Missed and Erroneous Diagnoses Common in Primary Care Visits

John Brush jebrush at ME.COM
Sun Jan 12 21:24:51 UTC 2014

Dear Bimal,
	This has been an interesting interchange, but we probably need to wrap this up. Perhaps we can discuss at the SIDM meeting or by email off line.
	I’m not sure I understand your analogies and your rejection of the Bayesian approach for diagnosing individual patients. As I said before, Bayesian reasoning is not perfect and is not applicable to every diagnostic situation. But I think that Bayesian reasoning is a useful model for thinking that provides a systematic and more quantitative (albeit often semi-quantitative) approach. I think it captures what is often called “expert intuition” and provides some structure that helps us teach good cognitive processes to trainees. I don’t know of an alternative. In making a diagnosis, we are trying to reason from effect back to cause, rather than from cause to effect. Deductive reasoning won’t work. We have to use inductive reasoning (or abductive reasoning). As Hume said 300 years ago, we use induction because it is our custom.
	As clinicians, we have to reconcile the need to make rapid decisions with the uncertainty of practice. We have to reconcile Systems 1 and Systems 2 thinking. We have to reconcile our need to approximate, with the desire to be precise and accurate. We have to reconcile that we are applying knowledge about populations and science to the care of particular individuals. We do the best we can. The most effective diagnosticians are ones who combine knowledge, experience, thoroughness, thoughtfulness, and patient-centeredness with a systematic and orderly approach. Bayesian reasoning describes one systematic and orderly approach. 
	A really good book on this is by Ian Hacking called An Introduction to Probability and Inductive Logic.

John E. Brush, Jr., M.D., FACC
Professor of Medicine
Eastern Virginia Medical School
Sentara Cardiology Specialists
844 Kempsville Road, Suite 204
Norfolk, VA 23502
Cell: 757-477-1990
jebrush at me.com

On Jan 6, 2014, at 3:09 PM, Bimal Jain <bjain at PARTNERS.ORG> wrote:

Dear John, Thank you for your very perceptive comments which touch on a number of important points in clinical diagnosis.
As I see it, Bayesian diagnosis is essentially application of a mathematical theorem, Bayes' theorem, to a real world process, clinical diagnosis. Application of mathematics to the real world is problematic even in simple cases as we see in the following example:
Let us consider the well known arithmetic theorem, 1+1=2. It has been employed for addition of objects of all sorts of objects such as persons, apples, animals, balls etc. for thousands of years, so that it seems to be a universally applicable theorem. Yet we find we are not able to apply it when we add a drop of water to another drop in a teaspoon as they coalesce together. We realize then this theorem has an implicit assumption that obects to be added need to retain their individual separate status after addition for it to be applicable. The fact that it is applicable to so many other objects is irrelevant to its non-application in the water drop situation.
Similarly, Bayes' theorem, derived from axioms of probability theory is applicable with success to a number of situations such as in epidemiology, life insurance as it leads to long run accuracy or gain. But, and this is the important point, this long run accuracy is attained at the cost of errors in individual instances, which is acceptable in epidemiology, life insurance. It is an inherent feature of probabilistic application that it only assures long run success and not success in every individual instance.
In clinical diagnosis, on the other hand, our aim is accuracy of diagnosis in every individual patient, which simply cannot be achieved by a probabilistic approach. This fact was neither appreciated by Ledley, Lusted in their 1959 paper on application of Bayes' theorem to diagnosis nor in subsequent discussions. A Bayesian approach has continued to be advocated, it seems to me simply by habit and custom and due to simplicity and elegance of Bayes' theorem. This situation is similar to belief in planetary orbits being circular purely on mathematical grounds of beauty and symmetry of a circle as a geometrical figure for over 2000 years. It was only corrected in the 17th century when Kepler actually observed orbit of Mars and found it to be elliptical. Similarly, I believe, we need to look at diagnosis in actual practice and see how diseases are actually diagnosed and if a Bayesian approach could improve accuracy. I have given several examples where diagnosis is done in a non-Bayesian manner such as diagnosis of acute MI in a young woman with acute EKG changes, physicians reading EKGs diagnosing MI from EKG changes without knowledge of prior probabilities.
It is well to heed Einstein's words in this regard which I quote in full:
Concepts which have proved useful for ordinary things easily assume so great an authority over us, that we forget their terrestial origin and accept them as unalterable facts. They then become labelled as ' conceptual neccesities', a priori situations etc. The road of scientific progress is frequently blocked for long periods by such errors. It is therefore not just an idle game to exercise our ability to analyse familiar concepts and to demonstrate the conditions on which their justification and usefulness depend.In this way they are deprived of their excessive authority. Concepts which cannot be shown to be valid are removed. Those which had not been coordinated with the accepted order of things with sufficient care are corrected, or they are replaced by new concepts when a new system is procuced which, for some reason or other, seems preferable.
I hope we start doing conceptual analyses of probability and clinical diagnosis and look more carefully at diagnosis in actual practice.



HTML Version:
URL: <../attachments/20140112/160ecb0d/attachment.html>

More information about the Test mailing list