Showing posts with label parallel distributed processing. Show all posts
Showing posts with label parallel distributed processing. Show all posts

Sunday, July 28, 2013

Pattern Matching in Psychiatric Diagnosis

I first heard about pattern matching and the importance it has in medical diagnosis over 30 years ago.  A friend of mine who was in medical school at the time told me about one of his professors who was always interested in the Augenblick diagnosis or the diagnosis that  could be arrived at in the blink of an eye.  He gave me examples of several diagnoses that could be either made immediately or within minutes based on a set of features that would lead to immediate associations in the mind of the clinician without an extensive evaluation.

I had many encounters in my medical training with the same phenomenon.  I can recall being on the Infectious Disease consult team and being asked to see a patient with ascites for the possible diagnosis and treatment of spontaneous bacterial peritonitis.  The consultant with an expert in Streptococcal infections and after patiently listening to the resident's presentation he asked what we thought of the rash on the patient's leg.  The patient had lower extremity edema with a slightly erythematous hue and a slight exudate in areas.  What was the diagnosis?  Without skipping a beat the consultant said this was streptococcal cellulitis and suggested sending a sample to the lab for confirmation.  It was subsequently confirmed and treated.  Why was the attending physician able to hone in on and diagnose this rash when it escaped the detection of two Medicine residents and two medical students?  He was an Infectious Disease specialist and that may have biased him in that direction but is there something else?

One of the ways that physicians and probably all classes of diagnosticians arrive at Augenblick diagnoses or efficiently clump and sort through larger amounts of information faster is by pattern matching.  Pattern matching is also the reason why clinical training is necessary to become an adequate diagnostician.  That will not happen with rote learning alone.  It is one thing to read about heart sounds and actually experience them and to have that skill refined by listening to hundreds and thousands of normal hearts and hearts with varying degrees of pathology.  Rashes are classic examples and several studies have documented that the speed and accuracy with which dermatologists can make an accurate diagnosis of a rash is much higher than the average physician.  In pattern matching a recognizable feature of the patient's illness triggers an immediate association with the physicians experiences from the past leading to a facilitated diagnosis.

Probably the best conceptualizations of pattern matching comes from the fields of philosophy and cognitive science.  My favorite author is Andy Clarke and his book Microcognition.  He addresses the issue of biologically relevant cognitive science and the model of parallel distributed processing.  A simplified diagram drawn from this model is shown below:


In this case we have a very practical problem of a patient with known bipolar disorder and a question of whether or not they have had a stroke.  In this case the respective clouds (there are many more) represent collection of features of medical diagnoses that may be relevant to the case.  Unlike a textbook, these features represent a lot of varied information including actual events and nonverbal information like the clinicians past history of diagnosing strokes and caring for people who have had strokes.   Each cloud here can contain hundreds or tens of thousands of these features.  These features are unique aspects of the clinician conscious state and the only way to control for variability between clinicians is to assure that physicians in the same speciality have similar exposure to these experiences in their training.  Even in the ideal situation where all specialists have an identical exposure to the same illness there will be variability based on different levels of ability and other capacities.  An example would be a Medicine resident I worked with whose examination of the heart with a stethoscope predicted the echocardiogram results.  It became kind of a joke on our team at the time that all he had to do was hold his stethoscope in the air in a patient's room and it was as good as an ultrasound.

The basic idea in pattern matching is that the clinician immediately recognizes one of the features they know and that allows for a rapid diagnosis or plan based on that feature.   Looking how that works in the hypothetical case we can look at a few features in the map:


 For the purpose of this discussion consider that our patient B is a 60 year old woman with a 35 year history of known bipolar disorder.  She has known her psychiatrist for years.  One day the husband calls with the concern that the patient seems to have developed a problem with communication.  She seems to be talking in her usual voice but he can't comprehend what she is saying.  She does not appear to be manic or depressed.  The psychiatrist listens to the patient on the phone and concludes that she has a fluent aphasia and recommends that they take her to the emergency department as soon as possible.  Ongoing care requires that the psychiatrist talk with the emergency department physician and hospitalist to make sure that acute stroke is high in their differential diagnosis and eventually go in to the hospital and examine the patient to confirm the diagnosis.

Practically all cases of psychiatric diagnosis require some measure of this pattern matching process with varying degrees of medical acuity.  I would go so far to suggest that it is the most important aspect of the diagnosis.  Keep in mind that the pattern matching also applies to the purely psychiatric part of the diagram.  Despite all of the recent criticism and focus on the DSM 5 the elaboration of pattern matching leads us to several important conclusions:

1.  Psychiatric diagnosis is a much more dynamic process than rote learning from a diagnostic manual.  The average clinician should have many more features of diagnoses than are listed in any manual.

2.  Psychiatric diagnosis requires medical training.  There is no way that our psychiatrist in the example could have made the diagnosis of aphasia and remain involved in the diagnostic process to its conclusion without medical training and previous exposures to these scenarios.

3.  The training implications of these scenarios are not often made explicit.  Every medical student, resident and practicing physician needs to be exposed to a diverse population of patients with problems in their area of expertise in order to develop a pattern matching capability.  They can also benefit by asking attending clinicians about how they made rapid diagnoses, but at that level of training the question is not obvious.

4.  Removing physicians with these capabilities from the diagnostic loop reduces the capability of that loop.  The best example I can continue to think of is the primary care process where the diagnosis and ongoing treatment of depression or anxiety depends on the results of a checklist that the patient completes in less than 5 minutes.  This assumes that there is an entity out there called depression that is based purely on a verbal description and pattern matching is not required.  It actually assumes that there is a population of people with this affliction.  Despite all of the hype about how this is "measurement based care" - I don't think that a single person like that exists.

5.  Pattern matching blurs the line between objective and subjective.  There is often much confusion about this line.  Are there "objective criteria" that can be written in a manual somewhere that captures even the basic essence of diagnosing a stroke in a patient with bipolar disorder?  Is there an "objective" checklist out there somewhere that can capture the problem?  Obviously not.  For some reason people tend to equate "subjective" with "bad" or "unscientific".  In the example given and any similar example, the subjective state with the most experience diagnosing strokes is probably the "best" diagnostician - subjective or not.  An "objective" rating scale doesn't stand a chance.

So consider pattern matching to be an important but unspoken part of the diagnostic process.  For obvious reasons it is more important than diagnostic criteria in a manual.  The most obvious of these reasons is that you really cannot practice medicine without it.

George Dawson, MD, DFAPA

Clark A.  Microcognition.  London, A Bradford Book, 1991.