[EXTERNAL] [IMPROVEDX] Bayesian diagnosis

Michael.H.Kanter at KP.ORG Michael.H.Kanter at KP.ORG
Sun Jul 20 00:23:48 UTC 2014

In my view, part of the problem with Bayesian approaches is that of 
obtaining the prior probability.  Most examples attempt to use the 
prevalence of the disease in the population giving the appearance of it 
sounding scientific.  This is probably okay when screening asymptomatic 
people for certain diseases where the physician actually never sees 
individual patients but is instead looking at the population.   Once the 
patient is seen by the physician, the prior probability changes as not all 
patients are alike. 
.  In true Bayesian statistics, the prior probability is a subjective 
probability. (this is also one reason why many statisticians do not feel 
Bayesian statistics are "scientific").  So, if a patient presents with 
certain symptoms, the prior probability should ideally be determined by 
the physician who uses all information about the patient other than the 
various tests/symptoms that are used to modify the prior probability to 
create a final conditional probability that the patient has a disease 
given certain symptoms or test results.  This then becomes an exercise in 
subjectivity where the physicians' intuition, prior experience, and "gut 
feel" may and should be used in determining prior probabilities.   Most 
advocates of a Bayesian approach simply select a prior probability based 
on a few variables and require that all physicians use that prior 
probability.  This is a fundamental flaw, in my view. 
 In the example below (item 6) where Bayesian approaches may be 
problematic due to failure to consider a rare disease, this  failure to 
consider a rare disease can be viewed as simply using the wrong prior 
probability for the more common disease.  Physicians who consider the rare 
disease would have a much lower prior probability of the non rare disease 
than those who do not consider the rare disease. 
Thus, the concept that a Bayesian approach is more objective may not be 
correct.   One can force it to be objective by requiring use of a certain 
prior probability but this then creates other potential errors and in 
theory is less accurate because it may not consider variables that are 
available to the physician seeing the patient but not in the model. 
Advocates of a Bayesian approach seem to forget that the prior probability 
is generally not known but is merely a subjective "guess".   So, comparing 
a Bayesian approach to the usual approaches is something that would be 
real interesting to do yet, one must be careful in defining exactly what a 
Bayesian approach means. 

Michael Kanter, M.D.
Regional Medical Director of Quality & Clinical Analysis
Southern California Permanente Medical Group
(626) 405-5722 (tie line 8+335)
THRIVE By Getting Regular Exercise

NOTICE TO RECIPIENT:  If you are not the intended recipient of this 
e-mail, you are prohibited from sharing, copying, or otherwise using or 
disclosing its contents.  If you have received this e-mail in error, 
please notify the sender immediately by reply e-mail and permanently 
delete this e-mail and any attachments without reading, forwarding or 
saving them.  Thank you.

From:   "Graber, Mark" <Mark.Graber at VA.GOV>
Date:   07/19/2014 10:03 AM
Subject:        Re: [IMPROVEDX] [EXTERNAL] [IMPROVEDX] Bayesian diagnosis

I'd love to see the kind of comparative study Bimal is advocating.  One of 
the suggestions on how to improve diagnosis that seems like it should work 
is exactly what Bimal is advocating: expanding the differential diagnosis 
and weighing the options appropriately.  Hardeep's work found that in 
cases of dx error there was no differential diagnosis listed 80% of the 
time.  Would this be an antidote to dx error, as Jason Maude has suggested 
(Differential diagnosis: The key to reducing diagnostic error?   DIAGNOSIS 
2014; 1:107-109).

Just to make a point though:  The Ledley\Lusted approach was to use the 
full range of conditional probability analysis in diagnosis, which 
includes first deriving all the possible candidate diagnoses, then 
weighing the evidence pro and con, and using tests to adjust probabilities 
to narrow the list down..  It seems that when people refer to Bayesian 
analysis these days they are referring to the narrower usage of Bayes 
theorem to calculate the probability of a SINGLE disease probability. This 
focus on just one entity is exactly the opposite of what Ledley\Lusted 
were advocating, and to the extent that it discourages thinking about all 
the other diagnostic options, could work against diagnostic accuracy.

From: Lorri Zipperer <Lorri at zpm1.com>
Reply-To: Society to Improve Diagnosis in Medicine 
<IMPROVEDX at LIST.IMPROVEDIAGNOSIS.ORG>, Lorri Zipperer <Lorri at zpm1.com>
Date: Thu, 17 Jul 2014 17:05:08 -0400
Subject: [EXTERNAL] [IMPROVEDX] Bayesian diagnosis

**** moderator forwarded ****

-----Original Message-----
From: Jain, Bimal P.,M.D. [mailto:BJAIN at PARTNERS.ORG]
Sent: Wednesday, July 16, 2014 5:35 AM
Subject: RE: [IMPROVEDX] [EXTERNAL] [IMPROVEDX] Patient-Provider
Communication and Diagnostic Error

Great references. Here are some thoughts on a somewhat unrelated topic--
Bayesian diagnosis.

