IBM has always intended Watson--the natural language processor best known for beating out its puny human competitors on the game show Jeopardy!--to do more than answer trivia questions. And the complex, high-stakes, data driven healthcare industry is one of the areas where the company sees a potential gold mine.
But there's a lot of work to be done before Watson comes to a doctor's office near you. Indianapolis-based WellPoint has enlisted a team of clinicians, computer scientists, statisticians, pharmacologists, IBM engineers and others to "teach" Watson to provide decision support.
The project is currently under development at the Cedars-Sinai Samuel Oschin Comprehensive Cancer Institute in Los Angeles, with additional pilot sites to be named later.
WellPoint's Elizabeth Bigham, vice president of care management and Ashok Chennuru, director of technology, sat down with FierceHealthIT at the annual meeting of the Health Information and Management Systems Society in Las Vegas to talk about the project's progress so far--and the plans for the future. (Here's a hint: They aren't trying to replace doctors with machines but they are mulling different business models.)
FierceHealthIT: How does Watson work as a decision-support tool?
Bigham: Watson has the ability to take in infinite amounts of content. In order to make a treatment decision support system work we are investing in Watson clinical evidence and clinical guidelines so we can build up the world of scientific knowledge of what is out there. And we are building on our own a longitudinal patient record with clinical data from your provider.
We're really giving Watson a very basic medical education. We're building in some basic medical terminology and so on. We're starting with oncology. But it can't just focus on oncology.
It will also include longitudinal patient records--and not just from hospitals and doctors. We're integrating it with our own data, including claims data. We already know our patients' diagnoses and the procedures that they got and the date of service and the medications that they're taking.
If you as a patient are seeing your oncologist, Watson would pull out your specific data from our record and compare it to the world of evidence and guidelines and pick out a personalized selection of treatment recommendations for you in particular. Not people like you, but for you in particular based on your characteristics.
Chennuru: There are so many variables. Watson has the capacity to put all this together in a way that a person just can't. And it can stay current with all the literature. Physicians can spend eight hours a week looking at ten different journals for one patient-and it's not patient specific. That's what's gotten us traction. It's all about integrating it into their workflow. It has to enhance it.
FHIT: What will the doctor-Watson relationship look like?
Chennuru: As a payer we are going to provide tools to enable the physician versus trying to do their jobs. If it tries to replace the doctor it will fail.
Bigham: We don't intend to build a system that tells them what to do. This is definitely a decision-support tool. So it will not say "Here is the right answer." It will say "Here are some recommendations."
And it's probabilistic recommendations, not deterministic--it doesn't say "These are the only three things you should do," but "These are the things that have the highest likelihood." We're not in any way requiring them to choose the recommendations.
Chennuru: And in fact we want physicians to provide input so we can train it and make it better. It's a learning system. That's what really intrigued us to work with IBM. Because it's only going to get better. And with the power of technology and the processing power we can turn all the information and make it more useable. A physician might be looking for three lines of information in a journal. Watson can find the relevant information a lot faster than a human.
FHIT: So what's the business model? Who pays?
Bigham: We haven't figured out the business model yet. One of the reasons is that we're still fine-tuning exactly what it's going to do. At the beginning we're going to ask providers to use it and help us improve it.
This is new even for us. We don't have another machine-learning tech system, so we're trying to figure out what that means.
Chennuru: The easy answer to your question for now is we do want to charge providers. But the way we're looking at it is we want to first get the penetration. Once providers find it valuable, they're going to pay for it. But we don't want that to happen until after penetration. So we want to start with working with the small pilot group, making it stable, making it valuable and then expanding it."
Bigham: Whether the provider pays or not is not the only question. Maybe we'll give them some of the tools and then if they want other ones they would pay for those. There are a lot of possibilities. We're also flirting with other ideas that are a lot less developed, but maybe we'll decide to make this a consumer-facing patient-education tool.
FHIT: How will physicians get a return on any financial investment they might make?
Chennuru: We are going to go toward paying for outcomes. This will be an enabler. As we start to get into bundled payments and ACO arrangements, this is going to be the key driver there.
Also, it's patient-centered and enables provider access to more information. If you can show the information derived from Watson in an easily-consumable manner, in a PDF, and give it to the patient, then he or she is better informed. That could cut down on readmissions, improve outcomes, and improve patient satisfaction.
FHIT: When might we see a Watson decision-support tool in docs' offices?
Bigham: We don't have a firm date that we're shooting for to get this out there. We are definitely focusing on getting it right. Physicians are going to be using this to make treatment decisions for their patients and it's important to get it right. The wrong treatment can have devastating effects.
Editor's note: This interview has been edited for length and for clarity.