I have enjoyed and learned from the conversation about pre-test
probabilities. I too have asked and have been asked "where do pre-test
probabilities come from?" Here's some more spice to add to the chili:
a. There's a growing body of literature examining the ability of people to
estimate probability accurately, including the ability of clinicians to
estimate disease probability. At risk of oversimplification, it shows that
while we may be OK at relative statements of probability [e.g. disorder A is
more common than disorder B in patients with symptom X], we have more
trouble when stating probability in absolute terms. Our "gut" probability
estimates are presumably from remembered cases, and there can be several
distortions that arise in our memories that affect our estimates [e.g. our
most recent case]. For more on this, you might start with:
Tversky A, Kahneman D. Judgment under uncertainty. Science 1974; 185: 183 -
187.
Dawson NV, Arkes HR. Systematic errors in medical decision making: judgment
limitations. J Gen Intern Med 1987; 2: 183 - 187.
b. Fortunately, there's a growing body of literature examining the frequency
of underlying disorders found when patients with a given clinical problem
are examined carefully. Examples include the following:
Weber BE, Kapoor WN. Evaluation and outcomes of patients with palpitations.
Am J Med 1996; 100: 138 - 148.
Kroenke K, et al. Causes of persistent dizziness: a prospective study of 100
patients in ambulatory care. Ann Intern Med 1992; 117: 898 - 904.
c. Since these studies exist and might be diagnostically useful if they were
valid, a group of us put together a set of Users' Guides for this sort of
study:
Users' guides to the medical literature. XV. How to use an article about
disease probability for differential diagnosis. JAMA 1999; 281: 1214 - 1219.
d. Are you familiar with the distinction between CER (control event rate)
from a trial of therapy and PEER (patient-expected event rate) that we
estimate for our own patients? I think a very similar distinction holds for
probabilities:
"Disease probabilities" are what the external evidence can show.
"Pre-test probabilities" are what we estimate for our patients, and this
can be informed not only by our remembered cases, but also by such things as
practice databases, population prevalence and incidence data, as well as
research evidence noted above. An EBM Note in the upcoming May/June issue of
EBM explores the strengths and limitations of these different sources.
e. I agree with another writer that sometimes we may start with a pre-test
probability range rather than a single point estimate. If you have the
likelihood ratio nomogram handy [see the CEBM website or the bright green
pocket card of the Sackett et al EBM book], you can show your learners the
following:
Get your group to estimate pre-test probabilities.
Start with one of these estimates on the left, pre-test probability scale.
Anchor straight edge at likelihood of test you're considering in middle.
Pivot straight edge on the LR up and down on the pre-test probability scale
to include all the differing priors your learners mention, to see what
effect this has on the post-test probabilities on the right scale.
Most of the time when I've tried this with learners we end up seeing that
our different point estimates of pre-test probability don't result in
meaningful differences in clinical action, since the resulting post-test
probabilities are usually all on the same side of our testing or treatment
thresholds.
f. Evidence about disease probability comes in the form of proportions, that
represent fractions with cases of the disorders as the numerators. To be
most clinically useful, what should the denominator be?
Many will say that the denominator should be the general population, so
that the fraction represents the frequency of disease in the community. Much
as I have tried to, I can't quite agree. For the clinician seeing a patients
with syncope, who wonders whether or not to consider the diagnosis of
pulmonary embolism, I would have thought the proper denominator should be
patients with syncope, not the community.
Thus, while for prevalence and incidence the proper denominator is the
general population, for clinically useful measures of disease probability
the proper denominator should be patients who have a specified clinical
problem [whether single symptom, a cluster of symptoms and signs, a whole
syndrome or even a disorder].
Hope these 6 bits clarify more than they confuse. I look forward to learning
more from you on this topic.
Cheers!
WSR
W. Scott Richardson, M.D.
Audie L. Murphy Memorial Veterans Hospital *******************
7400 Merton Minter Blvd. The crane's legs
San Antonio, TX 78284 have gotten shorter
T: (210) 567-4808 spring rain
F: (210) 567-4423 Basho
Email: [log in to unmask] *******************
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|