1. Since Ledley and Lusted introduced a probabilistic (Bayesian) approach 
clinical diagnosis in 1959 (SCIENCE 1959), innumerable books and papers 
been written advocating its use in actual practice.

2. However, this does not seem to have happened to any great extent in 
55 years. For example, CPCs in NEJM continue to be discussed in a manner
that is not Bayesian. Also discussions about diagnosis in clinical problem
solving exercises on actual patients published in journals do not employ a
Bayesian approach. In the occasional patient in whom it is employed, it 
not come out looking too good as it seems to suggest an erroneous 
(Pauker, NEJM 1992).

3. There is no published evidence it is superior to the usual method of
diagnosis in actual practice which is about 85 percent accurate (Graber 
2013). There are no head to head trials which demonstrate superiority of
Bayesian over usual method of diagnosis.

4. It is ironic, in this day and age of evidence based medicine, Bayesian
diagnosis continues to be advocated without any evidence for its
If it were a treatment, this issue would have been settled a long time 
by a controlled trial.

5. In all studies on diagnostic errors (Gordon Schiff, Hardeep Singh, Mark
Graber), the method of diagnosis has not been identified as a source of
diagnostic error. How would switching to a Bayesian approach reduce error?

6. One of the major causes of errors found in these studies is a failure 
suspect a disease in patients with atypical presentations. This error has
potential to increase with a Bayesian approach in which a low pretest
probability may be interpreted as low plausibility or low pretest evidence
leading to ruling out a disease without testing.

7. A Bayesian approach appears attractive, it seems, due to consistency 
elegance of underlying Bayes theorem which it employs.But reliance solely 
mathematical criteria without observation can be misleading as seen from 
well known example of planetary orbits. For over two thousand years,
planetary orbits were believed to be circular due to mathematical beauty 
elegance of a circle as a geometrical figure. This belief was found to be
erroneous when Kepler observed positions of Mars and calculated its orbit 
be elliptical.

8. I believe we need to have observational evidence of some sort, probably
from a controlled trial, to decide if Bayesian diagnosis is superior to
usual diagnosis or not. Only if it is found to be superior can it be
recommended for widespread use.


Bimal Jain MD
NorthShore Medical Center
Lynn MA 01904

-----Original Message-----
From: Graber, Mark [mailto:Mark.Graber at VA.GOV]
Sent: Tuesday, July 15, 2014 5:54 PM
Subject: Re: [IMPROVEDX] [EXTERNAL] [IMPROVEDX] Patient-Provider
Communication and Diagnostic Error

Barbara - Thanks for asking a GREAT question.  I think we all know the
answer.   The bad news is that I don't think you'll find a paper that
clearly links effective communication with improved diagnostic accuracy.
The reason is that so far we don't have a good way to actually measure
diagnostic accuracy.  We really need that tool to start making progress in
studying diagnostic accuracy and error.  David Sofen (Palo Alto Foundation
Medical Group) and I have discussed this exact issue a couple of times, 
the possibility of looking to see if there is correlation between
communication style and diagnostic accuracy just amongst all the 
in a large practice.  Hopefully he will weigh in on his own as well.

The good news is that there is PLENTY of INDIRECT evidence to support the
relationship between effective communication and diagnostic quality:

1)   Communication breakdowns are a common cause of diagnostic error.
(Graber et al.  Diagnostic Error in Internal Medicine.  Arch Int Med 
165:1493-99).  Hardeep Singh also has a series of publications on 
errors arising from breakdowns in lab communications, which is a related

2)  As Charlene Weir points out, there is evidence that the diagnosis
emerges from the history alone in the majority of cases.    (Peterson MC,
Holbrook JH, Von Hales D, et al. Contributions of the history, physical
examination, and laboratory investigation in making medical diagnoses. The
Western Journal of Medicine. BMJ 1992;156:163-5.  Wahner-Roedler DL, 
SS, Bauer BA, et al. Who makes the diagnosis? The role of clinical skills
and diagnostic test results. J Eval Clin Pract 2007;13:321-5.)  There is
also a famous Osler quote on this:  "Listen to your patient, he is telling
you the diagnosis".

