The Patient Doctor posted about a
website called MyElectronicMD that allows patients to diagnose their own medical problems. Such programs use various types of "artificial intelligence" to sort through signs and symptoms of disease and come up with what we in medicine call a differential diagnosis (abbreviated DDx). This is a wonderfully obtuse term used to describe a list of diseases likely to have caused a given constellation of such findings.
I thought it would be interesting to test it and see how well it worked. I clicked through its decision tree and entered information about the following imaginary patient (at each stage the user is given choices that narrow down the clinical picture):
- Step 1) Male
- Step 2) Chest symptoms
- Step 3) Chest pain with shortness of breath
- Step 4) Chest pain made worse with breathing, pain and tenderness at junction between rib and breast bone, shortness of breath- difficulties breathing
The following differential diagnosis was given:
- Costochondritis (inflammation of the cartilage between the rib and sternum)
- Pleurisy (inflammation of the lining of the lung usually caused by a virus)
- Pectoral myositis (inflammation of the overlying muscle of the chest)
- Pneumonia
- Pneumothorax (collapse of a lung or lung segment)
Please note that this is the very first (and only) fictitious patient I entered into this website's algorithm.
If this program is typical of such efforts, then I probably won't have to hang up my stethoscope anytime soon.
This DDx seems rather inadequate for several reasons. First of all, two exceedingly important and life-threatening possibilities were left out: acute coronary syndrome (which includes unstable angina and myocardial infarction) and pulmonary embolism (a blood clot in the lung). These are major omissions that would cause me to downgrade my evaluation of even a third year medical student let alone an intern or resident.
Why is this important? Because in establishing a DDx, it is axiomatic that one includes both the most likely and the most life-threatening diagnoses at the top of the list. We always tell our students and housestaff that "if you don't include it in your differential, you'll never make the diagnosis". To leave out likely or life-threatening possibilities means you'll never have the opportunity to correctly diagnose and appropriately treat the patient.
Some experienced clinicians may point out that this imaginary patient had "pain and tenderness at junction between rib and breast bone" suggesting that his problem is less serious. Tenderness, when used by clinicians refers to pain specifically elicited upon pushing on the area as opposed to simple subjective pain in the region (a distinction the average layperson using this website may not be familiar with). Most serious heart and lung problems are
not accompanied by chest wall tenderness thereby putting these problems lower down on the list.
However, there are studies that show that even in the case of proven acute coronary syndrome, some 8% of patients
still complain of chest wall tenderness making this symptom less helpful in discriminating serious from not so serious disease.
In addition, although the program has a nice, easy to use interface, some of the decisions the user has to make offer selections for which their may be
no optimally realistic choice. In this example, at stage 4, the user is required to choose among 17 possibilities none of which are mutually exclusive nor are they all-encompassing. Ultimately, the user has to make a choice that may not be at all indicative of what he or she is actually experiencing. Try my example to see what I'm talking about.
The data being entered may therefore be of poor quality leading to the "garbage in, garbage out" phenomenon. The program's results, though represented as being rather precise may be quite poor. Such disguised imprecision may give a very false sense of security to the user.
Lastly I have with this demonstration is the inclusion of
pectoral myositis. Now I don't doubt that histologically-proven cases of this entity do occur, but I would have to put this diagnosis in the category of a
zebra. This is a whimsically applied term used to describe exceedingly
unlikely diagnoses. It's frequently said that when hoof beats are heard, a seasoned clinician attributes them to a horse whereas a medical student imagines they're made by a zebra.
Including such zebras within the DDx may be a source of amusement to attendings during teaching rounds but can definitely be distracting and can lead clinicians down very fruitless pathways.
My Electronic MD may be cute and amusing and a successful anchor for all sorts of online advertising but this diagnostic algorithm is not only inadequate but may be dangerous as well. As one might expect, the website contains the perfunctory disclaimer (Final diagnosis can only be determined by your physician or other health care provider blah blah blah).
To me, however, the process of medical diagnosis is somewhat more than a parlor game or version of reality TV. Real patients are going to use such websites and may very well base decisions regarding their own health on their results. It seems to me that the purveyors of such delightful diversions should make it their responsibility to validate them in every way possible. I have little doubt that if such programs were ever marketed professionally to physicians, they would be considered by the FDA to be medical devices and would therefore require FDA scrutiny and approval.
Should their be any less such scrutiny because they're marketed to the lay public for fun?
Labels: Computers in Medicine, Medical Technology
