I thought members will be interested in this response sent to me by Dr Doggett.
Cheers
Amit Ghosh
> Dear Dr. Ghosh,
>
> Forgive the delay, but I just returned from vacation and noticed your
> questions concerning diagnostic testing. The responses that I saw
> posted did not really seem to hit the nail on the head. There is
> widespread understanding that post test probabilities (also called
> predictive values) are strongly influenced by pretest probabilities (or
> prevalence), and that Bayes' theorem explains the mathematical
> relationships. Unfortunately this has also lead to the equally
> widespread but erroneous belief that because sensitivity and specifity
> (also called likelihood ratios) are not affected by pretest
> probabilities they must be constant for a given diagnostic test. In
> fact, they are not constant over different test populations. For
> progressive diseases, most diagnostic tests are more sensitive for later
> stage disease than for early disease. Thus, sensitivity changes
> according to the spectrum of disease severity in the test population.
> And for an individual patient, a test will be less sensitive early on
> and more sensitive later. Also, specificity of a test changes according
> to the prevalence of comorbidities with confounding or overlapping
> symptoms. Thus, predictive values as well as likelihoods are only
> reliable if they were derived from a population similar to the
> population at hand in terms of disease prevalence, spectrum of disease
> severity, and prevalence of comorbidities with confounding symptoms.
> This explains why a typical case-control diagnostic test study
> overestimates predictive values, sensitivity and specificity. The
> prevalence is usually artificially set high by choosing the same number
> of cases and controls. The spectrum of disease severity is usually
> restricted to well defined disease cases; and the controls are chosen to
> have no overlapping symptoms from comorbidities. If any of this has
> been helpful to you it might also be useful for others, so feel free to
> post it. Some references discussing these problems are given below.
>
> Ransohoff DF and Feinstein AR, Problems of spectrum and bias in
> evaluating the efficacy of diagnostic tests, N Engl J Med 1978
> 299:926-30.
>
> Fletcher RH, et al., Clinical epidemiology: the essentials, Williams &
> Wilkins, Baltimore, Philadelphia, London, 1988, pp. 51-64.
>
> Gann PH, Prostate-specific antigen screening for prostate cancer: issues
> involving test validity, Endocr Related Cancer 1996 3(3):179-89.
>
> Brenner H and Gefeller O, Variation of sensitivity, specificity,
> likelihood ratios and predictive values with disease prevalence, Stat
> Med 1997 16(9):981-91.
>
> David L. Doggett, Ph.D., Medical Research Analyst
> Health Technology Assessment and Information Service
> ECRI, a non-profit health services research organization
> 5200 Butler Pike, Plymouth Meeting, PA 19462 USA
> (610) 825-6000 ext 509, FAX (610) 834-1275
> [log in to unmask]
>
> > -----Original Message-----
> > From: amit kumar ghosh [SMTP:[log in to unmask]]
> > Sent: Thursday, August 06, 1998 5:43 PM
> > To: [log in to unmask]
> > Subject: Dilemna of diagnostic testing of rare disorder.
> >
> > Dear all,
> >
> > While using a test for diagnosing the etiology of abdominal event
> > i.e, CT
> > scan for abdominal pain one wonders how effective is our math pretest
> > odd X LR =
> > Post test odd , useful in the diagnosis of rare disorders. Often rare
> > disorders
> > like aorto-enteric fistula as a cause of lower GI bleed are mentioned
> > in grand
> > rounds as a list completion diagnosis. Everyone things that the
> > person who made
> > the comment was really smart. However when it turns out the case is
> > actually
> > a-e fistula one wonders what is the best test, as 2 CT scan done
> > previous to
> > the current one were negative. In an individual patient with pretest
> > prob say Y
> > % the LR's and post test prob come out lower, earlier in the course of
> > disease.
> > Despite the notion that a diagnostic test should hold good across the
> > entire
> > range of symptom, in practice we find ourselves missing these cases
> > despite
> > doing the right diagnostic test. The application of Gold Standard
> > often does not
> > arise till very late in the course of disease or only at autopsy.
> > Doe this mean that the PPV changes in the same patient at different
> > times? Does
> > False negative rate of a test decreases with advancing disease? I
> > would think
> > so as we often diagnose a lot of condition pre-terminally using the
> > same test
> > we used a month ago!
> > Any alternative suggestions as to what kind of diagnostic test one
> > should be
> > applying in these case. Baysenian theorem doesnot always help in these
> > cases.
> >
> >
> >
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|