Watson Computer, M.D.

It was announced this week that Watson, the IBM AI computer system that defeated human counterparts Ken Jennings and Brad Rutter on Jeopardy a few months ago, was "hired" by Columbia University Medical Center for a new challenge - diagnosing complex medical conditions.

It really isn't that surprising.  Physicians go through complex algorithms to diagnose patients, just like computers running code.  "Tell me what brings you in today", or "How long has this been going on?".  "Can you describe the pain?".  "Where is the pain located?".  "Does the pain radiate or travel?".   Those of you who are my patients are used to me going through these "checkdowns", to use a sports analogy.  We decide which pieces of data are important, and which can be discarded.  After a few minutes of asking questions and getting quality answers, we have formulated a short list of possibilities, called a differential diagnosis.  A physical exam adds in more data, pinpointing the diagnosis or eliminating certain diagnoses from the list. Finally, if we are unable to make a diagnosis, we may order some tests, like blood work or an xray.  Then, 95+% of the time, voila.  The patient is diagnosed, treated, and improves (hopefully).

A computer can be trained to go through the same algorithms, go through the same checkdowns, per se, and order the same tests. And I am sure they can make the same diagnoses as well, maybe even pushing our 95% success rate.  Despite this, Watson and his computer brethren aren't going to take over medicine any time soon.

A computer cannot do a physical exam.  A computer can't (yet) tell inflection of voice, get a sense that the patient is in obvious discomfort, or understand certain human behaviors.  A good example is a 20 year old male coming in to see me with his parents.  20 year old males normally don't come to the doctor, much less with their parents.  This is a subtle yet important piece of information, one I obtain the moment I walk in the exam room.  It changes the way I approach the visit.

I am sure that Watson will be fantastic at helping diagnose really complex patients.  But for every day medicine?  Not yet.  The art of medicine, to efficiently, compassionately and correctly diagnose and treat a patient is something which requires a human to human interaction, between the doctor and the patient.  Nowadays, patients can research symptoms on the Internet to "diagnose" themselves.  The problem is that they can't do it correctly.  Patients call our office, wanting treatment for pink eye, but when we see them, they really have allergic conjunctivitis.  Or, a patient thinks they have heartburn, but after we evaluate them, we diagnose them with angina, a telltale sign of heart disease.

One of my famous teaching points is that common things are common, uncommon things are uncommon, and rare things are rare.  The vast majority of what we see is common.  Doctors don't need to memorize all the esoteric data that Watson has in its neural net to practice medicine, we just need access to it. This is why physicians shouldn't feel threatened by Watson getting his medical degree.  We should embrace these technologies because the practice of medicine is a gray science, with endless variables.  It's never black and white, and it constantly changes.  6.4 billion human machines, all similar, but none the same.  Systems like Watson can help us find patterns in disease presentation across different populations, allowing us to better diagnose patients.

My practice has used computers since 1997, when we transitioned from a paper to an electronic medical record (EMR) system.  Our information systems keep our patient data organized, remind us when patients are due for screening tests, tell us if we accidentally prescribe a medicine to someone who is allergic, and, most importantly, makes sure our notes are legible.  I can't imagine practicing medicine without these tools.  They are unbelievably useful.  I think every medical practice should adopt these technologies to improve patient care, despite the fact that implementation can be cost-prohibitive.

I am looking forward to seeing what Watson can do.  But, at the end of the day, Watson can have his massive trove of medical knowledge.  I'll take my smaller, more useful noggin, my Dell server with our patient database, and Mr. Grzyb's amazing mini-kolackys at Christmas time.  Watson doesn't know what he's missing.