Dear Klazien,
In both of your examples you are asking about positive predictive value
(PPV), not sensitivity. PPV is the proportion of patients who test positive
that are also proven positive by the gold standard. Sensitivity is the
proportion of all patients positive by the gold standard who test positive.
While these values are calculated from some of the same numbers, they are
completely different concepts and different numbers. In a 2x2 table, one is
calculated by reading down the columns, and the other by reading across the
rows. These values are related mathematically by Bayes' theorem, so that if
prevalence is known, either PPV or sensitivity can be calculated from the
other. Any good biostatistics text book will have a full explanation of all
of this.
In example 1 the gold standard will be a tabulation of who eventually comes
down with the flu illness after a sufficient followup period. In the 2nd
example the gold standard is the tabulation of who the flu virus can be
isolated from. Which gold standard is preferable depends on the purpose of
the study. If it is possible to harbor the flu virus and not become ill,
your gold standard will depend on whether you are interested in having the
test detect the number of people who actually become ill, or the number of
people who harbor the virus. If it is not possible to have the virus
without becoming ill, then these gold standards will be the same, as long as
followup is long enough for everyone with the virus to develop illness.
There is no one universal gold standard, rather the gold standard is chosen
to be appropriate for the question addressed. Sometimes it is appropriate
for the gold standard to be another test. If that test is not perfectly
accurate, then the result will be only relative to the second test, and true
PPV and sensitivity will not be known. Sometimes if there is no perfect
test, simply following up to observe for the pathology of interest is the
best gold standard; however, this will not be accurate if treatment
intervenes and eliminates some cases that would otherwise have occurred.
Because of these problems it is not always possible to determine true PPV
and sensitivity in clinical situations.
Two good books are:
Clinical Epidemiology, the Essentials, Fletcher RH, Fletcher SW and Wagner
EH (eds.), Williams & Wilkins, Munich, 2nd Edition, 1988.
Medical Uses of Statistics, Bailar JC III and Mosteller F (eds.), NEJM
Books, Boston, 2nd Edition, 1992.
Other books may be more available in Europe.
David L. Doggett, Ph.D.
Senior Medical Research Analyst
Technology Assessment Group
ECRI, a non-profit health services research organization
5200 Butler Pike
Plymouth Meeting, PA 19462-1298, USA
Phone: +1 (610) 825-6000 ext.5509
Fax: +1(610) 834-1275
E-mail: [log in to unmask]
> -----Original Message-----
> From: Klazien Matter-Walstra [SMTP:[log in to unmask]]
> Sent: Tuesday, August 01, 2000 6:25 AM
> To: EBH Discusision list
> Subject: Diagnostic tests
>
> Dear All,
>
> I am working on the subject "diagnostic tests" and the idea of the
> "goldstandart". Now I sometimes cannot find out to "what" sensitivity and
> specivity of a diagnostic test relate to. What I mean is the following:
> When
> you perform a test say to find out if somebody has a flu virus, what
> should
> the test measure ?
> 1. How many people with a positive result will get ill and have a flu ?
> (sensitivity and specivity relate to the outcome "Illness" yes or no), or
> 2. In how many cases with a positive test result can the flu virus be
> isolated (Goldstandart ?)
>
> As i see it sensitivity and specivity in both cases may be very different,
> but also the two cases may be of different clinical value, as many people
> whith a positive test in 2. may not become ill at all.
>
> How is this problem handled and can somebody help me with literature
> concerning this problem, specially what is the "Goldstandart" in 1 ?
>
> Thanks a lot for responses.
>
> With best regards,
> Dr. Klazien Matter-Walstra
> _________________________________________
> Stiftung Paracelsus heute
> Ilgenweidstrasse 3
> CH-8840 Einsiedeln, Switzerland
> tel +41(0)55 4188195, fax +41(0)554188196
> mailto:[log in to unmask]
> http://www.paracelsus-heute.ch
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|