September 12, 2011

Can IBM’s Watson Beat the Odds?
By Elizabeth S. Roop
For The Record
Vol. 23 No. 16 P. 14

Backed by superpowers previously unseen in a computer, the game show sensation has set its sights on helping cure what ails healthcare.

Having put Watson through its paces with a successful showing against two Jeopardy! champions, IBM has set in motion plans to turn its supercomputer loose on the healthcare industry. With its ability to parse natural language and analyze huge volumes of information in just seconds, Watson has the potential to revolutionize clinical decision support—but only if it can win over skeptical clinicians and overcome traditional barriers to market entry.

In a nutshell, IBM’s Watson is a deep analytics and natural language processing computer capable of processing and recognizing ambiguous words. It is able to “understand” the meaning and context of human language and rapidly process information to find precise answers to complex questions.

In the context of healthcare, this means Watson can provide enhanced clinical decision support by rapidly considering all texts, reference materials, prior cases, and the latest knowledge in journals and medical literature related to a potential diagnosis and treatment plan. Watson evaluates each possible answer based on the evidence it can gather through refined secondary searches, allowing it to provide a confidence level with each answer to guard against clinicians settling on a diagnosis before first considering all plausible alternatives.

Watson can “process the equivalent of 1 million books per second. That’s a lot of information,” says Andrew Clarke, IT director at New England Medical Transcription. “It’s not that they are reinventing the wheel. The power isn’t that Watson has new algorithms; it’s that it can process thousands of algorithms per second. That’s the power that makes it a strong tool in the medical world.”

Indeed, Watson’s speed is mind-boggling, says Clarke. “It processes thousands of algorithms at once for possible answers, looks for consensus, and then bounces that against a known entity like an encyclopedia. When you think about everything that went into making this work, that is pretty incredible,” he says.

But is it incredible enough to be a game changer in HIT? And can it be a financial success for IBM?

Unlimited Potential?
Many agree that Watson has the potential to reshape any aspect of healthcare that relies on huge volumes of data. In addition to diagnostic and medication decision support, this could include scientific research and even billing and coding, particularly once ICD-10 comes into play in the United States.

“With computing power like this, the likelihood of making mistakes will diminish over time. It will change the entire process of how you are triaged and admitted to the hospital. It will change the way you are diagnosed,” says Clarke.

For example, when a patient presents in the emergency department, the initial evaluation and the resulting preliminary diagnosis could be conducted by a clerk utilizing Watson. This would eliminate the “guessing game” and allow the patient to be immediately referred to the proper specialist for formal diagnosis and treatment. The result would be reduced costs and faster and “better care because you still have human interaction,” Clarke says.

To ensure the power of Watson can be fully leveraged in healthcare, IBM is collaborating with physicians at Columbia University to identify critical issues the supercomputer may be able to address. Meanwhile, physicians at the University of Maryland School of Medicine are working to identify how Watson can best interact with clinicians to provide maximum assistance.

IBM is also striving to close what some see as a critical capability gap: the lack of voice recognition. Its absence was noticeable in the Jeopardy! competition when Watson generated answers that had already been cited by the human contestants.

“You can help solve that problem by providing an effective voice interface,” says Nick van Terheyden, MD, chief medical information officer at Nuance Communications, Inc. “That continuous input of data is important … because this isn’t likely to be static. [Watson] will process information much like the human brain, the clinical brain, processes a case—differential diagnoses which are then refined with a physical exam.”

To that end, IBM and Nuance are collaborating on ways to combine IBM’s deep question answering, natural language processing, and machine learning capabilities with Nuance’s speech recognition and clinical language understanding solutions.

IBM did not respond to repeated interview requests. However, according to a press release announcing the collaboration, IBM and Nuance will jointly invest in a multiyear research initiative targeted to the applications of the Watson technology to assist in the diagnosis and treatment of patients in combination with Nuance’s voice and clinical language solutions. In addition, IBM has licensed access to the Watson technology to Nuance.

IBM and Nuance are currently engaged in a five-year joint research initiative designed to advance next-generation natural language speech technologies, the results of which will be commercialized by Nuance. IBM also named Nuance its preferred business partner for speech technologies and related professional services.

“I’ve seen both ends of the spectrum, the clinical and nonclinical, and I personally see this as an enormous potential leap forward,” says van Terheyden. “To be clear, the innovation IBM demonstrated in the Jeopardy! challenge was hugely significant, but that was a general application. Translating that into healthcare is part of what IBM and Nuance are working on now. Can we take that quantum leap and apply that in healthcare?”