3  There is substantial evidence that high levels of patient satisfaction
(typically reflecting communication) correlate with improved health 
in general. Articles that David sent me address this point:

§  Doyle, C., Lennox, L., & Bell, D. (2013). "A systematic review of
evidence on the links between patient experience and clinical safety and
effectiveness." BMJ Open, 3. Available at
http://bmjopen.bmj.com/content/3/1/e001570.full. This study finds 
positive associations among patient experience, patient safety, and 
effectiveness for a wide range of disease areas, settings, outcome 
and study designs. It finds that patient experience is positively 
with clinical effectiveness and patient safety, and the results support 
case for the inclusion of patient experience as one of the central pillars
of quality in health care.

§  Boulding, W., Glickman, S. W., Manary, M. P., Schulman, K. A., & 
R. (2011). "Relationship between patient satisfaction with inpatient care
and hospital readmission within 30 days", American Journal of Managed 
17(1), 41-48. Focusing on three common ailments-heart attack, heart 
and pneumonia-the authors measured 30-day readmission rates at roughly 
hospitals and found that patient satisfaction scores were more closely
linked with high-quality hospital care than clinical performance measures.
Hospitals that scored highly on patient satisfaction with discharge 
also tended to have the lowest number of patients return within a month 
all three specified ailments. Overall, high patient satisfaction scores 
more closely linked to a hospital's low readmission rates than a solid
showing on clinical performance measures.

§  Glickman, S. W., Boulding, W., Manary, M., Staelin, R., Roe, M. T.,
Wolosin, R. J., et al. (2010). "Patient satisfaction and its relationship
with clinical quality and inpatient mortality in acute myocardial
infarction", Cardiovascular Quality and Outcomes, 3(2), 188-195. Hospitals
use patient satisfaction surveys to assess their quality of care. The
objective of this study was to determine whether patient satisfaction is
associated with adherence to practice guidelines and outcomes for acute
myocardial infarction and to identify the key drivers of patient
satisfaction. The authors found that higher patient satisfaction is
associated with improved guideline adherence and lower inpatient mortality
rates, suggesting that patients are good discriminators of the type of 
they receive. Thus, patients' satisfaction with their care provides
important incremental information on the quality of acute myocardial
infarction care.

§  Isaac, T., Zaslavsky, A. M., Cleary, P. D., & Landon, B. E. (2010). 
relationship between patients' perception of care and measures of hospital
quality and safety", Health Services Research, 45(4), 1024-1040. The 
rating of the hospital and willingness to recommend the hospital had 
relationships with technical performance in all medical conditions and
surgical care. The authors found that better patient experiences for each
measure domain were associated with lower decubitus ulcer rates, and for 
least some domains with each of the other assessed complications, such as
infections due to medical care.

§  Jha, A. K., Orav, E. J., Zheng, J., & Epstein, A. M. (2008). "Patients'
perception of hospital care in the United States", New England Journal of
Medicine, 359, 1921-1931. This study assessed the performance of hospitals
across multiple domains of patients' experiences and found a positive
relationship between patients' experiences and the quality of clinical 
The authors found that patients who received care in hospitals with a high
ratio of nurses to patient-days reported somewhat better experiences than
those who received care in hospitals with a lower ratio, and hospitals 
performed well on the HCAHPS survey provided a higher quality of care 
all measures of clinical quality than did those that did not perform well 
the survey.

Mark L Graber MD FACP
President, SIDM
Sr Fellow, RTI International

From: Barbara Balik <Barbara at THECOMMONFIRE.COM>
Reply-To: Society to Improve Diagnosis in Medicine
Date: Tue, 15 Jul 2014 10:30:03 -0400
Subject: [EXTERNAL] [IMPROVEDX] Patient-Provider Communication and
Diagnostic Error

I am looking for citations that link effective patient-provider
communication and improvement in diagnosis accuracy.

Effective communication from the patient's view is often described as an
effective listener, understanding the patient's story, and respecting 
and choices

I appreciate any help

Barbara Balik, RN, EdD

Balik.Barbara at gmail.com
www.thecommonfire.com <http://www.thecommonfire.com>

Cell: 651.249.9237
Office: 505.797.8933

Common Fire Healthcare Consulting:
Partnering with healthcare leaders to forge quality and safety outcomes

Moderator: Lorri Zipperer Lorri at ZPM1.com, Communication co-chair, Society 
for Improving Diagnosis in Medicine

To unsubscribe from the IMPROVEDX list, click the following link:<br>
<a href="
" target="_blank">

Moderator: Lorri Zipperer Lorri at ZPM1.com, Communication co-chair, Society 
for Improving Diagnosis in Medicine

To unsubscribe from the IMPROVEDX list, click the following link:<br>
<a href="
" target="_blank">

HTML Version:
URL: <../attachments/20140719/fbbbc41e/attachment.html>

More information about the Test mailing list