disease prevalence (diabetes/hypertension) in 2
African communities.
There are data on continuous variables (body mass index and waist
measurement amongst others). I want to devise a cut-off value for
screening purposes that will decrease the number of people from whom I
have to take blood/measure the blood pressure after screening them.
We have explored the data, creating ROC curves for thresholds of
screening test values. I have also looked at positive predictive
values against these cut-offs.
These data give me useful information, for example if I screen the top
x% by waist measurement, what the sensitivity, specificity and PPV
would be. However, I feel that I should be able to optimise these
things by *modeling* in some way.
When I have my cut-offs, I intend re-testing the resultant
thresholds for screening in another African population.
Thanks in advance of anyone who can advise,
Regards
Terry
*********************************************
Dr Terry Aspray
Wellcome Laboratories
Royal Victoria Infirmary
Newcastle upon Tyne
NE1 4LP
Great Britain
tel +44 (0)191 222 5407 fax +44 (0)191 261 1763
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|