Med School to use Jeopardy’s Watson for symptom tracking

Watson, the Jeopardy-winning computer system, was named for IBM founder Thomas Watson, not for Sherlock Holmes’ famed right-hand man.

Yet Dr. Watson might be proud of his 21st-century namesake, as researchers at Columbia University Medical Center are working to adapt his technology into a tool to assist physicians and diagnose patients.

The researchers are working with IBM to create a virtual assistant that could help doctors to diagnose and potentially treat patients, which medical students and residents at CUMC will begin testing in as little as three to six months.

Watson processes questions in natural language by analyzing the contexts in which the key words in a question most frequently appear. According to Dr. Herbert Chase, a professor of biomedical informatics who is working on adapting Watson, this same process can apply to medical diagnosis with the aid of a database of medical terms.

You can give the system “four seemingly unrelated symptoms, and Watson can figure out that those four appear with the highest frequency in a paragraph that has the word Lyme disease,” Chase said.

Chase envisions a role for Watson in alleviating pressure on physicians, who are often strapped for time in managing the many uncertainties of each patient’s condition.
“If we simply give the doctors answers, their practice will improve,” Chase said.

Watson may also be able to assist in determining the best treatments for individual patients as well. Treatment decisions are particularly challenging for physicians, with enormously complex factors influencing what procedures or drugs are best for each patient.

Chase even hopes the Watson technology will be able to extract useful data about side effects from the vast world of patient-submitted information on blogs and discussion boards.

“A physician does not have the capacity to crunch all that data,” he said.

Medical school curricula could also feel the effects if the Watson technology becomes widespread, allowing a paring down of the tens of thousands of terms students must memorize. Chase described it as encouraging future physicians to “spend a little less time memorizing and a little more time analyzing.”

Noémie Elhadad, an assistant professor and specialist in natural language processing, said Watson’s power is the product of a new approach to computational linguistics that’s become prevalent in the past two decades, in which computers use statistical analysis to learn for themselves.

But some challenges remain in applying Watson to a clinical context.

“The clinical domain is completely different from a game,” said Elhadad, citing the obviously greater expectations of certainty and accuracy.

“You want the physician to trust the system, so you don’t want a black box,” she said, noting that the computational methods for determining a likely diagnosis and explaining that choice can often be different.

Chase also expressed concern about preventing Watson from getting in the way of the doctor-patient relationship.

“It’s a very personal, private activity, and so it’s not clear in my own mind how this is going to work,” he said, referring to the consultation between doctor and patient.

Both Chase and Elhadad emphasized that Watson would serve as a resource to help process the enormous complexity of modern medicine, while leaving decisions in the hands of physicians and patients. With Watson, “You have now a pool of experts, and one of them happens to be a machine. ... It’s not like machine against humanity,” said Elhadad.

“The idea that Watson ... can actually go find those pieces of information that are going to have a major impact on your patient’s life—it’s the holy grail,” Chase said.
henry.willson@columbiaspectator.com