IBM's Watson To Help Diagnose, Treat Cancer

Lucas123 writes "IBM's Jeopardy-playing supercomputer, Watson, will be turning its data compiling engine toward helping oncologists diagnose and treat cancer. According to IBM, the computer is being assembled in the Richmond, Va. data center of WellPoint, the country's largest Blue Cross, Blue Shield-based healthcare company. Physicians will be able to input a patient's symptoms and Watson will use data from a patient's electronic health record, insurance claims data, and worldwide clinical research to come up with both a diagnosis and treatment based on evidence-based medicine. 'If you think about the power of [combining] all our information along with all that comparative research and medical knowledge... that's what really creates this game changing capability for healthcare,' said Lori Beer, executive vice president of Enterprise Business Services at WellPoint."

Comments

Re:What could possibly go wrong?

[robot256]

The nice thing is that its result is not just spat out of a black box--it gives a pretty accurate confidence measure, and actually links back to the articles that led it to each conclusion. That means the doctor can go and read them for himself. He may find articles he never would have found otherwise, and become a better doctor for it. It also gives the basis upon which to challenge the computer in court, if it comes to that ("Toronto", anyone?). I think the hope of WellPoint is that it will allow doctors to learn from research faster and more efficiently, so that young doctors learn faster and old doctors stay current. The more the doctors on the front lines of medicine know, the better patient outcomes will be. Nobody is saying the computer is going to ever replace the doctors altogether.

Then the legal department will come and screw everything up, of course, but we can wait a little while before that happens.

Re:What could possibly go wrong?

[Trepidity]

Computers are actually better at certain kinds of diagnosis than the overworked GP, though, and have been for years. In particular, computers are very good at conditional probability, and at combining information from thousands of study results that a typical GP doesn't have time to keep up with. The MYCIN AI system beat most doctors in diagnosing blood infections over 30 years ago, but wasn't adopted in actual medical practice mainly for political reasons.

Re:What could possibly go wrong?

[ColdWetDog]

The nice thing is that its result is not just spat out of a black box--it gives a pretty accurate confidence measure, and actually links back to the articles that led it to each conclusion.

A question and a comment.

First, how do you know this? It really didn't say how it is going to get an 'accurate confidence measure'? Which leads me to my comment - the completely untested assumption here is that the answer lies in the current literature, if only you could wade through it. That isn't at all clear to me. Much of the 'best' data comes from double blinded placebo controlled studies. The big problem here is that the patients are typically carefully selected for having as few other co morbidities as possible. That's very useful from a research situation but makes the data poorly generalizable to the average patient on the street who has no particular interest in being 'simple' (or rational or compliant). The rest of the medical literature is basically crap. "Expert" opinion which turns out to be wrong as often as not. Observational studies which almost always inflate the efficacy of treatments and can only provide correlation. Much of medicine has really never been studied carefully at all.

Next, if they're using any insurance company's billing data, well you might just as well consult rabbit entrails. In the US the vast majority of that data is entered in an obfuscated, outdated and thoroughly whimsical system called ICD-9 (International Classification of Diseases, version 9). The REST of the world with the possible exception of North Korea is on 10 - a system that is much more useful. But even using one of the more sophisticated medical database systems is still unlikely to give you the detail you need to actually treat someone.

Of course, there is little useful information on how this will work - whether the doc will consult this wonderful oracle or if the insurance company will send you a form letter six months after the patient died saying you should have done something different. If they are going to go through with this,, I hope to hell they are going to carefully monitor it's success (or lack thereof) over time. And do that honestly. My money is that it won't help all that much.

Re:What could possibly go wrong?

[elsurexiste]

Wikipedia suggests that the real reason was technical: loading the required knowledge into the system was painful. This is more related to my own experience: ESs don't get adopted because UI is crappy at bests.