Dan, as you know it, calculatation of the probability of a disease is only useful if it will allow us to make decisions i.e. to act (that is, to convert the probabilities expressed on the continuum scale into categorical "yes/no" decisions). How high probability is "high" or, how low is "low", as you are well aware, is not only function of LR of a diagnotic test, but also of benefits and harms of treatment under consideration. As a classic teaching goes, "don't order a diagnostic test if it will not affect your decision" (or, as we showed, "don't order a diagnostic test if harms of treatments outweigh its benefits"). This is, of course, where classic Pauker& Kassirer's "threshold" model comes handy. Some years ago, we modified Fagan's nomogram (based on Paul Glasziou's work) to allow easy calculation of "rule-in" and "rule-out" thresholds (see Djulbegovic B and Desoky A. Equation and nomogram for calculation of testing and treatment thresholds. Med Decis Making 1996;16:198-199) (attached)
Note that the nomogram requires use of NET benefits and harms. As many different summary measures may be used to express benefits and harms, one can adopt the threshold model accordingly (see for example, Djulbegovic B, Hozo I and Lyman G. Linking Evidence-based Medicine Therapeutic Summary Measures to Clinical Decision Analysis. MedGenMed, January 13, 2000 http://www.medscape.com/Medscape/GeneralMedicine/journal/2000/v02.n01/mgm0113.djul/mgm0113.djul-01.html)
let me know if you think the nomogram attached can be useful in teaching/at bedside
thanks
ben
-----Original Message-----
From: Evidence based health (EBH)
[mailto:[log in to unmask]]On Behalf Of Dan Mayer
Sent: Sunday, August 14, 2005 10:24 AM
To: [log in to unmask]
Subject: Re: visual display of Bayesian logic
Giuseppe and the list.
I agree wholeheartedly with the sentiments of the letter. However, I
don't see the log transformation as being more intuitive than the
current (very difficult) use of odds and likelihood ratios. Personally,
I also thought that they were counter-intuitive when I first learned
them from Dave Sackett's book in the early 1990's. However, after
teaching Sensitivity and Specificity (terms that are truly
counter-intuitive; SpIN and SnOUT) for years, it finally dawned on me
that LRs are actually very intuitive.
Consider the odds of disease as D+ / D-. This ratio is how many people
are likely to have the disease to how many are not likely to have the
disease before the test is applied. It is pretty easy for patients (and
probably health care providers) to think of this; "of people with your
set of signs and symptoms, for every 3 who have the disease there will
be 2 who don't." This ratio is multiplied by another ratio (the
likelihood ratio) to give a final ratio. "After doing the test, for all
those with a positive test, for every six with the disease there will be
one who (has a positive test) and doesn't have the disease. This new
ratio; D'+ / D'- is the post test odds. The same would apply for a
negative test.
To me that is pretty intuitive. And another plus to this method is that
the LR is a good way to appreciate the "strength" of a test, analagous
to the Relative Risk or Odds ratio as measures of risk. As far as the
actual usefulness of the test, for probabilities in the mid range (20 to
70%) a large likelihood ratio is needed to really change the
probabilities significantly. However, for pretest probabilities in the
high or low range, we are just as happy using tests with much lower LR's
(positive between 2 and 5 and negative between 0.5 and 0.2). For high
pretest probability, these will nudge our probability up just enough
(make us just a little more certain, perhaps enough to be satisfied to
proceed) if positive and should get us to do more investigation if
negative (make us more uncertain and less satisfied to proceed with the
treatment for the diagnosed disease).
So, my feeling is to use the Fagan nomogram. Warren Browner from
Pacific Medical Center in San Francisco made a very nice "slide rule"
that can be used to solve this equation. And finally, most PDAs have
programs that give post-test probability when pretest probability and
LRs are entered.
I hope this helps.
Best wishes,
Dan
****************************************************************************
Dan Mayer, MD
Professor of Emergency Medicine
Albany Medical College
47 New Scotland Ave.
Albany, NY, 12208
Ph; 518-262-6180
FAX; 518-262-5029
E-mail; [log in to unmask]
****************************************************************************
>>> "Dr. Giuseppe Giocoli" <[log in to unmask]> 08/14/05 2:02 AM >>>
I would like to read this list's comments to the Lancet letter by Van
der
Ende et al about "applying Feynman-Tufte principle to to clinical
logic, to offer a visual representation of Bayesian logic".
J.Van den Ende et al. The trouble with likelihood ratios (Letter).
Lancet
vol 366 page 548 August 13, 2005
This means to represent likelihood-ratios as visual displays in order
to
permit clinicians to apply Bayesian logic without formal calculations.
In my opinion this sounds suggestive, but it even more complicated than
the
traditional way of presenting and using LRs. Perhaps the very trouble
with
LRs is the difficulty of carrying out studies on accuracy of diagnostic
tests.
Thanks,
Giuseppe Giocoli
GdL EBM AMCLI
Via Sarca, 19
25010 Desenzano d/G (BS) Italia
Dr. Giuseppe Giocoli
Via Sarca 19
25015 Desenzano d/G (Italia)
-----------------------------------------
###########################################################################
## This transmission may be confidential or protected from disclosure and
is only for review and use by the intended recipient. Access by anyone else
is unauthorized. Any unauthorized reader is hereby notified that any
review, use, dissemination, disclosure or copying of this information, or
any act or omission taken in reliance on it, is prohibited and may be
unlawful. If you received this transmission in error, please notify the
sender immediately. Thank you. #########################################
####################################
|