Dr. Watson, come here, I need you!
Try to make him feel welcomed. Reigning Jeopardy! champion and IBM robot, Watson, is devouring the material in medical textbooks and journal articles in hopes of landing a job as a physician’s assistant.
The trainee was sequentially presented the details of a fictitious patient: there’s an eye problem; vision is blurred; the family, living in Connecticut, has a history of arthritis. The trainee’s initial response was uveitis. More clues and the diagnosis was changed to Behcet’s disease until finally the trainee settled on Lyme disease. How sure was this seemingly hasty student of medicine? Seventy-three percent sure.
Medical trainees and doctors are not typically in the habit of quantifying their assessments with such Commander Data-like precision, but this trainee happens to share more qualities with the Star Trek android than the rest of the medical staff. Following its resounding victory on Jeopardy!, IBM’s Watson has been working hard to learn as much about medicine as it can with a steady diet of medical textbooks and healthcare journals. The mock case described above was part of a recent demonstration to the Associated Press showing just how much Watson has learned. The robot’s diagnosis was correct and it identified a link between symptom and cause that was “not common,” as one participating physician called it. After being told the patient was pregnant and allergic to penicillin, Watson suggested treating her with cefuroxine. Its human colleagues agreed. The demonstration was a success, and it is the hope of IBM and many medical professionals that Watson will one day soon lend doctors a helping hand as they perform their daily rounds.
The need for efficient use of medical information becomes more pressing as the amount of information amasses at an exponential rate. Dr. Herbert Chase, a Columbia University medical school professor, tells the Associated Press that it has been 30 years since doctors were last able to keep pace with the growing body of medical knowledge. With only so many hours in an often overwhelming day, doctors are hopeless to keep up with a body of knowledge that doubles every five to seven years. In addition to the struggles of keeping pace, the sheer volume of information presents a separate challenge to share that information effectively. Michael Yuan, a scientist that has worked with IBM, cites a 1999 study that found doctors field more than 1,100 questions a day and fail to answer 64 percent of them. The inefficient exchange of information leads to mistakes in any field. In the medical field, those mistakes could cost lives. A widely-noted–and hotly debated–report published in 2000 estimated that as many as 98,000 people die in a given year from medical errors occurring in hospitals. As the report notes, “that’s more than die from motor vehicle accidents, breast cancer, or AIDS.”
Now imagine Watson on the job. Never one to nap in the library, Watson’s database would be updated daily with the latest in research from potentially all online journals. The director of Health Informatics Center at the University of Southern California, Carl Kesselman, points out the need to track advances in genomics, specialized drugs, off-label uses, and the increasingly finer-grained classifications of diseases. Of course the physicians themselves can find the information, but the online searches would be labor-intensive and time-consuming. A physician’s assistant like Watson with realtime updates could simply be asked Jeopardy!-style questions to find answers or get second opinions. To make the interactions Jeopardy!-style, speech solutions developer Nuance is currently working with IBM to provide Watson speech recognition software customized with medical jargon. Doctors could query Watson’s database on the go by speaking into a handheld device.
At this early stage in its medical education Watson understandably, still makes mistakes. A team of medical students are working with Watson to catch mistakes and try to identify what causes them.
Its knowledge is not limited to science. Watson can also keep an eye on complex treatment guidelines that are often updated so the physician doesn’t have to. It can access medical records as well. However, for access to be completely streamlined they need to be digitized. Unfortunately the medical record digitization seems to be a change hospitals are thus far slow to adopt. Progress is being made, however, by companies like Practice Fusion, a maker of electronic health records (EHR) systems. Combining the latest medical knowledge with the patient’s medical history would give Watson the best chance to catch those rare cases that doctors might be slow to diagnose or miss altogether.
A major break from previous practice is IBM’s plan to include patient blogs among Watson’s data set. Much as they do now on websites such as carepages.com, patients can share symptoms, drug efficacy, drug side effects, relevant family histories, etc. Like a medical wikipedia, the data cloud that amasses could be mined by Watson to pull out obscure relationships that would normally pass under the radar of doctors concerned only with their patients. For example, cross-reactivity between two types of drugs that aren’t taken together very often. In essence, Watson would be conducting its own studies without a priori goals or limitations.
But the data is anecdotal, you say? Dr. Chase agrees, but argues that doctors are already using anecdotal data when they take medical histories. The patients’ descriptions are anecdotal, and the doctors don’t listen any less.
To me the patient blogosphere is the most exciting of Watson’s resources. What sort of insights into medicine and disease will we gain simply by blogging about our own experiences? As Wikipedia shows us, there’s truth in numbers. A major challenge to mining those insightful gems is blogs that are easily understandable to Watson. It’s one thing for a doctor to have a one-on-one conversation with Watson and refine his query when he inevitably runs up against misunderstandings. But it’s quite another thing to glean underlying facts from thousands of blogs from all over the world. A universal format for the patients would help, perhaps including a basic list of yes or no questions.
A know-it-all robotic physician’s assistant that you can talk to from anywhere with a handheld device. Reminds me, again, of Star Trek, “Computer, across how many worlds has the epidemic spread?” (Yes, that’s two Star Trek references in the same blog!). But one company would argue that it is years ahead of IBM in bringing AI to the forefront of medical diagnosis support–and her name is Isabel. Isabel Healthcare’s founder Jason Maude’s named the database program after his daughter who as a child was misdiagnosed with chicken pox (instead, she had two rare chickenpox-related complications). Created a decade ago, the company’s mission is to decrease misdiagnoses, and it performs essentially the same functions as Watson: symptoms are entered and the computer sifts the database to produce a list of the most likely causes. Isabel asks questions that the doctor might not think to ask, indicates the gold standard treatment, and lists relevant medical literature.
So, what do we need Watson for?
IBM executives point out that Watson is much faster than Isabel and much better at understanding terms that it hasn’t memorized from a textbook. Watson would know, for instance, that “difficulty swallowing” is “dysphagia.”
Watson has a ways to go before it makes the grade. IBM estimates that they are still a couple years away from making a marketable Watson. Doctors, IBM execs say, should not feel threatened by their fast learning student. The clinician’s role is to practice medicine and Watson’s role is to support the clinician, to act as a library. I have to admit, I’m skeptical. Medical students are a pretty ambitious lot. It may just be that beating two of the world’s best at Jeopardy! on primetime television isn’t enough for Watson. Maybe its done playing games and wants to contribute in a meaningful way to society.
Or, maybe Watson just wants to be called Doctor.