Like Clarke, van Terheyden envisions multiple applications for Watson technologies. For example, its ability to rapidly and intuitively parse huge volumes of data can go a long way toward taming the “burgeoning content of medical records. We’ve created a bigger and bigger haystack and being able to access that [data] in a more intelligent way is clearly an opportunity from a research standpoint,” he says.

What van Terheyden says Watson won’t do is replace the clinician. What it may do is create a new computerized version of the collaboration that once took place in the hospital physicians’ lounge, which was already being driven toward extinction by EHRs, texting, and instant messaging. But it cannot replace human interaction.

Rather, “It becomes a curbside consult for clinicians,” van Terheyden says. “Human involvement remains important, as we saw in the Jeopardy! challenge. [Watson] becomes a trusted colleague. There is still a trust factor that must be established. That will happen over time with good data and good analysis. We must demonstrate the value proposition and, most importantly, the capability of bringing this vast amount of knowledge to our patients … to bring this knowledge to bear with technology that sits next to us and offers the medical intelligence at the point of care and at the time of consultation.”

Barriers to Entry
Despite Watson’s seemingly infinite potential in healthcare, there are several fairly significant barriers to market entry that IBM must overcome. Clinician adoption is one.

Not only are physicians hesitant to adopt any new clinical decision-support technology until they are convinced the data it searches are trustworthy, but there is also the issue of how well Watson will align with existing clinical processes and best practices.

“Will they need to throw out those best practices? Physicians have built all their workflows around [clinical decision-support] tools that have already been developed and now there is something new to consider,” says John Moore, managing partner of Chilmark Research, an industry analyst firm focused exclusively on the HIT market. “There are a lot of solutions in the market already. IBM will see competition from solutions that may come in at lower price points or that may be more tightly integrated within the context of the EHR so they are fully embedded in the physician workflow.”

Indeed, price point may be the most significant obstacle of all. With some estimating an initial price tag in excess of $1 million, early adoption of Watson is likely to be limited to very large academic medical institutions and hospital chains.

The delivery model will be another obstacle IBM must overcome. Will Watson be a cloud-based subscription service? What are the sources of the medical content it will tap and how will that data be kept current?

“It shows promise. Natural language processing is very state-of-the-art, and Watson can handle a much broader array of nuances and descriptions than [the technology] we have today. The question is what it will take to bring it to market and when they do, what the price point will be,” says Moore. “How are they going to productize Watson into a form and at a price point that physicians will adopt and use?”

Moore notes that many of these questions will be answered as the collaboration between Nuance and IBM progresses. Until then, “It’s one of those things that we won’t know what we have until something is delivered to the healthcare sector,” he says.

The Time Is Right
Despite what will likely be a steep uphill climb, Moore says it’s an excellent time for IBM to act on the market potential it foresees for Watson because there is “a lot of money floating into healthcare right now dedicated to IT. The waters have been chummed with federal stimulus dollars, and the sharks are coming in. IBM is one of them.”

But other factors are at play that suggest now is the best time for a game-changing HIT solution. One is IBM’s decision to set aside its past reliance on closed, proprietary technologies and instead use industry standard technologies and algorithms.

“That will likely help them because people will be more likely to adopt the technology and also it makes it easier to enter all the data sources,” says Clarke.

Most importantly, he says, the power of computers has grown exponentially in recent years and shows no signs of slowing. The ability to combine that with advanced voice recognition and leverage it to process into something meaningful the huge volumes of data that are threatening to bury healthcare promises to change everything—assuming the barriers to entry don’t prove to be too much to overcome.

“As computing power grows and algorithms grow, computers can learn from their mistakes and they will continue to grow. … Everything that science fiction imagined is starting to come true,” says Clarke. “The ability to network so many computers to get their combined power can revolutionize healthcare.”

Doctors can absorb only so much data, adds van Terheyden. “If you look at healthcare and the [volume of] knowledge that we as clinicians are presented with, for a physician to keep up-to-date, he’d have to be reading journals 70 to 80 hours a week. That’s clearly impractical,” he notes. “What this represents is the ability to balance that gap and return us to what we, as patients, expect: best care based on the latest validated information.”

— Elizabeth S. Roop is a Tampa, Fla.-based freelance writer specializing in healthcare and HIT.