WellPoint, one of the nation’s largest health insurers, and IBM have agreed to work together on a new method of bringing medical evidence to bear on diagnosis and treatment. The companies will use IBM’s Watson supercomputer to apply insights from the medical literature to clinical information about particular patients; the system will then recommend the most probable diagnosis and treatment options to physicians and nurses, linking them to the original studies.

Eventually, if the approach works, it will be made available to all of the insurance companies that belong to the Blue Cross Blue Shield Association, a WellPoint spokesman said.

Watson, which can “understand” concepts in human language, is celebrated for having beaten human contestants in the TV game show “Jeopardy.” Watson can sift through and analyze 1 million books or 200 million pages of data in three seconds, according to a press release. This provides an enormous advantage in the medical field, where more than 800,000 new studies are published each year.

IBM has been actively seeking ways to use Watson in health care. Earlier this year, the company announced a collaboration with Nuance Communications, which specializes in voice recognition software and has developed its own form of clinical language understanding. Physician researchers at Columbia University and the University of Maryland are also working on medical applications of Watson.

WellPoint has high hopes for Watson. Its executives envision a system that could speed diagnosis and ensure that patients receive treatments based on the most up-to-date evidence. It also believes Watson could be useful in reviewing complex cases and deciding whether particular treatments merit coverage.

“What that means is that the review of cases is based on medical evidence,” explained Andrew J. Lang, CIO of WellPoint, in an interview with InformationWeek Healthcare. “The more consistent and higher usage of medical evidence both in the physician office and the health plans will streamline that process between the providers and the care reviewers. If the medical evidence supports certain procedures or protocols, our clinical review would be streamlined.”

Up to now, most physicians have not used the available clinical decision support tools in their diagnosis and treatment of patients, notes Mark H. Ebell, MD, an evidence-based-medicine expert who is deputy editor of American Family Physician and editor-in-chief of Essential Evidence. “Sometimes, it’s because they’re too complicated,” he noted. “Other times, they prefer to use their own intuition over a decision support tool.”

So why should physicians trust recommendations from a computer–especially when that advice comes to them from an insurance company?

Physicians who have seen the early work on the Watson application have had an “overwhelmingly positive” response, said Lang. “The key is that it’s just providing the information to the physician to help them make better decisions. Over a million new medical articles come out every year, and the volume of data is overwhelming for physicians to keep up with and to apply the relevant information to their patients. We think this is a real game changer to provide that information to them in a relevant fashion.”

Ebell said he believes Watson could potentially make “a unique and important contribution” to evidence-based medicine. But it’s not going to be easy, he said. The researchers at Essential Evidence (formerly known as Inforetriever) have been combing through the medical literature for 15 years in search of important studies, he noted. “We’ve been looking at the 100 top journals every month in a systematic way, and only about 1 in 100 studies, even in those top journals, should really change clinical practice and is really relevant to answering a question at the point of care.”

One reason is that clinical trials usually target a very narrowly defined population. They might exclude patients with certain variants of a disease, those who have additional issues, and very old or very young patients. “Yet the hard questions come up when you have a patient with heart failure and depression and they’re alcoholic and have liver failure and several other things going on,” Ebell observed. “The studies don’t report data for subsets of patients like that.”

In addition, he pointed out, many studies have endpoints that are irrelevant to real-world clinical issues. To help a doctor at the point of care, he said, a study must be free of bias and find improved patient-oriented outcomes.

“It’s not just whether a drug improves your blood pressure and blood sugar; it’s ‘does it help you live longer and better?’ Will Watson be able to distinguish studies that matter from those that don’t, and will it be able to distinguish well-designed studies from poorly designed studies?” Ebell asked. “It’s a do-able challenge, but they have to ask the question in the right way and be thinking about these issues.”

Posted September 19, 2011.