Friday, February 18, 2011

Will Doctors Be Relevant in the World of Watson?

Like so many other people, I watched Jeopardy this week with rapt attention as I saw IBM's uber-computer Watson clean the clocks of Jeopardy's two greatest champions, Ken Jennings and Brad Rutter. My reaction--that of fascination bordering on the point of awe simultaneously mixed with a not insignificant amount of apprehension--was probably also shared by a good number of viewers. Did we witness the creation of something worthy of comparison to the Brandenburg Concertos, or Frankenstein? No way to know at the moment. Check back with the Billy Rubin Blog in about four or five decades. I promise to do a follow-up entry at that time.

While watching the informational clips during the Jeopardy show in the first round, where various Big Bluers explained the mission and the challenges of developing Watson, I was struck by how Watson could be used in medicine. Give it a patient who explains their symptoms, then feed Watson with basic lab or other clinical data, and you could easily produce a computer doc that would rival the best in the biz. IBM seems to have had the same thought, as Carey Goldberg of the CommonHealth blog notes, as they have partnered with Nuance Communications of Burlington, MA to set up computer systems built on the advances gained from the making of Watson.

Leave aside the rather terrifying fact that this could deprive me of a job in the decades to come, it's not hard to see the ways in which a Doc Watson could outperform a flesh-and-blood physician. Take, for instance, the news story this week about the neurologic event of TV anchor Serene Branson, which earned speculation from ER physicians as well as a Happy Hospitalist as to the diagnosis (for a prescient discussion of the media portrayal, see Gary Schweitzer's HealthNewsReview entry here). Leave aside the dramatic videotape of Ms. Branson for a moment. What's the diagnosis?

The immediate speculation was that Branson suffered a Transient Ischemic Attack or "mini-stroke." To me, this never quite made sense for the simple fact that Branson is young, and TIAs are typically a disease of people in their 6th decade and beyond. Of course, there are outliers, as demonstrated by the sad story of professional baseball pitcher Darryl Kile, who died in his sleep in midseason at age 33 and was found at autopsy to have obstruction of two major coronary arteries, which is extremely unusual for one so young. So although a TIA seemed possible, I thought that the list of possible diagnoses included things like Multiple Sclerosis, and since I'm an ID doc I immediately thought of neurologic diseases associated with HIV. Other docs weighed in on the blogs and several proposed the possibility that this was a complex migraine headache This is now the official diagnosis of the authorities who have cared for her at the UCLA medical center (though, as Happy notes, this pronouncement came from a neurosurgeon, which from a medical standpoint is very odd and not entirely appropriate).

The diagnosis might be spot-on but it's a tough one to make. What could a Doc Watson do that we couldn't? Watson would be able to take the patient's age, take the information from all the data generated by her care (vital signs, physical exam findings, the history of garbled speech, any pertinent information from her family and social history, labs and radiologic tests) and arrive at a mathematically precise risk assessment for each of these various diagnoses, just as it did in trying to answer Alex Trebek's questions. (Yes, I know, it's the other way around in Jeopardy. Let's just move on.)

That is, I can note in a vague way based on my learning in medicine that TIAs are less likely to occur in younger people, but I have no idea the precise numbers; for Watson, I would assume that it would be a small matter to incorporate hundreds of epidemiologic studies allowing it to calculate multiple probabilities for all sorts of diseases. It's not that Watson would be doing anything different than what we do every day when we form differential diagnoses on patients, it's just that Watson would have three huge advantages: it could access considerably more information than we can (we are limited by how much we can read, and then remember, while for Watson that's just a download), it can then take that information and process it orders of magnitude more quickly, and it can describe the likelihood of an uncertain diagnosis with a precision that is virtually impossible for a lone human physician to do in live time while seeing a patient.

I'm not a computer specialist and don't know the inner workings of Watson, but I don't think this is wild speculation on my part. I suspect that we're not far away from being able to feed information into Watson-like programs that will help our diagnostic accuracy increase tremendously in the years to come. I also suspect that there will come a point where Watson-like programs will obviate the need for physicians. Hope I'm retired by that point!

Hat tips to Grunt Doc, Gary Schweitzer, Happy Hospitalist, CommonHealth.


  1. I don't think you need to worry. Medicine is a lot more than just taking a symptom list and assigning probabilities.

    Just for starters, there's the facial expression, tone of voice and other gestures when the patient relates symptoms.

    Then there's the stuff that they are not going to tell you unless they trust you. Like the chronic pain patient who finally revealed that her husband was forcing her to engage in very distressing sexual activities.

    Finally there's is the effect on the patient that comes from the personal relationship between you. That influences everything from how well they comply with treatment to intangibles that come from expectation and caring and are probably in the field of psychoneuroendocrinology.

    I use 95% of my knowledge for the 5% of the time that things are simply not what they seem or are not out of the book.

    A computer that wins at Jeopardy does not worry me. Now a computer that wins at Go, well that may be another matter.

  2. Some other thoughts.

    Watson would be helpful in forming a differential diagnosis, and in offering the "evidence based" treatments. However Watson would not be able to determine how well the treatment would fit the particular individual. Statistics are based on the group and the larger the group the less relevant to the individual. It also is questionable if Watson could sort through the spin that goes into "evidence-based" treatment recommendations.

    Watson would be helpful if there is only one treatment for the condition, like Jeopardy in which there is only one treatment. I don't see Watson helpful in forming treatments for poorly defined conditions which have few effective treatments, ex: fibromyalgia.

    Since a large part of treating chronic disease involves helping the patient make appropriate lifestyle changes I don't see Watson being helpful. Computers make lousy teachers for complex tasks like behavioral change.

    I guess the role of Watson is determined by how one sees medicine. Is medicine the process of assigning the patient to the correct check box and then picking the procedure or prescription that goes with that, or is medicine the art of understanding how the patient's symptoms arise out of the relationships between the patient's internal and external environments and interacting in a way that evokes healing or palliation? Watson could do well at the former, not the latter.

  3. This raises another good question. Will alternative medicine evolve to a point where doctors take the backseat? I know chiropractors are gaining prominence.