[EXTERNAL] [IMPROVEDX] Bayesian diagnosis

Ross Koppel rkoppel at SAS.UPENN.EDU
Wed Jul 30 14:39:08 UTC 2014

Wonderful post.  Thank you Bimal (and thank you Ted).

Two notes to add:

1. Aristotle also put a dying man on a scale to see how much lighter he 
would weigh when his soul left him.   But if the soul is lighter than 
air (goes to heaven?) then the man should be heavier upon death  :)

2. Aristotle, in explaining why people had such longer intestines than 
other animals,* concluded that the extra length was there so that man 
could spend more times between meals.  Why?  So that many could spend 
more time thinking!  Clearly he failed to predict video games, couch 
potatoes, and Fox news.


* 1. Seems unlikely that the good Macadonian philosopher ever disected larger animals.

Ross Koppel, Ph.D. FACMI
Sociology Dept and Sch. of Medicine
Senior Fellow, LDI, Wharton
University of Pennsylvania, Phila, PA 19104-6299
215 576 8221 C: 215 518 0134

On 7/30/2014 8:44 AM, Jain, Bimal P.,M.D. wrote:
> Hi, everyone,
> Here are some more thoughts on clinical diagnosis:
> 1.Even though clinical diagnosis is performed daily by physicians all 
> over the world, its method in actual practice is not known precisely.
> 2.Without knowing about the method actually used, a probabilistic 
> (Bayesian) approach has been proposed for diagnosis.
> 3.It has not been adopted in actual practice to any significant 
> extent, as I have pointed out earlier.
> 4.The current situation regarding diagnosis is similar to that 
> regarding motion in the seventeenth century in many respects.
> 5.Even though everyone had encountered falling and  moving bodies at 
> that time, the precise laws governing their motion were not known 
> precisely.
> 6.The prevailing view was Aristotelian in which heavier bodies fell 
> faster than lighter ones and every motion required a mover or force to 
> maintain it.
> 7.There were certain observed facts however which cast doubt on the 
> Aristotelian view such as absence of a mover or force which maintained 
> motion of an arrow in flight.
> 8.It was Galileo who first discovered the laws of falling and moving 
> bodies by a process which essentially created the scientific method.
> 9.The scientific method, which was perfected by Newton, is universally 
> recognized and applied as the most reliable method of gaining true 
> knowledge about any phenomenon.
> 10.The decisive step taken by Galileo was to study actual motion and 
> not develop ideas about it from preconceived notions as was done by 
> others before him including Aristotle.
> 11.For example, Aristotle believed it was ‘natural’ for a heavy body 
> to fall faster than a light one and developed his theory of falling 
> bodies on its basis.
> 12.Galileo started by first dissecting and analyzing actual motion by 
> employing very simple but insightful arguments which are described in 
> his great book Discourses on the two new sciences.
> 13.For example, he asked if a heavy and a light body tied together 
> fell faster or slower than each body falling separately. Also, what 
> happens to motion of a perfectly smooth metal ball rolling on a highly 
> polished smooth floor.
> 14.He then performed experiments to test ideas gained from his 
> analysis. Some of them like dropping a heavy and a light body at the 
> same time from Tower of Pisa  and studying motion of balls rolling 
> down inclined planes are well known.
> 15.From his analysis and experiments on actual motion, Galileo 
> discovered his two well known laws; (a) All bodies, regardless of 
> weight fall at the same rate (b) Motion in a moving body is maintained 
> by inertia and therefore does not require a mover or force.
> 16.I suggest we can learn about method of diagnosis in actual practice 
> by employing an approach similar to Galileo’s.
> 17.When we analyze actual diagnosis, two features about it stand out:
> (a)The widely varying clinical presentations and therefore widely 
> varying pretest probabilities of a given disease in different patients
> (b)The basic clinical aim of diagnosing a disease correctly in every 
> individual patient.
> 18.These two features require   we should be able, while diagnosing, 
> to give a correct yes or no answer about presence of a disease in 
> every individual patient regardless of pretest probability.
> 19. In general, a clinical presentation, regardless of pretest 
> probability of disease , is not informative enough to allow us to give 
> a definitive yes or no answer about a disease in a given, individual 
> patient.
> 20.Therefore a disease is usually only suspected and 
> assumed(postulated) to be present.
> 21.A test is then performed to evaluate the suspected disease.
> 22. A disease is diagnosed definitively if a test result with 
> likelihood ratio (LR) of 10 or higher is observed.
> 23. Some examples of diagnosis in actual practice in this manner are:
> (a) Definitive diagnosis of pulmonary embolism in any patient wit5h 
> positive chest CT angiogram (LR 21)
> (b)Definitive diagnosis of acute myocardial infarction in any patient 
> with acute Q wave and ST elevation EKG changes (LR 13)
> (c) Definitive diagnosis of deep vein thrombophlebitis in any patient 
> with positive venous ultrasound study (LR 19).
> 24. In the Bayesian approach, a disease is diagnosed from a post test 
> probability generated by combining a pretest probability and LR of a 
> test result in a given patient.
> 25. In this method, a disease is diagnosed definitively from a high 
> post test probability. How high , 95 or 90 or 85 percent is not clear.
> 26. A high post test probability could be generated if the pretest 
> probability is high and the LR of a test result low say 2 or 3. Would 
> a disease be still diagnosed definitively in such a patient?
> 27. Analytically, a probabilistic (Bayesian) approach appears to pose 
> a lot of problems if it is employed as a method of diagnosis in actual 
> practice.
> 28. Let us now examine by observation and experiment if our ideas 
> about diagnosis derived from analysis are correct or not.
> 29. I suggest, we retrieve medical records of about one hundred 
> consecutive patients  diagnosed with a certain disease, say pulmonary 
> embolism at a large institution.
> 30. Let us note the following points about diagnosis of pulmonary 
> embolism in each of these patients
> (a)Was the presence of a highly informative test result such as 
> positive chest CT angiogram (LR 21) or a ‘high probability’ perfusion 
> lung scan (LR 14) essential for diagnosis in each patient?
> (b) Or was it diagnosed from a high post test probability even if a 
> highly informative test result was absent?
> 31. Experimentally, summaries of these medical records could be given 
> to physicians who would be asked to diagnose pulmonary embolism from 
> information available.
> 32. It is difficult to know in advance what the results in these 
> observational and experimental studies will be. But whatever they are, 
> they will advance our knowledge about diagnosis in actual practice. If 
> they confirm our analytical findings, they would be made more secure 
> and if they do not, we would need to go back to the drawing board and 
> seek other explanations for the results.
> This is a long email but it discusses a number of important issues 
> regarding clinical diagnosis. Please review it and comment. Thanks.
> Bimal
> Bimal Jain MD
> Pulmonary-CriticalCare
> NorthShore Medical Center
> Lynn MA 01904

HTML Version:
URL: <../attachments/20140730/573317c8/attachment.html>

More information about the Test mailing list